Value Perception and Pricing Strategies for B2B Generative AI Solutions PDF Free Download
1 / 126/126
100%
Value Perception and Pricing
Strategies for B2B Generative AI
Solutions
Oscar Löfqvist
School of Science
Thesis submitted for examination for the degree of Master of
Science in Technology.
Espoo 26.05.2025
Supervisor
Associate Professor Jukka Luoma
Advisor
Dr. Jan Emmanuele
Copyright ©2025 Oscar Löfqvist
Aalto University, P.O. BOX 11000, 00076 AALTO
www.aalto.fi
Abstract of the master’s thesis
Author Oscar Löfqvist
Title Value Perception and Pricing Strategies for B2B Generative AI Solutions
Degree programme Master’s Programme in Science and Technology
Major Industrial Engineering and Management
Supervisor Associate Professor Jukka Luoma
Advisor Dr. Jan Emmanuele
Date 26.05.2025 Number of pages 109+6 Language English
Abstract
The rapid adoption of generative AI is reshaping how B2B firms perceive, create,
and seize value. Despite heavy investments, many companies struggle to translate
generative AI solutions into sustainable revenue and profits. Despite the central
role of pricing in commercial success, little guidance exists for monetising generative
AI-driven offerings. This research examines how companies evaluate the value of
generative AI solutions and translate that value into effective pricing strategies.
Drawing on 31 semi-structured interviews with buyers and suppliers indicates that
the highest perceived value of generative AI solutions comes from automating tasks
and time savings. Although initial market enthusiasm may temporarily elevate will-
ingness to pay, respondents anticipate that generative AI capabilities will ultimately
become commoditised and become an expectation rather than a value-adding feature,
reducing price premiums. Moreover, interviews reveal a shift from conventional
software licensing toward value-based pricing. However, it is constrained by persis-
tent challenges in measuring and quantifying generative AI, and respondents remain
divergent in their perception of inference cost.
These insights were synthesised into a double-dimensional pricing matrix. The
framework outlines four pricing strategies: value-based, usage-based, traditional SaaS,
and credit-based pricing strategies. It maps them against specific cost considerations
of large-language-model powered generative AI solutions and customer-perceived
value. This gives managers a practical tool to align pricing decisions with product
characteristics, market expectations, and commercial viability.
This thesis bridges a critical gap in modern pricing by integrating customer-perceived
iv
value and strategic pricing theory with the realities of generative AI. It contributes to
existing research by expanding it into the generative AI area and proposes a practical
framework for managers to price generative AI offerings. In doing so, it advances
academic understanding of value dynamics in the emerging area of generative AI. It
equips managers with a framework to choose and adapt pricing models that protect
margins and capture the value of generative AI.
Keywords Generative AI, Value Perception, Willingness-to-Pay, Pricing Strategies,
B2B
Aalto-universitetet, PB 11000, 00076 AALTO
www.aalto.fi
Sammandrag av diplomarbetet
Författare Oscar Löfqvist
Titel
Värdeuppfattning och prissättningsstrategier för generativa AI-lösningar inom
B2B
Utbildningsprogram Magisterprogram i Teknikvetenskapp
Huvudämne Produktionsekonomi
Övervakare Professor Jukka Luoma
Handledare Dr. Jan Emmanuele
Datum 26.05.2025 Sidantal 109+6 Språk Engelska
Sammandrag
Det snabba införandet av generativ AI omformar hur B2B-företag uppfattar, skapar
och fångar värde. Trots stora investeringar har många bolag svårt att omvandla
generativa AI-lösningar till hållbara intäkter och vinster. Även om prissättning är
avgörande för kommersiell framgång, finns det begränsad kunskap om hur AI-drivna
lösningar kan kommersialiseras. Denna studie undersöker hur företag bedömer värdet
av generativa AI-lösningar och omvandlar det till effektiva prissättningsstrategier.
Baserat på 31 semistrukturerade intervjuer med köpare och leverantörer framgår det
att det största värdet i generativa AI-lösningar är automatisering och tidsbesparingar.
Även om marknadsentusiasm tillfälligt kan höja betalningsviljan förväntar de inter-
vjuade att generativ AI-funktionalitet så småningom kommer att bli en förväntad
standard och inte längre ett värdeskapande särdrag, vilket minskar prispremier-
na. Intervjuerna visar dessutom en omställning från konventionella licensmodeller
mot värdebaserad prissättning. Detta försvåras emellertid av utmaningar att mä-
ta och kvantifiera generativa AI:s faktiska bidrag. Dessutom kvarstår skillnader i
uppfattningar om kostnaderna för AI-inferens.
Dessa insikter har sammanförts till en tvådimensionell prismatris. Ramverket beskriver
fyra prissättningsstrategier: värdebaserad, användningsbaserad, prenumerationsbase-
rad och kreditbaserad prissättning. Dessa placeras längs två axlar: värdetydlighet,
hur väl fördelar kan mätas, och kostnadspålitlighet, stabilitet i AI-inferenskostnader.
Detta ger chefer ett praktiskt verktyg för att anpassa prissättningen till produktens
egenskaper, marknadsförväntningar och lönsamhetsmål.
vi
Avhandlingen fyller en kritisk kunskapslucka kring modern prissättning genom att
integrera värde- och strategisk pristeori med generativa AI-lösningars verklighet. Den
bidrar till befintlig forskning genom att utvidga forskningen till AI-området och
föreslår ett praktiskt ramverk för att prissätta generativa AI-erbjudanden. Därigenom
fördjupar studien den akademiska förståelsen av värdedynamik inom generativ AI och
förser beslutsfattare med ett verktyg för att välja och anpassa prissättningsmodeller
som skyddar marginaler och fångar verkligt affärsvärde.
Nyckelord
Generativ AI, Värdeuppfattning, Betalningsvilja, Prissättningsstrategier,
Företag-till-företag (B2B)
vii
Contents
Abstract iii
Abstract (in Swedish) v
Contents vii
Acknowledgments x
Abbreviations xi
1 Introduction 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 ResearchQuestions............................ 3
1.3 ScopeoftheThesis............................ 4
1.4 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Overview of Generative AI 6
2.1 From Classical Machine Learning to Transformers . . . . . . . . . . . 6
2.2 Generative AI in Products . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Generative AI in Services . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Literature Review 12
3.1 UnderstandingValue ........................... 12
3.1.1 Historical View on Value . . . . . . . . . . . . . . . . . . . . . 13
3.1.2 Customer Perceived Value . . . . . . . . . . . . . . . . . . . . 14
3.1.3 ProductValue........................... 16
3.1.4 Quantifying and Determining Value . . . . . . . . . . . . . . . 18
3.1.5 Perception of Value for Generative AI Created Work . . . . . 20
3.2 Customers’ Willingness to Pay . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Pricing................................... 22
3.3.1 PricingOverview ......................... 22
3.3.2 Value-Based Pricing . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.3 Cost-Based Pricing . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.4 Competition-based Pricing . . . . . . . . . . . . . . . . . . . . 29
3.3.5 Pricing Models for Digital Businesses . . . . . . . . . . . . . . 30
3.3.6 Services-Based Pricing . . . . . . . . . . . . . . . . . . . . . . 31
3.3.7 Pricing of New Products . . . . . . . . . . . . . . . . . . . . . 33
3.3.8 Pricing in the Age of Generative AI . . . . . . . . . . . . . . . 34
viii
3.3.9 Price Discrimination . . . . . . . . . . . . . . . . . . . . . . . 39
4 Research Methodology 40
4.1 ResearchApproach............................ 40
4.2 ResearchSetting ............................. 41
4.3 DataCollection.............................. 42
4.3.1 Semi-structured Interviews . . . . . . . . . . . . . . . . . . . . 42
4.3.2 DeskResearch........................... 46
4.4 DataAnalysis............................... 47
4.4.1 Semi-structured Interviews . . . . . . . . . . . . . . . . . . . . 47
4.4.2 DeskResearch........................... 50
4.5 Pricing Framework Development . . . . . . . . . . . . . . . . . . . . . 50
5 Findings 53
5.1 GenerativeAIValue ........................... 53
5.1.1 AddingValue ........................... 53
5.1.2 Outcome Versus Process Value . . . . . . . . . . . . . . . . . 55
5.1.3 The Value of Human-in-the-Loop . . . . . . . . . . . . . . . . 58
5.1.4 MeasuringValue ......................... 60
5.2 Generative AI as a Competitive Advantage . . . . . . . . . . . . . . . 62
5.3 Long-Term Pricing Pressure in Generative AI . . . . . . . . . . . . . 65
5.4 Willingness to Pay for Generative AI Solutions . . . . . . . . . . . . . 67
5.5 Pricing Generative AI . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.5.1 Traditional SaaS Pricing . . . . . . . . . . . . . . . . . . . . . 70
5.5.2 LLM Cost Consideration . . . . . . . . . . . . . . . . . . . . . 72
5.5.3 Moving Towards Value and Outcome-based Pricing . . . . . . 77
5.5.4 Pricing of Services . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5.5 Challenges with Value and Outcome-based Pricing . . . . . . 81
5.5.6 Indirect Cost Considerations . . . . . . . . . . . . . . . . . . . 83
6 Discussion 86
6.1 Answering the Research Questions . . . . . . . . . . . . . . . . . . . . 86
6.2 Capturing Generative AI Value with New Pricing Framework . . . . . 89
7 Conclusion 96
7.1 Managerial Implications . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.2 Contribution to Literature and Future Research . . . . . . . . . . . . 97
7.3 Limitations ................................ 98
7.4 Conclusion................................. 99
x
Acknowledgments
I want to express my sincere gratitude to Jukka for stepping in at short notice to
supervise and support my thesis. Your insights and encouragement were invaluable
to the project and its final outcome.
Secondly, I would like to thank Jan for his guidance throughout the spring. Your
perspective added real depth to my work, and our conversations will stay with me
for a long time. I hope this will not be the last time we collaborate.
I also wish to extend my sincere thanks to all the interviewees who generously took
the time to speak with me and contribute to this thesis. Your input made this
research possible.
Looking back, I still remember the first day I walked onto the Aalto University
campus on a warm, sunny September morning, filled with excitement about what
lay ahead. Time at university went quickly, quicker than expected. Yet, I had the
time to explore my interests, build experiences, and form long-lasting friendships
with many people. I leave with a wealth of knowledge gained through my studies,
volunteering, and work experiences.
It was fun, I can say no less. The next adventure has already begun, and I can only
look back at the great times on campus, enabled by the fruitful discussions among
friends. Thank you all!
In seat 1A at 33 000 feet somewhere over Europe
Oscar Löfqvist
17.5.2025
xi
Abbreviations
AI Artificial Intelligence
AWS Amazon Web Services
B2B Business-to-Business
CPV Customer Perceived Value
GCP Google Cloud Platform
Gen AI Generative Artificial Intelligence
KPI Key Performance Indicator
LLM Large Language Model
LTV Labour Theory of Value
ML Machone Learning
PE Private Equity
R&D Research & Development
ROI Return on Investment
RQ Research Question
SaaS Software as a Service
WTP Willingness to Pay
YC Y Combinator
1
1 Introduction
The rapid advancement of generative Artificial Intelligence (gen AI) fundamentally
reshapes how value is created, perceived, and captured across industries. As gen AI
become embedded in products and services, it challenges traditional theories of value
capture and pricing strategies. Despite the widespread adoption of generative AI
and its anticipated economic impact, a notable gap remains in academic research at
the intersection of gen AI, value, and pricing.
Classic SaaS pricing theory assumes low, predictable marginal costs and homogeneous
usage, but generative-AI solutions face high initial training costs on top of standard
development costs. In addition, each user interaction with an LLM is associated with
an inference cost that can vary significantly between users and between different
underlying models. This turbulence, coupled with an industry racing ahead on hype
cycles, potentially renders cost- and competition-based pricing strategies obsolete.
At the same time, value-based strategies, generally regarded as the gold standard in
pricing, can struggle to isolate and quantify the value that gen AI actually creates.
This Master’s thesis addresses this gap by examining how value is perceived and
captured in a B2B context where generative AI plays a central role in the solution
offered. Motivated by a personal and academic interest in the intersection of emerging
technologies and business theory, the thesis examines both the customer’s valuation
of gen AI solutions and the pricing strategies firms can employ to optimise value
capture. The research builds on foundational theories of economic value and pricing,
adapting them to the realities of generative AI-driven businesses.
1.1 Background and Motivation
The concept of economic value theory originated from the classical economist Adam
Smith in the 18th century with his work The Wealth of Nations where the Labor
Theory of Value (LTV) was conceptualized (Sandmo, 2011). Over time, the concept
of value has evolved, encompassing neoclassical value theory, game theory, and, most
recently, the study of value within network effects of modern social media platforms.
Every evolution of value theory has occurred within a shifting world order, from the
Industrial Revolution to the rise of computers and the internet.
The ongoing advancements and world adoption of generative AI are changing the way
we think about and conduct business. This transformational shift into gen AI that
business leaders such as Alphabet CEO Sundar Pichai are calling the biggest shift in
our lifetime and Bill Gates referred to as having the potential to change the world in
2
ways we can’t even imagine (Powell, 2025; Marr, 2023), poses yet a new opportunity
to redefine and understand how value is perceived in the age of generative AI. Yet, as
this technological revolution unfolds, a central question remains: how can companies
effectively capture the value it creates?
Capturing product value has long been recognised as a challenge (Hinterhuber, 2004),
and it is evident that capturing the value of gen AI solutions remains difficult.
Industry leaders such as OpenAI have acknowledged that they are incurring losses
on customer usage of their latest models, despite users paying up to 200 dollars
per month for access (Altman, 2025). Further, already in the fall of 2023, Sequoia
Capital, one of the leading venture capital firms, posted the question "Where is all
the revenue?" and reiterated the same question in mid-2024 (Chan, 2024). Chan is
referring to the 600 billion dollars in AI revenue that is missing to justify the current
spending on AI infrastructure.
Research has also started to ask how gen AI technology should be priced (Mahmood,
2024). It is not only product firms that are asking how to capture the value of gen
AI, the professional services industry is also at risk of value destruction. The current
pricing models in the professional services industry no longer fit in the gen AI world.
They pose a tremendous risk of revenue leakage when gen AI technology is being
introduced (Biermann and Petersen, 2024b).
Given this evidence, it is not trivial to say how companies should capture the value
of generative AI or how gen AI should be priced to drive revenue (Santoro and Hare,
2025). A recent AI survey by Bain & Company indicated that nearly 90% of the
companies surveyed are developing or have deployed some level of generative AI
into their business (Rapoport et al., 2024). The survey also indicated that between
one-fourth and nearly one-half of a company’s AI use cases are being addressed by
buying generative AI solutions instead of internally developed products (Rapoport
et al., 2024). Additionally, McKinsey & Company reports an even larger percentage
with over 50% of companies surveyed buying off-the-shelf gen AI solutions (Singla
et al., 2024). Thus, it is evident that the technology has already entered the broader
business landscape, yet the revenue seems to be missing. This is particularly alarming
as over 50% of US C-suit executives are expecting that gen AI will deliver more than
5% revenue growth to their business in the next three years, and 17% are expecting
over 10% increase (Mayer et al., 2025).
A review of current research in the field of AI reveals a heavy focus on either
the technological advancement, such as Multimodal, or the societal impact of the
technology and its relation to regulation, ethics, and safety. However, there is a gap in
3
the academic literature regarding the understanding of the value of gen AI solutions
offered by companies. In addition, the perception of value must be quantified and
integrated into a strategic framework for businesses to take advantage of it. Current
research also lacks understanding of the pricing strategies surrounding products and
services that incorporate generative AI as part of their offerings. Lastly, if Sequoia
can be trusted, it seems evident that many companies are struggling to define the
value of generative AI and determine optimal pricing strategies to maximise revenue.
The goal of this thesis is to illuminate the perceived value of generative AI in a
B2B context. Additionally, this paper addresses how value should be captured and
priced within a generative AI setting. The thesis contributes to the current academic
literature by examining value and pricing theory from the perspective of generative
AI. The aim is also to close the gap in the current understanding of how gen AI
changes value perception. It further provides the reader with practical theories to
apply when integrating gen AI into products and services.
1.2 Research Questions
The theory of value has been foundational for businesses for centuries, and there is
no indication that it will not continue to be so. Customers buy products and services
based on value, the difference between what companies provide and what they charge
(Leszinski and Marn, 1997). Current research still lacks understanding of how value
is perceived in the shift towards generative AI product features. To both advance
the academic knowledge of value in a gen AI context and to help businesses capture
the value of AI, this paper will first address the research question:
RQ1: How do buyers and suppliers perceive, measure, and
anticipate the evolution of value in generative-AI solutions?
To further understand the impact of gen AI, the first research question is broken
down into three sub-questions that will help answer the first research question.
RQ1.1: Where and how do buyers and suppliers perceive incremental
value from generative AI solutions, and which attributes drive that
perception?
RQ1.2: Which metrics and methods are companies using to measure the
value of generative AI?
RQ1.3: Which factors shape buyers’ and suppliers’ perceptions of the
long-term value of generative-AI solutions, and how do those factors exert
influence?
4
The second part of this paper will address the quantification of value in a generative
AI setting that ties into pricing. Pricing is one of the most critical business functions,
yet it is widely overseen or mismanaged by companies (Baker, Marn, and Zawada,
2010). In addition, according to Hinterhuber (2004), pricing has received far too little
attention in academic research, given its importance for business success. Particularly
given the impact that generative AI is expected to have on businesses. Thus, the
second research question answers how gen AI should be priced and what factors
determine the pricing strategy. The aim is to enhance the understanding of pricing in
academic research and to help practitioners better comprehend the pricing dynamics
in a generative AI setting. As a result, the second research question is:
RQ2: Which technological and market factors shape suppliers’
pricing decisions for B2B generative-AI solutions?
The second research question is also broken down into three sub-questions to illu-
minate further the dynamics involved in generating AI value capture and pricing.
These sub-questions are:
RQ2.1: Which factors raises or constrains customers’ willingness-to-pay
for generative-AI solutions?
RQ2.2: How well do prevalent SaaS-derived pricing models capture
generative-AI value?
RQ2.3: How should companies capture and price generative AI solutions?
The research questions provide a clear scope for the thesis and serve as the backbone to
add value to the current academic literature and business managers. The researcher’s
questions are answered by first understanding the current literature on the topics.
Second, interviews are conducted to understand how companies currently value and
perceive generative AI pricing. Lastly, the academic research is combined with the
interviews to create a framework for value capture and pricing of generative AI
solutions in a B2B setting.
1.3 Scope of the Thesis
The thesis focuses on companies developing generative AI products and services, as
well as companies that have or are going to procure generative AI solutions. Products
in this context are software solutions that are either a gen AI-first solution or a
current software product that is adding a generative AI component. Services refer to
professional or technical service providers that are using humans to deliver a service
and are incorporating gen AI to improve the service offering. Lastly, generative AI is
5
defined as probabilistic software based on large language models, extending beyond
just a chat-based application.
This thesis focuses exclusively on a business-to-business (B2B) context, examining
both suppliers and buyers. Given that market dynamics in B2B differ significantly
from those in business-to-consumer (B2C) settings, the analysis is limited to the
B2B domain.
Further, the thesis focuses on value and pricing. Value is characterised as the surplus
between the benefits a buyer receives and the price they surrender — the classic what
I get for what I give. Price is that monetary sacrifice, positioned between a cost-based
floor price, the minimum price needed to cover expenses, and a value-based ceiling
price, the maximum price that can be charged based on the benefits received by the
customer. Hence, pricing is the mechanism that converts created value into revenue.
Lastly, certain aspects surrounding pricing, such as regulation, have been excluded
due to its constant evolution and local differences (Baker, Marn, and Zawada, 2010).
In addition, demand estimation under different utility functions is also excluded due
to the early stages of gen AI and the lack of data.
1.4 Structure of the Thesis
The thesis is divided into seven main sections. Section 2 Overview of Generative
AI introduced the reader to generative AI, its developments and the current imple-
mentation of gen AI into products and services. This chapter provides background
information for readers to understand the unique characteristics of generative AI
better. Chapter 3 Literature Review introduced the literature review. It encompasses
value in various contexts, in addition to the customer’s willingness to pay, and
introduces relevant pricing literature.
Next, Chapter 4 introduced the research methodology, data collection, and how the
pricing framework was created. This thesis employs a qualitative, interview-based
approach as its primary research methodology. Chapter 5 Findings walks through
the interview findings and ties them to the research questions and literature review.
Chapter 6 discusses the findings and Section 6.2 introduced a 2x2 pricing framework
for pricing generative AI solution in a B2B context. Lastly, Chapter 7 concludes
the thesis by discussing the managerial implications of the findings, contribution to
academic literature, future research and the study’s limitations, before concluding
the thesis.
6
2 Overview of Generative AI
Already a month after the release of ChatGPT, news articles appeared that tried
to help companies navigate the use of generative AI assistants and LLM-powered
chatbots. These articles suggested everything from drafting emails to customer
onboarding, and translation (Marr, 2022). This section focuses on the practical
implementation of LLM-based technology into business products and services.
First, this section introduces the reader to generative AI, its background, and its
current implementation in businesses. It serves as the background for later discussions
on how generative AI connects to both value and pricing theory. Second, it provides
an overview of how gen AI is being incorporated into products and services.
2.1 From Classical Machine Learning to Transformers
The invention of artificial intelligence (AI) and machine learning (ML) can be traced
back to Alan Turing. In his paper Computing Machinery and Intelligence, he proposed
the question "Can machines think?" (Turing, 1950 [2009]). This can be traced to be
the starting point for scholars to think about machine intelligence and it helped to
frame the view on what would develop into machine learning (E. Y. Zhang et al.,
2023).
The next major leap in computer understanding was the development of Hidden
Markov Models in the mid-1960s, followed by Recurrent Neural Networks in the
80s (E. Y. Zhang et al., 2023). In the nineties Convolutional-Neural-Networks was
introduced, enabling high-quality computer vision recognition (LeCun et al., 1998).
In addition, the general public got a new view on the computer’s power when IBM’s
Deep Blue defeated Garry Kasparov in chess (IBM, n.d.). Mr Kasparov stated after
the match that "For the first time in the history of mankind, I saw something similar
to an artificial intellect”.
The 2000s saw massive improvements in ML and AI, driven by the introduction of
frameworks, ways of thinking, and computing power. Google DeepMind’s AlphaGo
defeated the Go world champion player in 2016, and the following year, Google Re-
search published the foundational paper for language model understanding, Attention
is all you need, introducing the transformer (Google Cloud, 2024; Waswani et al.,
2017).
The transformer architecture started what we now call generative artificial intelligence,
or gen AI for short. Gen AI is defined as intelligent systems that have the capability
7
to generate new text, images, and other forms of media by utilising pattern and
structure recognition from training data (Sengar et al., 2024). OpenAI was founded
two years before the publication from DeepMind. In the fall of 2022, they publicly
released ChatGPT (OpenAI, 2022), marking a paradigm shift in the capabilities of
computers to perform complex tasks at a human level. By 2023, the major companies
and universities were releasing their own LLM model, with several models published
weekly (D. Zhang et al., 2024).
Gen AI has already been adopted across numerous industries and applications, in-
cluding education, media, marketing, finance, accounting, healthcare, manufacturing,
and logistics (Google Cloud, 2024; Rosário, 2024). Goldman Sachs estimates that
more than 200 billion dollars will have been invested in AI by 2025 (Goldman Sachs,
2023). The impact that generative AI is having is already at a tremendous scale, and
it can only be forecasted to continue. A Harvard Business Review article highlighted
already in 2019 that AI will add 13 trillion dollars to the global economy by the end
of the 2020s (Fountaine, McCarthy, and Saleh, 2019). Bain & Company AI survey
from 2024 indicates that around 50% of companies have integrated or are currently
integrating generative AI capabilities into their products or services (Rapoport et al.,
2024). The adoption of gen AI has been one of the fastest technological revolutions
and is impacting a wide range of industries and people.
2.2 Generative AI in Products
The following section examines how companies are integrating generative AI into their
product offerings, highlighting recent trends and patterns in adoption. A McKinsey
& Company survey on AI implementation into business processes and products,
displayed in Figure 1 shows a steady increase of AI adoption over the past 8 years
and a doubling of generative AI implementation. Companies are applying gen AI in
various business functions and products.
To further understand the development of gen AI products, one can turn to Y
Combinator (YC). The prominent startup accelerator in California that has founded
some of the most notable companies in the last two decades (The Economist, 2015;
Geron, 2013). YC has consistently provided a valuable lens for examining emerging
technology trends, and generative AI solutions are no exception.
Compiling data from companies that have attended Y Combinator reveals the high
growth of startups that are supplying generative AI solutions to other businesses.
The data was gathered directly from YC’s Startup Directory database, using the
search keys Generative AI and B2B (Y Combinator, n.d.). Figure 2 shows the high
8
growth of companies supplying a gen AI B2B product, particularly the growth from
mid-2022, resulting in over 300 companies in the space by the winter batch of 2025.
During the same time period, Y Combinator classified 1951 companies as B2B, and
out of these, over half (1048) were classified as AI, which includes gen AI-specific
companies.
Figure 1: Organisations that have adopted AI and gen AI in at least one business
function, % of respondents (Singla et al., 2024)
Figure 2: Cumulative numbers of Y Combinator (YC) companies that are classified
as supplying a gen AI product in a B2B context
9
Further, when analysing the industry where gen AI is being implemented, it is clear
that specific industries have been seen as more fit and up for disruption than others.
Figure 3 shows the gen AI B2B companies and the industry they are operating in.
Engineering, Product, and Design is the clear leader with around 30% of all gen AI
companies, followed by infrastructure and sales. Rosário (2024) highlights that gen
AI plays an ever-increasing role in product development by helping developers with
code development and troubleshooting.
Figure 3: Industry classification for Y Combinator (YC) companies that are classified
as supplying a gen AI product in a B2B context
Given that YC has only been accepting between 100 and 400 companies yearly, with
less than half being classified as B2B (Y Combinator, n.d.), the actual number of gen
AI companies operating globally is several magnitudes larger than these numbers
suggest. Still, this data gives an overview of both the increase in companies supplying
gen AI solutions and the industries and use cases being targeted.
On top of the growth in startups targeting the gen AI space, the large established
companies are also incorporating gen AI into their current product offering. For ex-
ample, GitHub launched its programming co-pilot already back in 2021 (N. Friedman,
2021). 2023 saw releases from the CRM behemoth Salesforce announcing Einstein,
a CRM copilot that directly integrates into its product (Salesforce, 2023). Further,
Adobe introduced its creative gen AI product Firefly (Adobe, 2023). 2024 saw the
incorporation of gen AI into SAP, the global ERP provider, introducing gen AI
features into its current product offering in addition to new products such as Joule
Studio (SAP, 2024).
10
2.3 Generative AI in Services
This section focuses on how generative AI is being implemented in the professional
services industry. Akter et al. (2023) highlights that AI is going to have a trans-
formative impact on the services industry. However, they conclude that current
academic research is scarce when it comes to the area. AI innovation in services
can be defined as a technology that can learn and improve while meeting customer
expectations via interconnected systems (Akter et al., 2023).
Researchers such as Rosário (2024), Pattanayak (2020), in addition to consulting
companies such as Simon Kucher (Biermann and Petersen, 2024a) note that Gen AI
has started to make its way into the professional services industries, such as business
consulting, accounting, law, finance, marketing, and healthcare. Law and tax firms
are also starting to adopt gen AI technologies (Warren et al., 2024). The impact gen
AI is having on professional services industries is also more profound than previous
technologies and it is now reaching a state of productivity (Biermann and Petersen,
2024a).
For example, the auditing giant Ernst & Young (EY) has incorporated AI into the
auditing process. They argue that with AI they can supply higher quality services
and increase the value they provide to their customers (Delarue and Jeschonneck,
n.d.). Nevertheless, according to research conducted by Thomson Reuters, gen AI
is yet not widely adopted among professional services firms (Warren et al., 2024).
Still, the report indicates that 23% of the professionals surveyed indicate that their
organisations are incorporating gen AI into the work. According to a Simon Kucher’s
survey, 27% of professional services companies are already piloting gen AI tools with
an additional two-thirds planning to do so (Biermann and Petersen, 2024b).
Gen AI is providing significant productivity improvements for the professional services
industry by being able to speed up the work (Gotts, 2024). As many professional
services firms rely on billable hours as their main way of generating revenue, this
productivity increase can require substantial shifts in their business models. However,
over 50% of legal, tax, and accounting professionals believed that gen AI is not a
major threat and is not going to affect the price they are charging clients (Warren
et al., 2024). Still, over 10% see gen AI as a major threat to the firm’s revenue, while
a third see it somewhat as a threat. 10% of tax and law firms believe that gen AI
will decrease the ability to bill clients. On the other hand, over 20% believe that
gen AI can help firms increase the rates they are charging customers. Nevertheless,
for the time being, the majority of companies see gen AI as a way to reduce cost as
opposed to increasing revenue (Biermann and Petersen, 2024b).
11
Researchers have also highlighted that gen AI can increase the gap between large
consulting providers and small ones, due to the ability of larger ones to invest,
understand, and deploy generative AI (Eulerich et al., 2024). Thus, the professional
services industry may move towards organisations that have the capability and capital
to harness gen AI solutions. It can also result in a consolidation of the top providers
of services. On the other hand, smaller firms have the ability to quickly adopt
new technologies that can considerably drive down the cost and increase the value
that they supply. Professional services are generally constrained by labour-intensive
requirements, high billable hours, and access to privileged information (Kaplan,
2024). All of these factors are disreputable with gen AI solutions.
Yet, figuring out how to price generative-AI work within professional services is still
proving difficult. It is clear that firms’ revenue models are going to be affected by
gen AI (Biermann and Petersen, 2024b; Kaplan, 2024), and there is a need to start
asking the question of how services should be priced when the industry integrates
generative AI.
12
3 Literature Review
This section introduces the reader to the current literature in the field of value and
pricing. The first section 3.1 introduced the reader to the different parts of value.
Second, section 3.2 bridges the gap between value and price. Lastly, section 3.3
introduces the reader to pricing and the different pricing strategies and pricing models
used.
3.1 Understanding Value
A company’s success is strongly tied to its ability to create customer value (Graf
and Maas, 2008). Thus, comprehending value and how it is perceived has always
been important for both researchers and business leaders. This section discusses the
economic or business value and how it is tied to a company’s ability to generate
revenue.
In its simplest form value can be thought of as the benefits that customers seek to
gain from a product or service (Flint, Woodruff, and Gardial, 1997). Zeithaml (1988)
defines four definitions of consumer value: value is low price, value is whatever I
want in a product, value is the quality I get for the price I pay, and value is what
I get for what I give. From these definitions of value, one can see that value is
subjective and tied to what each person tries to gain from a product or service.
Gale (1994) builds on Zeithaml definition and highlights that customer value is
the perceived quality adjusted for the product’s price. This view incorporates the
previous elements of value into a standardised form. However, other scholars have
indicated that companies can no longer only compete on quality (Woodruff, 1997).
Thus, the understanding of value is more complex than scholars initially thought.
Further, Woodruff (1997) combines prior definitions of customer value, including
Zeithaml (1988) and incorporates additional research on the perception of value.
They define customer value as a customer’s perceived preference for and evaluation
of those product attributes, attribute performances, and consequences arising from use
that facilitate (or block) achieving the customer’s goals and purposes in use situations.
This definition indicates that value is a function of several underlying factors or
product features that the customer perceives as vital for the goal they are trying to
achieve. Further, it indicates that value can be both an enhancing and a destroying
factor in pursuing the customer’s goal.
Scholars have also highlighted the importance of apprehending that value is a dynamic
concept (Woodruff, 1997; Leszinski and Marn, 1997; Flint, Woodruff, and Gardial,
13
1997; Graf and Maas, 2008) that changes as situations and time progress. It is also
essential to understand the product characteristics customers value and how they
can influence each attribute’s importance (Parasuraman, 1997). On the other hand,
Flint, Woodruff, and Gardial (1997) also argues that value can stay constant as long
as people’s situations remain similar.
The following sections first give a brief overview of how value has changed over the
centuries and how it has been tied to the current state of the world at the time.
After that, customer-perceived value is discussed, followed by a product view of
value. After that, the quantification and how a company should determine value are
presented. The last section presents how value is perceived for generative AI work
versus human-generated work.
3.1.1 Historical View on Value
Understanding how economists have conceptualised value is an essential driver in
understanding value and value capture. The following section is based on the book
Economics evolving: A history of economic thought by Sandmo (2011).
The modern discussion of “value” in economics begins with the classical economic
theory. In The Wealth of Nations (1776), Adam Smith argued—alongside contem-
poraries such as David Ricardo—that, in the long run, the exchange value of a
commodity is anchored in the quantity of labour required to produce it, a view later
labelled the Labour Theory of Value. Smith also distinguished a good’s natural
price, built up from wages, rent, and profit, from the market price that prevails
at any moment, thus recognising early on that cost and price need not coincide.
This labour–cost perspective dominated while production was still largely manual; it
already lagged behind the emerging realities of the Industrial Revolution. Still, it
provided the first systematic framework for thinking about price formation.
During the mid-nineteenth century, John Stuart Mill softened the strict cost doctrine
by treating price as the outcome of both supply and demand. The decisive break
came with the so-called “marginal revolution” of the 1870s: William Stanley Jevons
and Carl Menger shifted attention from producers to consumers, recasting value
as a subjective assessment rooted in marginal utility. In the neoclassical view, a
commodity is worth what the last unit is worth to the buyer, not what it previously
cost the seller. This turn from objective labour costs to subjective preferences laid
the groundwork for modern price theory and still underpins contemporary value
discussions.
14
3.1.2 Customer Perceived Value
The introduction of the subjective theory of value introduced by neoclassical economists
is the foundation for customer-perceived value (CPV). When it comes to customer
value, it is essential to recognise that value is determined by the consumer and not
by the seller (Woodruff, 1997). Customer-perceived value for a product comprises
several attributes that together work to form the understanding of value. Thus, it
is the customer’s perceived value that guides a purchasing decision for a product.
Managers can leverage CPV as a performance measure to understand the drivers
behind customer value (Blut et al., 2024).
Zeithaml (1988) introduces a means–end model that links consumers’ perceptions
of product attributes with their personal values. The framework highlights how
perceived quality and the sacrifices associated with price interact to shape the
overall value consumers derive from a product (Zeithaml, 1988). By modifying these
underlying drivers—improving quality, lowering monetary or non-monetary sacrifices,
or both—firms can enhance or diminish customer value.
Building on this foundational understanding of customer-perceived value and its
determinants, it is crucial to examine how these factors influence consumer behaviour.
Although the means–end model clarifies the relationship between quality, sacrifice,
and value, it does not specify the exact attributes that constitute value in a given
context.
Blut et al. (2024) conducted a large meta-analysis on the CPV model developed by
Zeithaml and examined underlying attribute and their relative importance in driving
customer value. The conceptual model developed by Blut et al. (2024) offers a more
tangible view of customer value and the underlying drivers. In addition, it introduced
the reader to the outcome state of value; word-of-mouth (WOM),satisfaction, or
repurchasing intention. Higher customer value has been shown to lead to higher
outcome success along all the three aforementioned areas (Vieira, 2013). Thus,
customer value results in a broader outcome than just the value for that product
and can help to drive future revenue.
Figure 4 gives an overview of the attributes that were studied. All the benefits
variables were found to be statistically significant in addition to sacrifice variables;
price, effort, performance risk, and security risk (Blut et al., 2024). Particularly,
convenience/efficiency,excellence, and relational benefits were found to have stronger
positive effects on overall value in a B2B setting. Thus, managers should focus
on these attributes to increase customer perceived value. However, it has to be
15
highlighted that the overall explainability of all the variables used explained just
under 50% of the overall value (Blut et al., 2024). As such, more underlying factors
affect customer value than the ones studied.
Figure 4: Conceptual model of drivers behind customer perceived value. Including
drivers increasing customer perceived value and drivers decreasing value. Based on
Blut et al. (2024)
Woodruff (1997) introduced the notion that customer value should be thought of as
a hierarchy, where the value evolves from the features being offered to customers to
the value gained by the customers by using the product to solve their problems. This
model, called the Customer Value Hierarchy Model, is displayed in Figure 5. Further,
it explains how customers perceive value and how customer satisfaction is tied together
on each level. Customers evaluate products at three levels: attributes, (features and
performance), consequences, (practical benefits in use), and goals (alignment with
broader objectives). Satisfaction is assessed similarly, whether the product meets
expectations at each level.
Contrary to what Zeithaml (1988) found that some customers buy on price alone,
Leszinski and Marn (1997) argue that this is not the case. Instead, they argue, that
customers are buying according to their personal perceived value of the product
and that customer value is the difference between the value the customer gets from
16
Figure 5: Customer Value Hierarchy Model. Connecting desired customer value with
received value. (Woodruff, 1997)
the product and the price they are paying for it (Leszinski and Marn, 1997). This
relationship is shown in Equation 1.
Customer Value =Customer Perceived Benefits −Customer Perceived Price (1)
As such, there is always a trade-off between the perceived value and perceived benefits,
and each customer has their own interpretation of the relationship. The relationship
does not have to be linear. Some products and services have a higher tendency to lead
to a higher customer value for a slight increase in the perceived benefits (Leszinski
and Marn, 1997). They also highlight that there are price-capped customers, who,
no matter the increase in perceived benefits, are not willing to spend more to get it.
3.1.3 Product Value
The previous section evaluated customer value from a customer’s perspective. Graf
and Maas (2008) highlights that value can also be considered from a company’s
product perspective. Zeithaml (1988) argue that value in a product is a very personal
and idiosyncratic aspect, and consumers are looking at more than just the cost of
the product. Nagle and Holden (2018) establishes the value of a product as the
price of the customer’s best alternative, the reference value, plus the worth of the
17
differentiation for that product over other options, the differentiation value. See
Equation 2. This view on value is also shared by Graf and Maas (2008).
Product Value =Reference Value +Differentiation Value (2)
To better understand the pillars of product value, it can be broken down into four
segments that together are referred to as the value stick (Stobierski, 2022). The value
stick comprises four parts: the customer’s willingness to pay, the price charged, the
fully loaded cost, and the company’s willingness to sell point. The value stick graph is
displayed in Figure 6. The value stick helps to think about the relationship between
the different aspects that make up the product value and how it is connected to the
price charged. This strategy is referred to as value-based pricing and is discussed in
detail in the pricing Section 3.3.2.
Figure 6: Value stick graph, working up from willingness to supply (WTS) to
willingness to pay (WTP) and showing value capture areas. (Stobierski, 2022)
Companies can increase what is referred to as customer delight by adding product
features or services that customer perceives as fundamental, beneficial or of unique
value to them (Ravald and Grönroos, 1996). Ravald and Grönroos (1996) argue that
besides adding features, which are most often associated with a cost for the producer
and could reduce the firm margin in the value stick, companies can instead focus on
lowering the customer-perceived sacrifice. On top of reducing the actual price charged
18
this can involve making the purchase more convenient, reducing uncertainties e.g.,
delivery time, or cognitive effort related to the purchase for the customer (Ravald
and Grönroos, 1996).
Consequently, customer value is not only a function of the cost versus benefit but
also includes other indirect factors that can affect the perceived value. However, as
mentioned earlier, value is not static. The perceived value to the customer varies
based on the context and circumstances under which they are considering the product
(Woodruff, 1997).
3.1.4 Quantifying and Determining Value
The previous sections introduced value understanding, how to think about value in
different contexts, and some initial views on value quantification in various settings.
To further understand the drivers of value, one can turn to Grönroos (1990) CPV
formulas. The five equations below quantify customer-perceived value in a service-
based setting.
CPV1=Episode benefits +Relationship benefits
Episode sacrifice +Relationship sacrifice (3)
CPV2=Transaction value ±Relationship value (4)
CPV3=Core solution +Additional services
Price +Relationship costs (5)
CPV4=Core value ±Additional value (6)
CPV5=Long-term revenue-generation support
Price +Relationship costs (7)
From the above Equations 3 to 7, it is possible to determine different strategies that
can be used to drive up customer value. For example, the price in the denominator is
a factor of value creation. As such, companies can increase the CPV by reducing the
total cost associated with acquiring and using the product or service (Salem Khalifa,
2004). As discussed earlier, companies can increase customer value by adding product
features (Ravald and Grönroos, 1996). The total perceived value can increase by
adding value to the core product, and any additional features added can either
increase or decrease the value perceived. In a services-based business, managers can
more easily emphasise the different aspects of the offering to drive up the overall
perceived customer value (Blut et al., 2024).
Woodall (2003) introduced five key concepts that determine the value for a customer.
First is what is termed net value for the customer, which is the balance of benefits
19
versus the sacrifices. Second is derived value for the customer, which refers to the
use or experience outcome from the product. Third is the monetary difference from
an objective reference point, this is called rational value for the customer. Fourth, is
the sale value for the customer, which is the option determined primarily on price.
Lastly is what is called marketing value for the customer, which is the perceived
value for the product attributes.
To better help managers find and evaluate the underlying drivers of customer value
Woodruff (1997) introduced a customer value determination process framework. The
framework is displayed in Figure 7. The process framework is designed to help
managers understand what value the customers are expecting, rank the importance
of each value, and then understand how well the company’s product is fulfilling these
values. Lastly, value is not a static object as highlighted by Parasuraman (1997),
Graf and Maas (2008) and others, the framework also highlights the importance of
understanding how the customer values are expected to change in the future. This
understanding helps companies create new value strategies for their clients as time
progresses (Woodruff, 1997)
Figure 7: Customer Value Determination Process (Woodruff, 1997)
Further, (Sheth, B. I. Newman, and Gross, 1991) investigated human value drivers
when it comes to making product decisions. They define consumer choice behaviour
as consisting of five parts. These are functional value, condition value, social value,
emotional value, and epistemic value. A person’s decision may be influenced by one
or more of these consumption values. Functional value is established through the
20
presence of key functional, utilitarian, or physical attributes and is assessed based
on a set of relevant choice attributes. Social value is the perceived utility acquired
from that alternative as seen by a specific group of people, and it is based on how
people perceive it. Emotional value is the benefit gained from an alternative’s ability
to evoke emotions or affective states. Epistemic Value is the utility acquired that
arouses curiosity, provides novelty or satisfies a desire for knowledge. Lastly, the
conditional value is the utility acquired due to the specific situation or the set of
circumstances present at the time. Understanding the human value dimensions helps
business leaders identify sources of consumer value and tailor offerings accordingly.
3.1.5 Perception of Value for Generative AI Created Work
The previous sections have discussed the measurement of value in an economic
context. This section discusses the perceived value of generative AI-generated work.
It incorporates both an economic value perspective to the extent that is possible,
in addition to looking at how gen AI can be seen in a social value setting when
comparing it to human-generated work.
Bellaiche et al. (2023) conclude in their extensive study that AI-produced art tends
to be valued less than human-created art along four dimensions, namely liking, beauty,
profundity, and worth. The study indicated that humans prefer art that is labelled
as human-created compared to art labelled as AI-created. The authors note that
this was particularly strong in areas that communicated a deeper meaning of the art,
such as profundity and worth. Other studies have indicated that when text is labelled
as AI-written versus human-written, it is seen as less credible (Lermann Henestrosa
and Kimmerle, 2024).
Other research settings have found similar results. In evaluating human and AI-
generated poems, people slightly preferred the human-generated version (Köbis and
Mossink, 2021). An interesting discovery from the researchers was that the difference
in preference did not change regardless of whether participants were informed or not
about the origin of the poem (AI or human). Generally, people tend to be averse
toward algorithmic decision-making when they are aware of it (Köbis and Mossink,
2021). However, in this setting, the participants’ aversion toward AI-generated text
when knowing that it was AI-generated was not higher than when they did not know
it.
It has also been found that the setting where people first get a taste of AI-generated
content substantially affects their view of it. If people have negative surprises when
using AI, it reduces a person’s willingness to use the technology in the future (Hong,
21
2021). In addition, the level of consumers’ prior knowledge affected the perceived
value of a gen AI-designed product. A higher level of consumer knowledge decreased
the willingness to pay for generative AI design (H. Zhang, Bai, and Ma, 2022).
Researchers have also suggested that generative AI can enhance the perceived value
of AI-generated content. Particularly the value of intangible and high-involvement
products saw a boost when AI was used in the creative process (Tigre Moura,
Castrucci, and Hindley, 2023). The researcher particularly highlights that the use of
AI has a positive effect on the perception of novelty when employed in the design
process. In addition, consumers have indicated that they are prepared to pay more
for utilitarian products that have been designed by AI (H. Zhang, Bai, and Ma,
2022).
3.2 Customers’ Willingness to Pay
Customer-perceived value directly influences willingness to pay (WTP), which in
turn determines the effectiveness of different pricing strategies. A higher perceived
value allows for premium pricing, while a mismatch between price and perceived
value can result in revenue losses. Companies need a deep understanding of WTP to
properly benefit from a pricing strategy (Breidert, Hahsler, and Reutterer, 2006).
Homburg, Koschate, and Hoyer (2005) ties customer value and WTP into a model
and showcases how customer satisfaction affects a customer’s willingness to pay. As
discussed in the previous section, customer satisfaction results from customer value,
as highlighted by Blut et al. (2024).
There is a strong relationship between WTP and customer satisfaction and it takes
the form of an inverse S-shaped curve (Homburg, Koschate, and Hoyer, 2005). The
shape of the curve indicates that customers have the most substantial incentive to
either pay more or less for a product at the extremes of their satisfaction level, be it
high or low. As such, the influence for companies to benefit from a higher willingness
to pay relies on providing high customer satisfaction. Even with a slight increase in
satisfaction levels, it does not affect the perceived WTP.
Companies can utilise this understanding to charge a premium from satisfied cus-
tomers (Homburg, Koschate, and Hoyer, 2005). The researcher highlights that this
is particularly true in professional services, where the offering is non-standardised
and usually tailored to one customer. The premium can also be seen from Equation
1. If the perceived customer benefits grow faster than the perceived price, companies
can increase prices while ensuring that customers still obtain the additional higher
total value.
22
Research is still emerging regarding the willingness to pay for gen AI products. Given
that gen AI can substantially enhance performance, it is not far-fetched to assume
that companies are willing to pay to gain the efficiency of gen AI substantially. For
example, recent research on the implementation of gen AI in customer support roles
indicates a 15% increase in worker productivity and even higher among younger
professionals (Brynjolfsson, Li, and Raymond, 2025)
As discussed in Section 3.1.5 the perception of gen AI generated outcome also varies.
This variation can have a significant impact on the willingness to pay for gen AI
products and services. The difference in generational perspective on generative AI
may also influence organisations’ willingness to adopt and invest in generative AI
products and services. Particularly, younger professionals that have adopted gen AI
products at a higher rate can be more willing to pay for the products and services
that seasoned employees (Bick, Blandin, and Deming, 2024).
3.3 Pricing
The previous section examined the concept of value and explained how consumers’
perceptions of it affect their willingness to pay. Continuing on this discussion leads
to pricing and how companies should most effectively capture the value they are
supplying to customers. This chapter gives an extensive overview of pricing strategies
and pricing models in different settings. First, an overview of pricing practices, the
most common current pricing strategies used, and the pricing models deployed in
digital businesses is discussed. After services-based pricing is introduced, followed by
pricing of new products, and pricing in the age of gen AI. Lastly, price discrimination
is examined.
3.3.1 Pricing Overview
Quantifying customer value is one of the most critical capabilities for companies to get
right (Hinterhuber and Snelgrove, 2022). Increasing prices, assuming everything stays
the same, have a far greater impact on profitability than the same improvement in
other operating levers such as sales volume, variable cost, or fixed cost (Hinterhuber,
2004; Baker, Marn, and Zawada, 2010). To understand the importance of pricing,
one can turn to a simple example. A 1% increase in price leads, on average, to an
8% rise in profits (Baker, Marn, and Zawada, 2010). This ratio is similar to what
other scholars have identified (Hinterhuber, 2004). The exact ratio depends on the
firm’s operational structure.
23
Pricing can be understood from multiple perspectives. Firstly, price can be looked at
from the customer’s perspective as the sacrifice given to obtain a product (Zeithaml,
1988). Thus, the price can be thought of as the value provided to the customer, and
price relies on the customer recognising the value of the product (Nagle and Holden,
2018). From a company’s perspective, the price must cover the costs associated with
the product or service. Monroe (2003) relates these two sides and concludes that
the price lies between the maximum customer value, price ceiling, and the cost of
the product, price floor. This is graphically displayed in Figure 8. In reality, the
price difference is pushed closer together due to competition and company objectives,
leading to a narrower price band than would be suggested by the price floor and
ceiling.
Figure 8: Conceptual Orientation to Pricing. Showing the relationship between the
costs of supplying a product and the value delivered, in addition to market factors
determining the price level. (Monroe, 2003)
Baker, Marn, and Zawada (2010) further connects the relationship between pricing
and value, through what they refer to as Value Profiles, displayed in Figure 9. It
shows the relationship between perceived benefits or value and the perceived price.
In zone A in Figure 9, the customers perceive a common value delivered by the
company, but the perceived price varies greatly. In zone B, the customers have a
consistent perception of the price position, but customers perceive a broad spectrum
of benefits. The graph helps companies to map the customer’s understanding of
their products. A larger ellipse indicates that the customers are unclear about the
product’s price or benefit position (Baker, Marn, and Zawada, 2010). As such, a
24
company’s goal is to minimise the ellipse along both dimensions. Companies can
also segment the customers into smaller groups to minimise the error and, as such,
tailor the offer to each group better. This strategy is generally referred to as price
discrimination, and it is discussed at the end of this chapter.
Figure 9: Value Profile Graph, showing how customers understand a company’s
product offering for price over value. Adapted from Baker, Marn, and Zawada (2010)
Moving to pricing strategies, researchers and the industry generally recognise three
main pricing strategies. These being cost-based,competition-based, and value-based.
For a comprehensive list of researchers recognising these three pricing strategies, see
Guerreiro and Amaral (2018).
Cost-based refers to setting prices based on the cost of production for the product or
service (Hinterhuber, 2008). This is a traditional pricing strategy. Its main strength
lies in the available data needed to set a cost-based pricing. However, the strategy
lacks depth and does not consider customer-perceived value or other market factors.
Competition-based pricing strategy determines prices based on competitors price level
(Hinterhuber, 2008). Again, the strength of competition-based pricing lies in the
25
availability of data and a simplistic model. On the other hand, it ignores customers’
willingness to pay, and any product differentiation is not accounted for.
Lastly, value-based pricing is based on the perceived customer value of the product
or service (Hinterhuber, 2008). The advantage of a value-based pricing strategy lies
in the fact that the company can charge for the actual value provided. However,
quantifying and measuring the value can be challenging. In addition, it can lead to
high prices compared to the market. Table 1 gives an overview of the three primary
pricing strategies, their strengths and weaknesses.
Table 1: Comparison of the three most common pricing strategies. (Hinterhuber,
2008)
Cost-based pric-
ing
Competition-
based pricing
Customer value-
based pricing
Definition
Prices are based
on cost accounting
data.
Prices are set
based on competi-
tor price levels.
Prices reflect the
value delivered to
customers.
Examples
Cost-plus, mark-
up, target-return
pricing.
Parallel, umbrella,
penetration/skim,
market-based
pricing.
Perceived value
pricing, Perfor-
mance pricing.
Main strength
Data readily avail-
able.
Data readily avail-
able.
Considers cus-
tomer perspective.
Main weakness
Ignores competi-
tion and customer
demand.
Ignores customer
willingness to pay
and differentia-
tion.
Data is hard to
obtain and inter-
pret. May lead to
high prices if not
communicated ef-
fectively.
In addition to these three pricing strategies, one can recognise price skimming
and penetration pricing (Tellis, 1986) as two other important pricing strategies.
Price skimming refers to a strategy where a company starts with a higher price
but knowingly reduces it each period (Noble and Gruca, 1999). The opposite is
penetration pricing, where companies start with a low price to gain market share
and subsequently start to raise it (Noble and Gruca, 1999).
26
The industry recognises these five pricing strategies as the primary pricing strategies
used in practice (Deland, 2022). However, both price skimming and penetration
pricing are less common. Together they comprise less than 3% of published research
in the field (Hinterhuber, 2008). On the other hand, both strategies have been
popular for new product launches, which are discussed below.
Pricing strategy has yet to see a uniform usage in the literature. The difference
between price strategy and pricing models tends to be blurred and mixed up in
literature (Ingenbleek and Van der Lans, 2013). Pricing strategy is defined as
observable in the market, whereas pricing models are hidden within the organisation
(Ingenbleek and Van der Lans, 2013). Noble and Gruca (1999) would define value
and competition-based pricing methods as pricing models, whereas, for example,
Hinterhuber and Liozu (2012) and Guerreiro and Amaral (2018) refer to them as
price strategies. This paper generally does not distinguish between the terminology.
It discusses pricing strategy as a high-level strategic view of the first pricing decisions
companies make, whether visible in the market or not. Second, this paper refers
to pricing models as a more tactical implementation that professionals use when
discussing monetising their products. Among professionals, it can also be noticed that
they mix both pricing strategies with pricing models when talking about monetisation.
As with customer value, pricing is also a dynamic concept that changes as time
progresses (Hinterhuber, 2004). Thus, companies need to stay agile and constantly
review their pricing to ensure that they remain competitive and can respond to
changing market conditions.
The modern challenge of pricing is related to the transparency of the cost side. For
many products, customers can find what the underlying costs are by searching the
internet (H. Friedman and Gerstein, 2014). Thus, the comparable advantage of
cost-based pricing strategies is diminishing, and companies are turning towards other
pricing strategies instead.
The following sections discuss value, competition, and cost-based pricing in more
detail.
3.3.2 Value-Based Pricing
Value-based pricing is, at its essence, a very simple pricing strategy to understand.
The price is set based on the value that the product or service delivers to customers
(Hinterhuber, 2008). Value-based pricing determines the price ceiling that customers
are willing to pay for the product before considering other alternatives (Ingenbleek,
27
Debruyne, et al., 2003). To successfully compete with a value-based pricing strategy,
firms need to be able to differentiate the product and quantify the value it is delivering
to customers (Hinterhuber and Snelgrove, 2022).
Many companies in several sectors recognise value-based pricing to be the best pricing
strategy for several different industries and products (Ingenbleek, Debruyne, et al.,
2003; Hinterhuber, 2008; Hinterhuber and Liozu, 2012) and a key tool to drive
increased profitability (Liozu, 2017; Hinterhuber and Snelgrove, 2022). Value-based
pricing has also emerged as a key strategy for launching new innovative products to
the market (Ingenbleek, Frambach, and Verhallen, 2013).
Figure 10 illustrates the power of value-based pricing. The figure shows the rela-
tionship between the price and value. As long as the customer value difference for a
company’s products is higher than the price difference, the company should have a
competitive advantage and be able to drive up the price until the price difference is
the same as the value difference. However, the risk is that a competitor undermines
the company on value differentiation and price.
Figure 10: Value quantification and value-based pricing (Hinterhuber and Snelgrove,
2022)
Implementing value-based pricing is far from straightforward. Researchers have
identified only a limited number of companies that have successfully implemented
value-based pricing (Hinterhuber, 2008). Yet, the strategy has gained traction in
practice, with approximately 25% of companies across various industries reporting
value-based pricing as their primary pricing strategy (Liozu, 2017).
28
However, challenges remain when adopting a value-based pricing strategy. The
two main challenges are value assessment and market and customer segmentation
(Hinterhuber, 2008; Liozu, 2017). Quantifying what price to charge is difficult if a
company do not know what value customers see in the product. In addition, accessing
reliable data for value quantification and understanding the customer’s willingness
to pay can be challenging (Hinterhuber and Liozu, 2012). Lastly, companies should
group customers based on their perceived value of the product (Hinterhuber, 2008),
this better helps to segment customers into similar groups and charge similar prices
within a group. Thus helping to minimise the ellipse area in Figure 9. By doing this,
companies can take advantage of price discrimination between the different groups,
which is discussed towards the end of the chapter. Other challenges highlighted for
value-based pricing strategy are value communication (Hinterhuber, 2008; Liozu,
2017), sales force management (Hinterhuber, 2008) and finding competitive price
levels and value drivers (Liozu, 2017).
3.3.3 Cost-Based Pricing
Professionals have long implemented cost-based pricing due to its simple structure.
Cost-based pricing builds on top of data from the cost accounting department
(Hinterhuber, 2008). Companies then add a margin on top of the total costs that
are associated with the product or service. Cost-based pricing helps companies to
understand the price floor of the product and thus the minimum price they need to
charge to make a profit (Ingenbleek, Debruyne, et al., 2003).
However, the downside of cost-based pricing is that companies tend not to understand
the value they are delivering to customers and as such can not specify the accurate
price for the value the product delivers (Simon, Butscher, and Sebastian, 2003;
Hinterhuber and Liozu, 2012). On the other hand, businesses facing difficulties in
quantifying value are more likely to opt for a cost-based pricing strategy (Amaral
and Guerreiro, 2019), as it is a simpler pricing strategy, requiring less resources to
develop.
Further, cost-based pricing also ignores the competitive landscape (Hinterhuber and
Liozu, 2012), potentially leading to major underpricing compared to the market. In
addition, as mentioned, cost has become very transparent in the digital age, and as
such, the competitive advantage of cost-based pricing has been reduced.
Another challenge with cost-based pricing in industries where cost is affected by
volume is that companies have to determine the unit cost before setting the price, and
to do this, they have to assume a predefined volume (Nagle and Holden, 2018). Thus,
29
it assumes that volume is not linked to price, a rudimentary flaw. The fundamental
economic relationship tying quantity and price together was theorised back in the
18th century by the classical economists.
It is also recognized that cost-based pricing leads to sub-optimal company performance,
underpricing, and ultimately harms profits (Hinterhuber, 2008; Nagle and Holden,
2018). Further, companies pursuing a cost-based strategy can rarely charge premium
prices for their products or services (Amaral and Guerreiro, 2019) and thus losing
out on higher profits.
On the other hand, Ingenbleek, Debruyne, et al. (2003) argues that cost-based pricing
works if the competitive intensity is high, but not as well when the competitive
intensity is low. In everyday situations, companies do not have to worry about the
margin on their products, particularly if a value-based strategy is used. However,
in a competitive situation when prices are being pushed down, having a better
understanding of the cost structure of the product or service can become a competitive
advantage (Ingenbleek, Debruyne, et al., 2003). In addition, in some industries, it can
be challenging for customers to understand value-driven pricing or that customers
are used to fixed cost pricing (Amaral and Guerreiro, 2019). Thus, in some specific
circumstances, a cost-based strategy can be used.
3.3.4 Competition-based Pricing
Competition-based pricing is the most widely adopted pricing strategy across sev-
eral industries (Liozu, 2017) and also the most researched (Hinterhuber, 2008).
Competition-based pricing strategy uses competitor prices as the main driver for the
company’s own price level (Hinterhuber, 2008).
Competition-based pricing can be effective as it enables companies to first understand
the competitive price position for its products or services that the customers see
as comparable and then set the price below to increase the likelihood of purchase
(Ingenbleek, Frambach, and Verhallen, 2013). A competition-based strategy also
enables companies to evaluate competitors’ offerings and, based on the position,
set an appropriate price. For instance, when a company offers a product with
marginally fewer features, it is advisable to undercut the competitor’s price modestly.
(Ingenbleek, Debruyne, et al., 2003).
Within competitive pricing, there are several strategies for companies to pursue. First
is cooperative pricing, which involves a complete price match with the competitors
(Deland, 2022). This is also the strategy that maximizes the return within an
30
industry (Griffith and Rust, 1997). Second, an aggressive pricing strategy also known
as envious pricing involves keeping a price distance from competitors and ensuring
that the company has the lowest price (Griffith and Rust, 1997; Deland, 2022). Lastly,
dismissive pricing can be used by market leaders who can charge a premium and, as
such, do not have to directly compete with competitors (Deland, 2022).
The disadvantage of competition-based pricing is the ignorance of the demand side
of the product (Hinterhuber and Liozu, 2012). In addition, with competition-based
pricing companies price very similarly, which means that it is more difficult to compete
on differentiation (Hinterhuber, 2008). Competition-based pricing can also lead to
price wars, driving down the price and evaporating all profits from an industry.
3.3.5 Pricing Models for Digital Businesses
The fundamental pricing strategies discussed above define how prices are determined.
To further understand how customers are charged, one has to turn to pricing models.
This chapter introduces modern pricing models and their connection to fundamental
pricing strategies.
The research on pricing models in the digital business landscape is limited. Thus, the
following section provides a non-exhaustive overview of the different pricing models
collected and cross-referenced from several industry websites based on Y Combinator
startup directory (Y Combinator, n.d.), in addition to the author’s own experience
in the field. Many of the mentioned pricing models can also work as recurring or
subscription-based models.
Flat-rate pricing: one pricing structure for access to all features or to a pre-
determined set of features. The price is not dependent on how much the customer
uses the product, but it can be limited to the number of users.
Tiered/ feature-based pricing: multiple pricing levels or product packages offered,
each with a distinct set of features and benefits. This model enables customers to
select the features that provide the highest value to them.
Usage-based/ pay-as-you-go pricing: customers are charged based on their actual
usage of a product or service. This pricing model aligns costs with consumption, and
different consumption metrics such as API calls, transactions processed, etc, can be
used.
Per-user pricing: customers are charged based on the number of users or seats
accessing the product. The model connects the usage to the estimated cost of serving
the users and works particularly well in industries with stable costs.
31
Outcome-based pricing: the company charges based on the actual value delivered.
Different metrics can be used to define the value delivered, such as time reduction,
resolutions successfully solved, or monetary savings.
Credit/ token-based pricing: customers purchase credits at a predefined price.
The credits can later be used within the product to access certain features. Different
features can consume different amounts of credits.
Freemium pricing: a basic version of the product is offered for free, allowing users
to access a limited amount of features at no cost. Additional or more advanced
features or functionalities are available for a paid upgrade.
3.3.6 Services-Based Pricing
Professional services businesses such as consulting companies have generally used
hourly billing for their services (Monroe, 2003). This is a straightforward way of
charging the client based on the number of hours a human is putting into the project
work. A normal hourly based contract can be tied to cost-based pricing as the
professional service firm uses the cost associated with the work and the human labour
and adds a margin to cover the additional costs while ensuring that they can turn a
profit.
In addition, more innovative pricing models have been introduced such as retainer
agreements,productized services,value-based pricing, and pay for results (Clark,
Cohn, and Goldsmith, 2019). Retainer agreements refer to clients paying a flat
monthly fee to get access to the services offered by the firm. Productised services
refer to billing according to a price list, which gives the client access to a specific,
tailored offering that is off the shelf. In this context, Clark, Cohn, and Goldsmith
(2019) refer to value-based pricing strategy as billing the customer based on some
agreed goal and the impact that the engagement has on a predefined business metric.
Value-based pricing has gained momentum and is been proposed as the new way of
pricing professional services (Rautakoura, 2024). Lastly, pay for results is somewhat
tied to value-based pricing, but is stricter. Pay for results is based on a performance
threshold and if the threshold is not reached, the client does not pay anything (Clark,
Cohn, and Goldsmith, 2019).
Consulting companies have already recognised the changes required when generative
AI is being increasingly implemented into day-to-day work. The reduced work hours
needed to complete the same amount of work with AI tools decrease the ability to
bill clients. Particularly, companies relying on hourly billable hours should start
32
to reconsider their business model (Biermann and Petersen, 2024a). Professionals
working in services-based businesses have also recognised the need to adopt new
pricing arrangements to stay competitive (Warren et al., 2024)
Simon Kucher is already working on new pricing models for professional services in the
age when generative AI is being adopted into the workplace. These are summarised
in Table 2. These pricing strategies are not necessarily new. Nevertheless, it
offers an overview of different pricing strategies that professional services firms are
implementing in the age of gen AI. The various pricing models can be adapted based
on the type of work provided by professionals and the specific classification of AI
products being adopted
Table 2: Summary of pricing models that are emerging for professional services in
the age of gen AI (Biermann and Petersen, 2024a)
Pricing Model Description Challenges
Increase Hourly Rates
AI reduces work hours, so
raising hourly rates main-
tains revenue.
Client resistance if com-
petitors do not follow.
Multi-Dimensional Model
Keep hourly pricing but
add software/technology
fees to cover gen AI invest-
ments.
Accept reduction in work
hours that need to be cov-
ered with software.
Credit-Based Model
Clients purchase service
credits (possibly via sub-
scription) and redeem
them for specific work.
Setting fair credit values.
Output-Based Pricing
Charge based on deliver-
ables rather than time.
Measuring service units ef-
fectively.
Flat Rate Pricing
Charge a fixed price for
services, ensuring pre-
dictability.
Defining service levels
fairly.
Outcome-Based Pricing
Price based on the true
value delivered to clients.
Measuring impact can be
difficult.
There is also an understanding among practitioners that consultancies are unlikely to
cut their fees even after gen AI is adopted (Warren et al., 2024; Howard, 2025). This is
because the firms have invested vast amounts of money into the technology and need
to recoup their investments. This could indicate a move towards a multi-dimensional
33
pricing model. However, there is yet no evidence on which pricing model professional
services firms are implementing.
3.3.7 Pricing of New Products
Pricing is a key determinant for the success of a new product. Researchers have
shown that using the wrong pricing strategy can destroy a new product’s market
advantage (Ingenbleek, Frambach, and Verhallen, 2013). A common mistake among
professionals is to price new products similarly to already existing ones (Piercy,
Cravens, and Lane, 2010). It has even been shown that value-based pricing is a
more important predictor for product success than the product advantage itself
(Ingenbleek, Debruyne, et al., 2003).
Price skimming and penetration pricing used to be the preferred pricing strategies for
new products (Tellis, 1986). However, scholars now argue that companies should also
consider other pricing strategies. Value-based, cost-based, and competition-based
have all been studied in the context of new product launches (Ingenbleek, Debruyne,
et al., 2003; Ingenbleek, Frambach, and Verhallen, 2013).
Ingenbleek, Debruyne, et al. (2003) first studied the effect of the three primary pricing
strategies for new product launches, and Ingenbleek, Frambach, and Verhallen (2013)
extended the research to include a wider variety of industries, including services-based
businesses. The findings under different scenarios, product advantage,product costs,
and competition are shown in Table 3. It can be concluded that value-based pricing
has no disadvantage in any setting compared to the other two strategies. These
findings are similar to what Ingenbleek, Debruyne, et al. (2003) identified in their
first study.
Table 3: The impact of the pricing strategy on the price and product performance
(Ingenbleek, Frambach, and Verhallen, 2013)
Product Relative Competitive Relative Price New Product Performance
Advantage Product Costs Intensity Value- Competition- Cost- Value- Competition- Cost-
based based based based based based
High High High + 0 – + – +
High High Low + 0 – + – –
High Low High + + – + – +
High Low Low + + – + – –
The results highlighted that value-based pricing helps to achieve new product market
performance and get closer to the price ceiling, defined by the value of the product
than the other two strategies (Ingenbleek, Frambach, and Verhallen, 2013). However,
34
it is important to state that companies must educate the customers about the value of
the product and can not expect that customers will immediately recognize the value
and be willing to pay value-based pricing for an innovative new product (Hinterhuber
and Liozu, 2012).
Cost-based pricing should generally be avoided, but can help with new products in
competitive markets. In markets with low competition, competition-based pricing
can be used. Ingenbleek, Frambach, and Verhallen (2013) suggests that competition-
based pricing can be combined with value-based pricing. Particularly, competition-
based pricing can help to undercut reference prices and reach a higher price level if
competitors are pricing well above the cost. Value-based pricing alone does not help
companies determine how low to set a price to boost market performance. Thus,
competition-based pricing can be used to guide a company’s understanding of the
customers’ reference price (Ingenbleek, Frambach, and Verhallen, 2013).
3.3.8 Pricing in the Age of Generative AI
Unlike traditional software, AI-based services introduce new challenges in pricing
due to their unpredictable cost structure, which is related to the inference costs for
running AI models, difficulty in demonstrating and quantifying the direct value, and
their effect on return on investment (ROI). As such, conventional pricing models
might not work with gen AI product features.
Some initial ideas have started to emerge on how to better price gen AI products,
features and services. Regardless, Boston Consulting Group (BCG) indicates that
companies tend to gravitate towards familiar pricing models that have long been used
in software pricing namely consumption based and subscription/ seat based (Pineda
et al., 2024). It is clear that pricing gen AI features remains challenging even in 2025
(Santoro and Hare, 2025), and there is no clear one-size-fits-all pricing model for gen
AI features (Pineda et al., 2024)
Gartner conducted research highlighting how companies should think about pricing
their generative AI product features. They underscored that pricing should be based
on customer value and driven by quantifiable metrics that indicate the ROI for the
specific feature (Santoro and Hare, 2025). The ROI calculations can be based on,
for example, productivity enhancements, reduced time-to-value or risk mitigation
factors.
35
Depending on the gen AI category, different pricing methods have started to emerge.
Table 4 gives a non-exhaustive overview of varying pricing metrics being used for
various product categories (Alexander and Higgins, 2024).
Table 4: Sample of AI Products and Pricing Metrics (Alexander and Higgins, 2024)
Category Example Products Pricing Metrics
AI or ML platform
•
Amazon: AWS Machine Learning
•Baidu: Brain
•Google: TensorFlow
•IBM: Watson
•Microsoft: Azure AI services
•Metered or unmetered
•User-based
•
Transactional, which may include
a free component or limited usage
Application suites
•
Microsoft: Copilot for Microsoft
365
•
Oracle: Fusion Cloud Applica-
tions
•Salesforce: Einstein
•
SAP: RISE with SAP S/4HANA
•Named user
•Named user + usage credits
•Employee-based
•Full user equivalent
Large language models
•Anthropic: Claude
•Google: Gemini Pro
•OpenAI: ChatGPT
•Tokens
•Characters
•
Provisioned throughput units
(PTUs)
Conversational
•Amazon: Lex
•Nuance Communications: Nina
•Environment
•Cost per session
•Application user
•Port-based
•
Transactional, which may include
a free component or limited usage
Coding assistants
•Amazon: CodeWhisperer
•GitHub: Copilot
•Individual user
•Professional user
•Enterprise (unlimited users)
Custom-built projects
•Accenture: myWizard
•IBM: Watson
•Tata Consultancy Services: ignio
•Wipro: HOLMES
•Resources per hour
•Project-based
•Monthly fee
•Business outcome
As can be seen from the table, some popular pricing methods seem to have emerged.
First, some version of a user-based or subscription pricing model, where companies
pay based on the number of seats/ users (Alexander and Higgins, 2024; Pineda et al.,
36
2024). This is a straightforward pricing model, but it exposes the company supplying
the product to potentially high costs as the usage of generative AI features can be
uncapped in this case. A second pricing strategy is related to usage or consumption,
such as token and transactional-based (Alexander and Higgins, 2024; Pineda et al.,
2024). This model ensures that the supplier does not accidentally lose money on high
usage. However, from a customer’s perspective, it is challenging to know precisely
how much the product is going to cost them.
Further, Pineda et al. (2024) and Lange, Nieuwenhoff, and Biljardt (2025) highlight
that outcome-based pricing model is on the rise, where customers are charged based
on the true value provided by the gen AI feature. Given the predicted power of gen
AI, this model can become very lucrative for the supplier if the company can quantify
the value delivered. However, quantifying value remains challenging (Hinterhuber
and Snelgrove, 2022).
When implementing a gen AI pricing strategy, companies need to consider the impact
their products will have on the business and how it should be reflected in the pricing
model. For example, if a new gen AI product feature is expected to reduce the number
of people needed to execute the same work by the customer, using a seat-based
pricing model would not be wise. Figure 11 gives a representation of the pricing
models to use based on the product’s impact and quantifiable value delivered (Lange,
Nieuwenhoff, and Biljardt, 2025). User-based metrics often use per-user/ seat-based
pricing and when moving upwards to outcome-based, pricing can be based on, for
example, customer support tickets resolved (Lange, Nieuwenhoff, and Biljardt, 2025).
Support tickets resolved are an easy metric to quantify whether the AI solves it or
not, and the product can replace human workers.
To further gain an understanding of the current pricing thinking, one can turn to
the large technology providers and examine their current pricing for gen AI features
that are incorporated into the core product. For example, ServiceNow has added
generative AI capabilities to its current product offering. Companies need to license
these capabilities, and research indicates that the gen AI features can be priced
upwards of 60% on top of the standard product price (Cook et al., 2024).
In addition, the gen AI capabilities consume "Assist Credits", with varying costs
depending on what the customer wants to use the gen AI for. Salesforce’s gen AI
product Einstein can be added to current enterprise plans and, similar to ServiceNow,
has a credit limit tied to using gen AI capabilities within the product (Martina,
Liversidge, and Decker, 2024). This contrasts with Microsoft, which charges a flat
fee per user for its AI copilot (Microsoft, n.d.).
37
Figure 11: Price Model Matrix (Lange, Nieuwenhoff, and Biljardt, 2025)
As seen in the example of ServiceNow and Salesforce, they are both trying to mitigate
the risk of skyrocketing costs of high customer usage that can occur with only seat-
based pricing when customer usage is not capped. In addition, they help their clients
better understand the cost of deploying gen AI features by charging a flat fee for
the initial product and charging more if the usage exceeds a predefined limit. This
approach benefits both the company and the user, promoting a fair product offering.
In contrast, Microsoft’s offering of a flat fee is even more transparent for the customer.
Still, it subjects Microsoft to potential losses if the cost of supplying all the features
surpasses the price.
No matter the pricing strategy used, it is still essential to understand the cost side
(Monroe, 2003) of supplying customers with gen AI product features. Figure 12 gives
a high-level understanding of how gen AI capabilities should be incorporated into
a company’s pricing strategy based on the delivery cost and customer benefits of
the feature. With products that are expensive to deliver, companies should supply
the gen AI features as add-ons to the current product or up-sell the features. With
the example above, Microsoft is using the add-on practice, whereas Salesforce and
ServiceNow gravitate more towards the upsell approach.
However, it is essential to remember that the strategy used should align with the
long-term product goal (Santoro and Hare, 2025). If the goal is to increase market
share, companies can focus on including the generative AI features in the current
offering to drive more demand for the product. On the other hand, if the company’s
38
goal is to drive increased customer revenue, the additional AI features should be sold
as an add-on or as an upsell.
Figure 12: Breadth of Value and Delivery Cost Influence Packaging Approaches
(Santoro and Hare, 2025)
Scholars have emphasised that when setting prices for innovative products, it is
crucial not to position them too low (Hinterhuber, 2004). This is also true with gen
AI, as Gartner proposes that cutting-edge gen AI products should be tested with a
premium price (Santoro and Hare, 2025). Thus, cost-based strategies are unlikely to
be adopted in the gen AI space. As value driven by AI can be substantial, providers
can try to capture this value via a value-based strategy while ensuring that they are
aware of competitors’ positioning. As the gen AI space moves so quickly, companies
must stay flexible and adapt to changing market dynamics. For example, Gartner
highlight that companies need to be prepared to adjust prices down for AI features
that customers start to expect as standard (Santoro and Hare, 2025).
39
3.3.9 Price Discrimination
Price discrimination was first introduced in the early 1920s. Price discrimination can
be broken down into three degrees of discrimination and it involves charging different
prices for the same product or service (Pigou, 2017). Ideal degree (first degree),
which charges the exact price the consumer is willing to pay for the product, the
demand price. Second degree refers to charging buyers different prices based on the
quantity consumed. Lastly, third degree of price discrimination involves dividing the
customers into various groups and charging a different price for each group. Pigou
(2017) highlight that price discrimination works best in settings where the customer’s
willingness to pay does not depend on what other customers are charged.
Price discrimination generally leads to higher revenue, as a single-price strategy may
undercharge for the customers that are willing to pay more (Nagle and Holden, 2018).
It can also be argued that all companies that have the capability to conduct price
discrimination are doing so (Varian, 1989). However, it is not trivial for companies to
know how many price segments should be offered. Nagle and Holden (2018) argues
that more segments are always better. This is because companies can then get closer
and closer to first-degree price discrimination and charge precisely the value that
each customer is getting out of the product or service.
Nevertheless, managing a large number of different price segments is generally
impossible. The complexity and cost of doing so is practically the limiting factor
(Nagle and Holden, 2018). In addition, identifying and getting acceptance from
customers with higher purchasing power is challenging.
Scholars have identified three conditions that should be met for price discrimination
to be possible (Varian, 1989). First, the company need to have market power to
carry it out. Second, the company needs to be able to sort customers into the correct
groups with similar willingness to buy and purchasing power. Lastly, the company’s
products must be designed to naturally prevent resale or exchange between third
parties that would otherwise undermine the value of price discrimination.
40
4 Research Methodology
The following section introduces the reader to the methodology used in this thesis.
It first discusses the research approach, followed by an overview of the research
setting and the data collection. Next, the data analysis is discussed, and lastly, the
development of the pricing framework is introduced.
4.1 Research Approach
The research employs a qualitative research approach, with data collection based
on semi-structured interviews. According to Gioia, Corley, and Hamilton (2013),
a qualitative research approach is suitable for this type of research, as it enables
the present thinking of the industry to inform the research, rather than basing it on
pre-existing ideas.
This research primarily uses the Gioia methodology. The Gioia methodology is
well-suited for new concept development with an inductive research setting and relies
on grounded theory (Gioia, Corley, and Hamilton, 2013). Further, Woodruff (1997)
noted that quantitative data is not enough when trying to comprehend the customer
value and, subsequently, the pricing. Consequently, supporting a qualitative research
approach.
The Gioia methodology relies on coding the interviews into first-order concepts
without trying to categorise the data (Gioia, Corley, and Hamilton, 2013). This is
done already after the first few interviews. As the research progresses, the concepts
are refined to identify similarities and differences within the categories. Lastly,
these are categorised into second-order themes that are used to answer the research
questions (Gioia, Corley, and Hamilton, 2013). After taking these steps, a visual
representation of the data can be created to illustrate the relationships at hand and
connect the responses to relevant theory.
The interviews are conducted as semi-structured interviews. In a semi-structured
interview format, the questions are prepared beforehand and discussed during the
interview, leaving room to explore areas that the interviewee brings up (Myers and
M. Newman, 2007). Keeping the interviews semi-structured allows for flexibility to
explore what the interviewee touches on, rather than strictly following a predetermined
script. Further, semi-structured interviews allow for modification as the interview
process progresses to ensure fit with the research questions and scope. Both the
supply and buy sides have their own interview plans, each with a distinct set of
interview questions.
41
4.2 Research Setting
A two-phase research design, combining semi-structured interviews with desktop
research, was employed to develop a comprehensive understanding of how firms
approach value capture and pricing strategies for generative AI.
First, 31 semi-structured interviews were conducted with industry professionals who
work at the intersection of generative AI and pricing. Participants were purposefully
selected to reflect a diverse range of professional backgrounds, thereby enhancing the
breadth and relevance of the thesis across both time and audience, particularly in
light of the rapidly evolving nature of generative AI.
In addition, the interviewees were selected to include individuals from similar in-
dustries, and roles were chosen to ensure that the views presented are not based
on a single person’s opinion but represent the broader understanding in the field.
The selected interviewees thus enabled triangulation of ideas, an essential aspect of
qualitative research (Myers and M. Newman, 2007).
All the interviewees represented companies that operate in a business-to-business
setting. This was a deliberate decision, as the pricing dynamics and complexity
involved in B2B pricing differ from those in a B2C setting. The companies mainly
operate in the technology or services sector, with a few operating in an industrial
setting.
The research focused on two types of people. One side was people who are currently
developing or have developed a generative AI solution. This segment is referred to
as the suppliers. These individuals typically came from startups and scaleups, as
well as a few professionals from more established consulting companies. A total of
13 interviews were conducted with people selling gen AI products and services.
On the other side are people who have procured gen AI products or services or are
looking to do so in the future. This group is referred to as buyers. The majority of the
people interviewed came from larger established companies, with a few representing
startups and scaleups. A total of 10 interviews were conducted among buyers. In
addition, 8 interviews were conducted, including discussions on both the supplying
and buying sides of gen AI.
The second part of the research is dedicated towards desk research. The desk research
aims first to comprehend the pricing of the foundational large language model
providers, as their pricing sets the price floor for the majority of application products.
Additionally, the desktop research gathered and analysed currently available pricing
42
data from startups and scale-ups to gain further insights into the pricing methods
presently used. Lastly, cost data from cloud computing is gathered and compared
against the cost of LLMs.
4.3 Data Collection
The data collection comprises two main parts. The first part consists of semi-
structured interviews with 31 individuals, totalling 19 hours of interviews. The
second part consists of desk research, including the foundational model providers’
API prices and the identification of over 300 gen AI startups and scaleups’ pricing
models.
4.3.1 Semi-structured Interviews
Semi-structured interviews were conducted with 31 individuals over a four-week
period, starting in mid-March. A total of 19 hours of interviews were conducted.
These 31 individuals represent 28 separate companies headquartered in four different
European countries, with a higher share originating from the Nordic countries of
Finland and Sweden. For the reader’s convenience, the full list of interviewees is
presented in Table 5. The table includes information on the company, industry,
interview role, area discussed, my relationship to the interviewee and the length of
the interviews. All the data is anonymised and classified into higher categories to
ensure complete anonymity. A more detailed explanation of the interviews follows
below.
The interviewees were primarily selected using a purposive sampling technique. Pur-
posive interview sampling involves selecting participants based on their characteristics,
who possess the required knowledge for the research (Etikan, Musa, Alkassim, et al.,
2016). Specifically, homogeneous sampling and expert sampling were used. This
ensured that enough depth could be obtained while simultaneously finding a broad set
of individuals who have directly worked with generative AI and pricing. In addition,
snowball sampling was used for a few interviews where an initial interviewee referred
a new person who would be of interest to the study (Parker, Scott, and Geddes,
2019; Goodman, 1961). Six of the 31 interviewees were recruited through snowball
sampling. All the referrals were examined beforehand to ensure that they could
provide relevant information for the thesis.
On the buying side, an emphasis was put on selecting people in procurement roles,
whether directly leading procurement, such as a procurement manager, or as a subset
of their other responsibilities, such as CEOs or COOs for smaller companies. On
43
Table 5: Descriptive data of the conducted interviews
Int. Role Area Company Industry
Relationship
Length
min
1 Manager Buyer Established A Physical Product First-time 40
2 Manager Buyer Established B Physical Product First-time 45
3 Manager Buyer Established C Physical Product First-time 40
4 Manager Buyer Established D Physical Product First-time 40
5 C-level Buyer Established E Services Acquaintance 50
6 C-level Buyer Established F Services Close Tie 40
7 Operations Buyer Established G Services First-time 35
8 C-level Buyer & Supplier Established H Services Acquaintance 25
9 Manager Buyer & Supplier Established I Services First-time 45
10 Manager Supplier Established J Services First-time 40
11 Manager Buyer & Supplier Established K Services First-time 50
12 Manager Supplier Established L Services First-time 40
13 Manager Buyer & Supplier Established M Tech Product First-time 35
14 C-level Buyer Established N Tech Product First-time 35
15 C-level Buyer Scaleup A Tech Product Acquaintance 20
16 Manager Supplier Scaleup B1 Tech Product First-time 35
17 C-level Buyer Scaleup B2 Tech Product First-time 30
18 Manager Supplier Scaleup C1 Tech Product First-time 40
19 Manager Buyer & Supplier Scaleup C2 Tech Product First-time 30
20 Manager Supplier Scaleup D Tech Product Acquaintance 35
21 Manager Supplier Scaleup E Tech Product Acquaintance 35
22 C-level Supplier Scaleup F Tech Product First-time 45
23 C-level Buyer & Supplier Startup A Services First-time 20
24 C-level Supplier Startup B Services First-time 30
25 Technical Buyer & Supplier Startup C Tech Product Close Tie 50
26 C-level Supplier Startup D1 Tech Product First-time 45
27 C-level Buyer & Supplier Startup D2 Tech Product First-time 30
28 C-level Supplier Startup E Tech Product Acquaintance 30
29 Technical Supplier Startup F Tech Product Acquaintance 30
30 C-level Supplier Startup G Tech Product Close Tie 30
31 C-level Supplier Startup H Tech Product First-time 45
the supplier side, the emphasis was on people directly involved in the pricing of the
product, including various C-level roles, pricing and strategy managers, as well as
technical people, depending on the company’s structure.
The interviews were conducted either in English or Swedish. Twenty-five interviews
were in English, and the remaining six were in Swedish. The Swedish interviews
were translated into English by the writer and cross-checked against Deepl Translate
to ensure accuracy in the translation. Furthermore, two interviews were conducted
in person, while the remaining interviews were held online via Zoom or Google Meet.
All interviews were audio recorded using both a phone and a computer to ensure that
any technical issues would not result in data loss. The phone audio was recorded with
the pre-installed iPhone app Voice Memos, and the computer audio was recorded with
44
Microsoft Word Voice Recording. Audio recording was selected over video recording
to alleviate the uncomfortable feeling of being recorded, resulting in better interview
results (Myers and M. Newman, 2007). For the purpose of this thesis, an audio
recording is enough, and a video recording would not have added any insights to the
thesis.
The interview plan followed a structured yet informal format. The emphasis was on
ensuring that the interviewee was comfortable sharing their insights while contributing
to the research. This is particularly important for this thesis, as discussing pricing
tends to be a very sensitive topic. At the beginning of the conversation, permission
to record the audio for transcription purposes was obtained. All the audio recordings
were deleted after the transcription.
The interviews were split into two segments, depending on whether the buying or
supplying side was discussed. Each segment had its own set of interview questions.
Otherwise, they followed the same interview plan. For the interviews that discussed
both areas, a combination of interview questions was used, covering the most relevant
questions from each segment. The interview questions are available in Appendix
Interview Guide and Questions in addition to the interview guide.
The interview questions were created based on the literature review and the writer’s
understanding of the market. In addition, the interview questions were updated as
the interviews progressed to ensure that they addressed the research questions and to
gain further understanding of topics that the early interviewees had touched upon.
The goal of the discussion with suppliers was to understand how these companies
approach the value capture of generative AI and their pricing strategies and models.
The discussion also included the cost side of supplying gen AI products. For the
buying side, the aim is to comprehend how companies determine the value added
and, subsequently, the willingness to pay for products and services that incorporate
or will incorporate a generative AI component. These individuals tend to be more
open to discussing what they are buying and how they interpret the value and price,
as it is not as sensitive as discussing their own pricing strategy.
In the interview Table 5, the first column represent the interview classification
number. The table is in an arbitrary order, and thus the numbers do not correspond
to the interview order. The next column describes the interviewee’s role. It is
classified as a higher-order role to ensure full anonymity. The classifications used
are C-level,Manager, and Operational/ Technical.C-level roles include CEO, CFO,
COO, and CTO. Manager roles encompass various vice president positions, including
45
procurement, strategy, pricing, and growth management. Lastly, Operational and
Technical roles include data analysts and product engineering roles. The different
roles provide a broader perspective on how value and pricing are perceived at various
levels within an organisation.
The area discussed is displayed in column three. This refers to either discussing the
Supplying side of gen AI products and services or the Buying side. With some people,
both sides were discussed. A total of 10 interviews focused on the buying of gen
AI products and services, 13 interviews discussed the selling side, and 8 interviews
combined both aspects.
Next, the companies represented by the participants can be classified into three
distinct categories: Startup,Scaleup, and Established. The Startup classification is
used for smaller new companies that are still emerging and developing their product.
The focus lies on companies developing a gen AI product. The term Scaleup is
used to describe more established startups that have achieved product-market fit,
have stable revenues, and have raised a Series B or equivalent level of financing.
In addition, the scaleups should have deployed or are actively looking to deploy
generative AI into their current solutions. Established companies are incumbents that
have been in operation for a long time and are generally much larger than scaleups.
The established sector primarily comprises publicly listed companies or companies
of a similar size, as well as tech consultancies. Additionally, a few private equity
(PE) companies were interviewed, as they possess a broad knowledge of emerging
trends. The PE companies are also classified under the Established category to ensure
anonymity.
The industries of the participating companies are categorised into three primary
segments: Technology Products,Physical Products, and Services, based on the com-
pany’s primary revenue-generating offerings. The Technology Products segment
comprises firms whose core product is software or other digital technologies that
constitute the primary source of revenue. Services segment encompasses firms that
generate income through the provision of human labour, including but not limited
to consulting, accounting, and creative services. The Physical Products segment
comprises companies that primarily generate revenue from the sale of tangible goods
and equipment. The majority of the interviews came from technology products and
services, 17 and 10. While 4 companies could be classified as providing physical
products. The industries are directly related to the company classification, and no
specific focus was placed on a particular vertical to ensure that sufficient interviews
could be conducted.
46
In addition, my relationship with the interviewee is disclosed, which follows what
Myers and M. Newman (2007) suggests, arguing that the relationship can influence
the depth and level of the discussion. The relationship is classified as First-time,
Acquaintance, and Close Tie.First-time interactions are with people whom I had not
met before in any role. Acquaintance is classified based on having met the person
before, but with limited interaction. A Close Tie is a person with whom I regularly
discuss either through work or personally.
Lastly, the length of each interview is displayed in minutes in the last column. The
time has been rounded up or down to the nearest five minutes for clarity. The
interview length is based on the audio transcription length. The scheduled time for
most interviews was 40 minutes, with a few scheduled to be a bit shorter due to time
constraints. The interviews lasted between 20 and 50 minutes, with the average time
just under 40 minutes.
4.3.2 Desk Research
In addition to the interviews, desk research is conducted to gain further insight
into the value and pricing practices of gen AI. The aim is to find current insights
into the pricing of gen AI software. The data collection is categorised into two
groups, with the first category comprising foundational model providers. Given that
many companies rely on these providers for the underlying large language models
(LLMS) used in their products or services, understanding their pricing structures
is essential for establishing a price floor. Second, an analysis of the leading gen AI
product companies is conducted to gain a broader familiarity with the pricing models
currently in use.
First, the leading foundational model providers are identified. This comprises
Anthropic, Cohere, Google, Meta, Mistral, OpenAI, and X. The data was collected
on March 19th for all companies by visiting their respective websites. Each of the
companies’ flagship and lightweight models is selected, and the current pricing of
the model is identified. The price data is standardised to US dollars per one million
tokens and further divided into input and output costs. Only models with API
access and API pricing are used in the analysis. Furthermore, additional fees, such
as fine-tuning, caching, or grounding, are excluded from the study.
To contextualise the LLM cost data, data on current cloud computing is also collected
to facilitate a comparison of the cost differences between the prices of gen AI
foundational models and those of normal cloud computing. This data was gathered
on April 28th.
47
Second, the Y Combinator Startup Directory Database is used to identify the top
B2B generative AI companies (Y Combinator, n.d.). This is the same dataset as
used previously in Section 2.2. The search terms Generative AI and B2B are used
to identify the companies relevant for this research. This results in 324 companies,
with their industry of operations displayed in Figure 3.
4.4 Data Analysis
4.4.1 Semi-structured Interviews
Both audio recordings from the interviews were auto-transcribed using Aalto Univer-
sity’s Speech2Text software and cross-checked against each other to minimise the risk
of any information loss. The transcript is uploaded to ATLAS.ti, which is used to
analyse the interview data. The data processing follows the Gioia method with a
few modifications.
The data processing followed a three-stage process. An initial coding was conducted
after the first 10 interviews to ensure that the scope and questions were relevant. This
included creating a preliminary first-order Gioia codes and the emerging second-order
themes. This allowed for a more straightforward overview of the aggregated findings
early on. It also allowed for minor modifications to the interview questions. A new
coding for both the first and second order was conducted after 20 interviews to ensure
alignment and structure for the findings. Already at this stage, the data started to
converge on similar themes. Lastly, the remaining 11 interviews were coded on an
ongoing basis after the interview.
The 20 initial interviews were uploaded to ATLAS.ti, and all the relevant quotes
were identified and grouped into preliminary themes manually. This resulted in
almost 400 direct quotes and 18 initial themes. The next step involved exporting
the data from ATLAST into Excel, where the next step of the data analysis was
conducted. This involved merging similar quotes to reduce the complexity of the
data and reorganising the themes and underlying quotes to better fit the research
questions at hand.
The remaining interviews were also uploaded to ATLAS, quoted, and thematically
managed in the software before being merged into the original 20 interviews. Careful
consideration was taken to ensure that any new emerging themes were not overlooked
when incorporating them into the original data structure. In the end, no new themes
emerged, and the last interviews followed ideas similar to those of the original ones.
48
Lastly, the final Gioia data structure was created to encompass all the findings
from the interviews. The corresponding data structure resulted in nine second-order
themes and three aggregated dimensions, which are displayed in Table 13 and Table
14.
Table 13: Gioia data structure (1/2), showing the first-order codes, second-order
themes, and the aggregated dimensions
49
Table 14: Gioia data structure (2/2), showing the first-order codes, second-order
themes, and the aggregated dimensions
Throughout the process, detailed information was obtained for each quote, including
the interview reference number, area discussed, and additional comments related
to the quotes at hand. Thus, each quote can be traced back to its original text
and context, making it easy to go back and understand further details. Lastly, the
50
illustrative quotes added in the Results section have been slightly modified to ensure
grammatical correctness. The original words and meaning have been preserved for
each quote.
4.4.2 Desk Research
For each YC company that fit the search criteria, a manual review of the website
was conducted to determine its pricing model. Ultimately, pricing information was
found for just over 40% of the companies. For the remaining firms, pricing details
were either not disclosed or unavailable because the companies were too early-stage
to have a monetisation strategy in place.
Identified pricing models were first classified based on the information provided on
the company’s website and then grouped into higher-order categories aligned with
those discussed in the literature review. These pricing categories include credit-based,
feature-based,flat-fee,open-source,seat-based, and usage-based pricing models. The
classification was done manually and based on the writer’s view of the pricing model
and how it aligned with the literature. A credit-based pricing model is characterised
by a predetermined amount of credit or tokens that can be used within a specified
period. In addition, hybrid pricing models are highlighted, such as usage-based
layered on top of feature-based pricing.
4.5 Pricing Framework Development
An outcome of this thesis is a pricing framework for managers to use to help guide
their initial generative AI solutions’ pricing. The pricing framework is a 2x2 matrix
that builds on current pricing literature, incorporating interview data and generative
AI-specific characteristics.
Quadrant one, value-based pricing, is seen as the gold standard in pricing literature
and is also mentioned by most interviews as the pricing model they prefer. It ties
to maximum value creation as identified by Monroe (2003) and Baker, Marn, and
Zawada (2010). Further value-based pricing is the most optimal pricing model under
all market conditions as determined by Ingenbleek, Frambach, and Verhallen (2013).
Consequently, value-based pricing models are placed in quadrant one to reflect these
preferred characteristics of the model.
Further, traditional SaaS pricing models such as feature, seat, or flat-fee pricing
models are still commonly used by companies supplying generative AI solutions,
as indicated by the YC data, see Table 6. In addition, many suppliers and buyers
51
highlighted SaaS-based pricing as a pricing model worth considering, particularly
given its well-known understanding in the market and its predictability for finance
teams. Consequently, it was essential to include the current SaaS based pricing
models and investigate under which conditions the models function for a generative
AI offering.
In addition, credit-based pricing models are a pricing model that has gained traction
among generative AI offerings, as highlighted by Biermann and Petersen (2024a) and
implemented by some of the larger corporations. In addition, the YC data in Table 6
indicated that credit-based pricing for gen AI offerings has become a popular model.
In addition, credit-based models tie directly into the price floor as it is based on the
inference cost of the underlying LLMs. Consequently, credit-based pricing can be
seen as a cost-based pricing model, which, according to Ingenbleek, Frambach, and
Verhallen (2013), can contribute to higher new product performance, particularly in
competitive industries.
Lastly, a usage-based pricing model is the most adopted non-traditional SaaS pricing
model as indicated by the YC data in Table 6. Further, usage-based pricing was
also indicated by some interviewees. Similar to credit-based, usage pricing can tie
into the price floor and consequently better align with the specific inference cost
characteristics of LLMs.
A double-dimensional axis guards the four quadrants. The x-axis corresponds to
value, and the y-axis to the underlying cost structure. Thus, the axis dimensions tie
together the value aspect of pricing, a central theme in both the literature review
and the corresponding interviews. Second, the cost structure guards the value and is
adapted according to the specific characteristics of LLMs. In addition, both axes are
hierarchy prioritised. The hierarchy is conveyed through spatial proximity to the
matrix centre. Dimensions positioned closer to the core are assigned higher priority
than those placed further out, ensuring that tiebreakers can be determined in edge
cases. The dimensions are also continuous, enabling companies to fine-tune and even
charge discriminating prices between different customer segments.
On the x-axis, value measurability has the priority over expected value delivered.
Both dimensions were frequently discussed in the literature, with value measurability
being a key priority and concern among both scholars and interviewees, particularly
in understanding how to measure, quantify, and segment out value. Second, the
expected value delivered is the move from traditional pricing models to the usage
outcome. Value measurability has priority over expedited value because value-based
pricing models cannot be used without being able to measure and quantify it.
52
Lastly, the y-axis corresponds to the cost of supplying a solution, tailored to the
specific circumstances of generative AI. With inference cost and inference variability
being the two dimensions. The higher priority dimension in inference cost refers to
the underlying cost of running the large-language models in practice in the particular
solution the company is providing. Simply put, you can not set a price for anything
without knowing what the cost will be to supply it in the first place. Consequently, this
draws on the cost-based pricing strategy as the base by ensuring that the underlying
cost structure is taken care of and understood before proceeding to capture the
upside value delivered. Inference variability accounts for the cost differences between
different customer use cases, and generative AI has a higher compute cost variability
than has been seen in traditional cloud computing. Interviews did not converge on a
single view of the inference side of LLMs. Thus, it is essential to ensure that any
pricing framework accounts for this uncertainty before it converges.
As a result, each of the four matrix quadrants is based on traditional pricing literature
and adapted based on the empirical analysis and gathered interview data. The
dimensions are grounded in both pricing theory and interview data and adapted to
account for the different scenarios in gen AI solutions.
53
5 Findings
This chapter introduces the main findings from the interviews and desk research.
First, section 5.1 discussed the value that generative AI adds and how it should be
measured. Second, section 5.2 Generative AI as a Competitive Advantage discusses
the competitive advantage of generative AI and how it affects the perception of value
and price. Next, section 5.4 discusses the interviewees’ view on the willingness to
pay for gen AI products and services. Lastly, section 5.5 presents the findings on
the pricing of gen AI. The section is further divided into seven sub-sections, each
discussing the topic more deeply.
5.1 Generative AI Value
Understanding the incremental value that generative AI brings to products and
services is crucial for establishing effective pricing strategies. To accomplish this,
companies must first comprehend which business areas customers actually value.
In addition, companies need to develop the capabilities to measure the impact of
generative AI in the area effectively.
To apprehend value creation, it is essential to examine where suppliers perceive that
gen AI contributes incremental value. Simultaneously, it is important to investigate
where and how customers experience and assess the added value enabled by generative
AI.
The first three subsections answer the research question RQ1.1 - Where and how do
buyers and suppliers perceive incremental value from generative AI solutions, and
which attributes drive that perception?. The last subsection answers the research
question RQ1.2 - Which metrics and methods are companies using to measure the
value of generative AI?
5.1.1 Adding Value
Interviewees consistently emphasised time savings and automation as the primary
sources of gen AI’s value contribution. The efficiency improvement can materialise
in faster internal processes and deliverables. Interviewees highlighted completing
tasks in a fraction of the time it used to take. In addition, the suppliers see that
customers emphasise the automation capability of generative AI. The interviews
exemplify this. This emphasis on saving hours mirrors Blut et al. (2024), showing
convenience/efficiency and sacrifice-reduction are the strongest positive drivers of
customer perceived value, and echoes Grönroos (1990) argument that lowering
54
perceived sacrifice often raises value more than adding new features.
Time savings is a huge value add, you can do what used to take 30 hours
in one hour.
—Interviewee 9, Buyer & Supplier, Established I
The main kind of attribute [of gen AI value add] is the saved work hours.
But it’s not so much of a quantitative analysis of saved work hours. It’s
more of a qualitative view.
—Interviwee 7, Buyer, Established G
They [the customers] see the potential of gen AI in automating a lot of
what they do. So we position ourselves as a tool that allows that.
—Interviewee 21, Supplier, Scaleup E
I think it’s really the automation part. That’s in general what they [the
customers] see when they can do work more efficiently.
—Interviewee 18, Supplier, Scaleup C
Time saving, cost saving and ultimately scalability, ..., you can work on
more things at scale.
—Interviewee 29, Supplier, Startup F
Furthermore, in service-based industries, buyers expect efficiency improvements to
result in faster project delivery. They also expect the time and cost savings to be
reflected in lower prices. Hence, looking back at Blut et al. (2024), both buyers and
suppliers view gen AI more as a means to mitigate sacrifices, such as time, effort,
and cost, associated with other processes.
If the supplier is using gen AI we can expect that they have a higher
productivity and that things are getting done faster. ... we can expect
that the project would then be completed in half the time.
—Interviewee 2, Buyer, Established B
So we are utilising [gen AI] tools internally. Customers are expecting
the benefits to be passed on to them through faster deliveries or, basically,
lower prices.
—Interviewee 12, Supplier, Established L
Additionally, both suppliers and buyers are utilising publicly available large language
models (LLMs) to identify efficiency improvements for their own internal processes.
The interviewees highlighted that they are finding considerable efficiency gains. These
55
efficiency gains typically outweigh the monthly cost of using the tools, often by a
substantial margin.
Our three developer team right now codes at the speed of probably five or
six people. So we’re saving two monthly salaries there, which is still quite
a bit more than the 20€a month for ChatGPT.
—Interviewee 25, Buyer & Supplier, Startup C
If we think about the OpenAI licence cost. The time the tool saves me at
least makes up for the monthly cost.
—Interviewee 14, Buyer, Established N
Interviewees see that gen AI’s core value lies in automating routine work to achieve
substantial time savings and scalability. Across buyers and suppliers, tasks that
once took dozens of hours can now be completed in a fraction of the time, often
yielding productivity gains that far exceed LLM subscription costs. These efficiency
improvements are expected to accelerate project deliveries for service-based firms and
support more competitive pricing. In sum, generative AI is embraced as a strategic
tool to alleviate the time, effort, and cost burdens of traditional workflows.
5.1.2 Outcome Versus Process Value
To further comprehend where companies see the value in generative AI, interviewees
were asked to evaluate the value differences between a process and its output, and how
these differences change when generative AI is integrated into the mix, particularly
if the outcome is valued separately from the process that produces it.
The interviews revealed a clear consensus among both buyers and suppliers: it is
the outcome, not the process, that is ultimately valued. This view was repeatedly
emphasised across sectors and roles. Most respondents noted that customers are
primarily concerned with the quality, impact, or effectiveness of the delivered product
or service, rather than how it was produced or whether generative AI was involved
in the process. This sentiment was captured by several interviewees who stated:
Our clients know that humans have not made this, but gen AI. Still,
they’re willing to pay quite a lot for the service because it generates results.
So they don’t care how it’s done as long as the results are good.
—Interviewee 24, Supplier, Startup B
56
It is not the gen AI part that is interesting, it is the business value. What
problems does it solve? Does it lower my cost, lower my risk or increase
the revenue and effectiveness? That’s it. Then, if it was a monkey or AI
that did it, doesn’t matter.
—Interviewee 14, Buyer, Established N
I don’t think whether it was generated by gen AI actually matters in the
case of me making a purchasing decision on a piece of software.
—Interviewee 17, Buyer, Scaleup B2
It’s the output and the price that matter. How you reached the outcome
is secondary.
—Interviewee 3, Buyer, Established C
Other suppliers also noted the same and added that, in their view, the outcome is
the only thing that is evaluated as long as the underlying technology is safe and
compliant with data security protocols.
Customers don’t care what is under the hood, they care about the outcome
as long as it is secure, compliant etc.
—Interviewee 12, Supplier, Established L
Nonetheless, interviewees also acknowledge that, even if they emphasise the value of
the output more than the process itself when asked about it, the reality might be
more complex. They suggest that they may be unconsciously attributing value to
the process, even if they know that the outcome is the only valid metric of value. As
one participant stated about a services-based process:
I would wish in a rational mind to say that it doesn’t matter to me how
you got the solution. It’s just that I got the result that I wanted. In reality,
I know that the world doesn’t really work like that at the moment.
—Interviewee 7, Buyer, Established G
In particular, when the process is more visible and transparent, there tends to
be a stronger emphasis on the steps involved rather than just the final outcome.
People often assign value to the effort required to produce a result, with the time
invested being a key driver of perceived value. Both buyers and suppliers consistently
emphasised that the time it takes to generate an output greatly influences how
valuable it is perceived to be. Indicating that a faster process with the same result
can be valued lower when there is an understanding of how long a process should
take.
57
I had a law firm sell me a document that they spent half an hour filling
in and wanted to charge me several thousand euros. I was just very
disappointed. I know it is the norm in the industry, but time doesn’t
justify the cost.
—Interviewee 6, Buyer, Established F
If I sell a consultancy project for 10 days. I do it in 2 days with the help
of gen AI, with the same output quality. If the customer knew that I only
spent 2 days on it, there would be really negative reactions.
—Interviewee 31, Supplier, Startup H
A lot of people feel that the grind that you have to do to learn and make
something work is almost as important as the outcome.
—Interviewee 11, Buyer & Supplier, Established K
An additional viewpoint highlighted by one interviewee was that, at the moment,
the process value for generative AI is emphasised a lot more than the outcome. It is
currently driven by the hype surrounding gen AI, and as such, people tend to place a
higher value on the technology itself and having access to it, rather than objectively
evaluating the end results.
At the moment, the process is valued higher than the true outcome value.
But, it will shift to be based on outcome. ... Everyone wants to do gen
AI now.
—Interviewee 10, Supplier, Established J
Lastly, in longer-term settings, maintainability and transparency of AI-generated
outcomes become critical. One tech consultancy highlighted the fact that short-term
efficiency gains from gen AI can erode the long-term value add of the technology.
This is due to the "black-box" nature of the LLMs and, consequently, the loss of
transparency of the process. Having people who have been closely tied to the process
can add notable value down the line, as they possess a comprehensive understanding
of the system, making it cheaper and easier to maintain.
I do value the process itself, the maintainability, testability and updatability
are very much linked to the fact that I can see and understand the process
that provides the output and AI components are notoriously black boxes.
—Interviewee 11, Buyer & Supplier, Established K
Overall, interviewees agreed that customers primarily value the end result of a service
or product, regardless of the use of generative AI in its development. Yet this apparent
outcome-focused approach coexists with nuanced considerations of process visibility,
58
effort and time invested, and the current hype around gen AI access. In practice,
stakeholders may unconsciously attribute worth to how quickly or transparently
a result is produced, and short-term gains risk being offset by long-term concerns
over maintainability and “black-box” opacity. Thus, while the outcome remains the
primary value metric, perceptions of process and its implications continue to shape
the true worth of generative AI solutions.
5.1.3 The Value of Human-in-the-Loop
A related theme emerging from the interviews was the role of a human-in-the-loop
technique for gen AI solutions. The discussions revolved around when and where
a gen AI system adds more value than having a human, and in which situations
involving a human increases the value of both the process and the end result.
A consensus emerged among many of the interviewees, emphasising that human
oversight remains critical in processes involving complex judgment or business-critical
decisions. Rather than replacing humans, gen AI is often used to augment their
capabilities. Companies building more advanced agentic AI emphasised that they
are not aiming to replace humans but instead work alongside humans to empower
the work that humans are already doing. As one participant framed it:
Is it either Bob [an arbitrary person] or gen AI, or is it Bob using AI
intelligently ... which means that it’s actually better than the sum of the
parts?
—Interviewee 21, Supplier, Scaleup E
This is a very, thin red line here, we are actually not aiming to replace a
human.
—Interviewee 25, Buyer & Supplier, Startup
However, some buyers note that when evaluating gen AI systems, one of the metrics
they consider is the potential cost savings from replacing a human with a more
automated process. Even in the absence of supplier-driven automation agendas,
some buyers frame the adoption of gen AI as a means of reducing labour costs and
headcount.
You have to put the value of the gen AI into relation to what a person’s
hour costs doing the same work. How much can we save?
—Interviewee 3, Buyer, Established C
59
We already see that certain business functions are being replaced by gen
AI, for example, in call centres.
—Interviewee 4, Buyer, Established D
Regardless, the consensus remains that a human-in-the-loop system is, at least for
now, preferred over a fully automated gen AI system. This is true particularly in
processes that are more complex or in situations that require decision-making. The
consensus is that gen AI systems are not yet capable of making decisions, and humans
are still needed to verify and sign off on the outcomes generated by a generative AI
system.
Using humans [together with gen AI] is still a very valid option, particularly
in business-critical areas.
—Interviewee 25, Buyer & Supplier, Startup C
Somebody needs to sign off on each decision and put their names under
it. I wouldn’t trust any AI system yet.
—Interviewee 26, Supplier, Startup D
In situations where human decision making adds value, we are prepared
to pay a bit more by having a human involved in the process.
—Interviewee 5, Buyer, Established E
Nevertheless, in specific operational contexts, having a human can be value-destroying.
This is particularly true for repetitive tasks that can be automated with a deterministic
AI system, primarily non-gen AI solutions. In these situations, having a human who
makes more errors is seen as a value destroyer instead of a value enhancer.
In some situations, removing the human can increase the value by de-
risking the process ... Humans are going to make more mistakes than the
AI and cost a lot more.
—Interviewee 5, Buyer, Established E
By removing the human components, we also reduced the errors. Our
customers loved it, and it also scales very well.
—Interviewee 29, Supplier, Startup F
However, having a human-in-the-loop can be challenging when it comes to measuring
generative AI value. A frequently raised issue was the difficulty in directly comparing
gen AI outputs with human-generated outputs. Many participants argued that such
comparisons are inappropriate due to the scale and scope advantages of gen AI
systems. Generative AI often enables activities that would be impossible to perform
60
manually or with traditional software.
You can not compare a human to an AI. When we take an AI into use,
it will do millions of tasks, something that a human could not do.
—Interviewee 2, Buyer, Established B
You cannot compare gen AI to humans or other software products. Because
most of the things that I do with gen AI are not something that would be
able to be done by either.
—Interviewee 7, Buyer, Established G
On the other hand, some interviewees highlighted that they would compare a gen AI
system towards other software products, be it AI or not, instead of comparing them
directly to human-generated output. However, this view was also disputed, as one
interview put it:
Gen AI should not be compared to anything other than itself. ... So you
compare it against the business value it can generate, the output, how
accurate it is and how authentic it is.
—Interviewee 12, Supplier, Established L
Collectively, interviewees affirmed that human-in-the-loop oversight is indispensable
for tasks requiring complex judgment or accountability, positioning generative AI as
an augmentative partner rather than a substitute. Fully automated gen AI systems are
deemed most effective in automating high-volume tasks, thereby reducing errors and
scaling operations. Although a comparative cost-benefit analysis of AI deployment
versus human labour costs is often undertaken, direct benchmarking against human
performance is challenged by AI’s unique capabilities and scope. Accordingly, the
evaluation of generative AI should be based on its own merits—business value, output
quality, and reliability—with human oversight retained where nuance, compliance,
and governance demand it, and full automation applied only to predictable workflows
to maximise both efficiency and trust.
5.1.4 Measuring Value
However, quantifying the value of gen AI remains a challenge. Both sides acknowl-
edged that current practices for measuring value are underdeveloped. This lack
of maturity—whether organisational or industry-wide—makes it difficult to link
generative AI capabilities to clear economic metrics. Tying the gen AI value to the
customer perceived value formulas introduced by Grönroos (1990) is thus challenging,
as measuring the added value remains difficult.
61
Nevertheless, many interviewees expressed optimism and shared plans for developing
KPI frameworks tied to known metrics such as time saved, cost reduction, or review
scores. However, the challenge of separating out the gen AI part and the overall
lack of maturity, either within the company or within the industry, is preventing
companies from quantifying the value of gen AI.
We have business KPIs for gen AI value, but they are still a bit hard to
measure.
—Interviewee 9, Buyer & Supplier, Established I
Our industry is not yet mature enough to quantify the extra value that
gen AI brings, and it won’t be for some time.
—Interviewee 16, Supplier, Scaleup B
It’s maybe not possible to separate the impact of just the gen AI part.
—Interviewee 18, Supplier, Scaleup C
We have not yet quantified the value gen AI adds, but we will add KPIs
once we reach that stage. We aim to quantify the value through e.g., time
saved.
—Interviewee 26, Supplier, Startup D
In addition, sellers seem to be aware of the KPIs they will use once they start
measuring the value of gen AI more extensively. Companies are not looking to build
any magic new KPIs but are sticking with simple and known KPIs in the new world
of gen AI products.
In my mind, it should be the same metrics for gen AI [evaluation] as the
other alternative that’s there.
—Interviewee 15, Buyer, Scaleup A
Customers should be able to compare the impact of gen AI versus the
current state.
—Interviewee 11, Buyer & Supplier, Established K
We assess the value of our gen AI product to that of humans based on
review scores.
—Interviewee 25, Buyer & Supplier, Startup C
We are quantifying the value based on economic metrics. We know that
our customers run A/B testing on our product compared to competitors
and how it affects their North Star metric.
—Interviewee 28, Supplier, Startup E
62
Tools like ours that have a legacy in just being automation and time-saving
tools, that’s kind of the number one value driver that we still look at with
gen AI.
—Interviewee 16, Supplier, Scaleup B
We tie the value directly to the results that we add or the costs that we
save for you.
—Interviewee 20, Supplier, Scaleup D
It is promising to see that companies, at least to some extent, have started to think
about the KPIs they are going to use to show the value that gen AI can bring to
customers. Buyers are increasingly expecting suppliers to quantify and communicate
the additional value created by gen AI, particularly when price premiums are involved.
Sellers should provide data metrics on the difference between gen AI and
non-AI solutions.
—Interviewee 6, Buyer, Established F
It is up to the service provider to quantify the value [of gen AI products]
in terms of process or quality improvement we would get.
—Interviewee 1, Buyer, Established A
I think whenever there’s an extra cost, people want to know what I’m
getting in exchange for this, in which case you need to be able to point
out the value.
—Interviewee 19, Buyer & Supplier, Scaleup C
Overall, interviewees agree that measuring generative AI’s value is still developing,
constrained by organisational and industry immaturity and the challenge of attributing
impact solely to AI. Nonetheless, there is optimism as both buyers and suppliers
prepare to deploy familiar economic KPIs — for example, time saved, cost reductions,
quality scores, and A/B test results—to capture AI’s incremental benefits. As
customers increasingly demand transparent evidence of AI-driven value, grounding
assessments in these established metrics offers a clear path toward more rigorous and
comparable evaluation.
5.2 Generative AI as a Competitive Advantage
The following section, combined with the subsection 5.3 discusses research question
RQ1.3. - Which factors shape buyers’ and suppliers’ perceptions of the long-term
value of generative-AI solutions, and how do those factors exert influence?. Currently,
63
generative AI is a hot topic that is gaining a lot of attention from companies and
individuals. Thus, it is interesting to explore whether companies can drive additional
value and charge higher prices if generative AI is integrated into their product
offerings.
In certain circumstances, possessing a generative AI offering is perceived as a prereq-
uisite, as it signals to buyers that the supplier is actively engaging with technological
developments and maintaining competitiveness within the industry.
You need to have basic gen AI features that the market has come to expect.
Otherwise, customers don’t see you as a viable option.
—Interviewee 16, Supplier, Scaleup B-1
If you don’t use gen AI you are no longer trustworthy. It has become a
must-have.
—Interviewee 2, Buyer, Established B
The prevailing consensus among the interviewees is that today, generative AI provides
a potential source of competitive advantage. However, many people also believe
that gen AI is going to lose its competitive advantage over time and that gen AI
is simply becoming the norm that everyone expects. Thus, in the long term, the
expectation is that generative AI will not be a competitive advantage. Rather, it is
anticipated that generative AI offerings will become a baseline expectation, serving
not as a differentiator but as a necessary condition for a company to be considered in
procurement processes. Returning to Nagle and Holden (2018) Product Value formula
in Equation 2, it can be foreseen that the reference value is expected to decline,
and the only added product value attribute is the differentiation value between the
different gen AI offerings.
Today AI is a competitive advantage, but it is not going to be for long
—Interviewee 20, Supplier, Scaleup D
At the moment, gen AI is giving a competitive advantage over our com-
petitors. But they are also working on gen AI tools, and we just need to
run faster to not get left behind.
—Interviewee 18, Supplier, Scaleup C-1
Long-term gen AI is just becoming a part of every product, the same way
as other software features today
—Interviewee 15, Buyer, Scaleup A
64
There will be a base layer of gen AI feature that you need to supply to
customers to stay competitive.
—Interviewee 22, Supplier, Scaleup F
Furthermore, many companies also acknowledged that gen AI is currently more of
a marketing term. Gen AI product features that have made it into production are
generally simple and remain in the exploratory phase. The true value-added features
of the gen AI seem to be still lacking. However, customers are still willing to pay for
access to gen AI features, even if they are not very advanced or truly value-adding.
Gen AI is more of a marketing term at the moment than a true value-
adding product feature
—Interviewee 17, Buyer, Scaleup B
A lot of what I have seen is that gen AI is added as a gimmick on top of
the current product.
—Interviewee 10, Supplier, Established J
Gen AI is still a new shiny thing, and it is easy to show that we have it
and it provides this much value
—Interviewee 20, Supplier, Scaleup D
Even if it’s something that you could have done with a simple filtering
function in Excel, if you do it with gen AI, it’s somehow cooler.
—Interviewee 16, Supplier, Scaleup B
Gen AI is a hot topic and nice branding that we look at during the
procurement value assessment
—Interviewee 1, Buyer, Established A
In summary, while generative AI currently serves as a notable differentiator, signalling
to buyers that a supplier is technologically adept and forward-looking, this advantage
is widely anticipated to diminish as AI capabilities become ubiquitous baseline
features. Interviewees note that many of today’s generative AI functions remain
rudimentary, often deployed more as marketing decorations than substantive value-
adds, yet still command buyer interest and willingness to pay. Looking ahead,
generative AI is expected to transition from a competitive lever to a mandatory
criterion for market participation, altering its role from a source of differentiation to
a fundamental expectation in procurement decisions.
65
5.3 Long-Term Pricing Pressure in Generative AI
Continuing to answer RQ1.3, this section discusses an area raised by many intervie-
wees, the long-term price pressure of generative AI solutions.
Given the current competitive advantage of generative AI, it could materialise in
price premiums when companies want to get their hand on gen AI products. However,
buyers expect that long-term general market dynamics and competition will keep
prices down, making it difficult for anyone to long-term capitalise on gen AI price
premiums.
Prices in the market are going to be automatically reflected if an efficient
AI comes out. Like someone is going to be challenging the market, and
then they will probably lower their prices.
—Interviewee 6, Buyer, Established F
The prices will adjust automatically because there will be so many people
[building gen AI]. In the end, economics will win, and the other guys will
compete against each other.
—Interviewee 15, Buyer, Scaleup A
At some point, someone is coming to the market that is cheaper and can
scale the product better. Basic market dynamics will also play here.
—Interviewee 7, Buyer, Established G
Suppliers share this view, expecting that market pressure will keep prices down,
particularly in competitive markets. Further, one supplier already mentioned that
they have started to see price pressure in their market, suggesting that any possibility
of premium prices has already passed.
Prices should be set based on the next best alternative. With gen AI there
will be a lot of supply, pushing down the next best alternative. This will
lead to lower prices. Not because the value is lower, but because the supply
is higher.
—Interviewee 10, Supplier, Established J
So if the competition is fierce, companies won’t be adding additional profit
margin, but will be decreasing their prices. If the competition won’t be
fierce there, I think they will use it as an additional profit lever.
—Interviewee 23, Buyer & Supplier, Startup A
66
We are already seeing more price pressure. At the same time, we are also
becoming more efficient internally. So it’s that way around, rather than
we have an efficiency gain, and we decide whether or not to pass that on
to our customers. It’s already starting to be baked into the expectation of
the customer.
—Interviewee 21, Supplier, Scaleup E
In addition, buyers expect that any efficiency gains a service provider can achieve
by utilising generative AI internally will materialise in lower prices as the internal
supplier costs have decreased. This, combined with market dynamics, is a clear
indication that charging premiums for gen AI will be very difficult without a clear
value add.
If the cost structure of the service provider is going down due to changes,
and we see that their margins are exploding. We have an opportunity to
negotiate.
—Interviewee 1, Buyer, Established A
If the cost of supplying the service I’m buying decreases. I would compare
the price to the market to see if I need to renegotiate.
—Interviewee 6, Buyer, Established F
It is very difficult for a company to be over-profitable over a long period
of time. Market dynamics kick in, and some of that profit will be shifted
to us. Otherwise, we will change suppliers.
—Interviewee 5, Buyer, Established E
On the other hand, one buyer acknowledged that while short-term profitability for
suppliers may be acceptable, it must balance out over the long term. This implies
that suppliers may currently demand a premium for generative AI features. However,
they must remain alert regarding competitive dynamics to avoid being underpriced
or replaced by rivals.
The interviewees anticipate that, despite any initial premiums for generative AI
solutions, competitive market forces and abundant supply will drive long-term price
erosion. Buyers expect efficiency gains realised by suppliers to be reflected in
lower fees, and are prepared to renegotiate or switch providers should margins grow
disproportionately. Consequently, sustaining a durable price premium for generative
AI will require a demonstrable and unique value proposition, since undifferentiated
offerings will inevitably have to surrender to standard economic dynamics.
67
5.4 Willingness to Pay for Generative AI Solutions
The following sections address research question 2 beginning with RQ2.1 -Which
factors raise or constrain customers’ willingness-to-pay for generative-AI solutions?
When it comes to willingness to pay for generative AI products, suppliers believe
that customers’ willingness to pay increases, particularly if the AI system is capable,
works well, and can innovatively add value.
Having an AI that works today, definitely increases the willingness to pay.
—Interviewee 16, Supplier, Scaleup B
AI can work as a sales enabler and provide justification for higher prices.
Not very high price hikes, but at least inflationary . . . More advanced gen
AI systems can drive higher willingness to pay.
—Interviewee 18, Supplier, Scaleup C
If we can save time and increase the human performance, the willingness
to pay increases.
—Interviewee 27, Buyer & Supplier, Startup D
Customers are willing to pay more for smart solutions, helping them do
more than previously.
—Interviewee 12, Supplier, Established L
Buyer willingness-to-pay for generative AI solutions reflects established purchasing
behaviour; organisations will invest more only when a solution measurably delivers
value. Accordingly, for generative AI solutions, the willingness-to-pay seems to align
with literature and what Homburg, Koschate, and Hoyer (2005) identifies in their
research. High WTP only occurs in the far right tail, where customer satisfaction
and expectation are the highest. As such, gen AI WTP is no different from the
current procurement practices of other solutions. If a product adds value, companies
are willing to pay for it.
If it brings value, we need to be ready to pay.
—Interviewee 1, Buyer, Established A
If [the gen AI platform] is actually good, then it wouldn’t be a problem to
pay like 10x or 20x the prices for them.
—Interviewee 7, Buyer, Established G
68
When we want to buy gen AI, we will buy gen AI, and the better the AI,
the more we are willing to pay for it.
—Interviewee 2, Buyer, Established B
If there is some smart gen AI in the platform that improves the work, we
might pay more for it. How much more, difficult to say.
—Interviewee 3, Buyer, Established C
So if you don’t use it and it doesn’t create value, it costs nothing. But if
you use it a lot and it creates a lot of value, then you are ready to pay a
lot for it.
—Interviewee 5, Buyer, Established E
However, it was also noted that suppliers struggle to demonstrate the additional
value that gen AI could bring to a customer. This would justify a higher willingness
to pay among customers.
The challenge suppliers are having is to show the value of the gen AI so
that we would be willing to pay more for it.
—Interviewee 1, Buyer, Established A
Both suppliers and buyers are somewhat more hesitant about setting premium prices
for products and services that use generative AI. As previously discussed, a company
must demonstrate value to justify premium prices.
I don’t see that gen AI would be a decision factor on what price the product
or the service that I’m buying should be.
—Interviewee 7, Buyer, Established G
Whether it has gen AI stamp on it or not doesn’t matter frankly. I think
that in the future, all products will have AI in them.
—Interviewee 15, Buyer, Scaleup A
It will be difficult to charge a premium price for gen AI features when
the market comes to expect that everyone has these features. If you don’t
have them, you are out of the game.
—Interviewee 16, Supplier, Scaleup B1
I’m expecting similar prices for an agency even if they use gen AI, at
least before I can see the actual upside and value of the AI.
—Interviewee 9, Buyer & Supplier, Established I
69
We don’t pay for something just to have access to it. It needs to be based
on value creation.
—Interviewee 5, Buyer, Established E
Finally, both suppliers and buyers emphasise that the innovative aspect of incorpo-
rating generative AI can, in itself, influence a higher willingness to pay. In some
cases, it may even be a decisive factor.
There is always value in being innovative. So when we utilise gen AI, I
actually have a feeling that our customers are willing to pay even more
as they see that it’s a sensible and smart step.
—Interviewee 24, Supplier, Startup B
Buyers are increasingly asking whether suppliers are incorporating gen
AI. They may request examples or roadmaps showing gen AI components,
not because they know exactly how to use AI, but to ensure the supplier
is evolving. ... Not having gen AI signals that a supplier might be falling
behind.
—Interviewee 13, Buyer & Supplier, Established M
Across cases, willingness to pay for generative-AI remains rooted in classic value-
based purchasing: buyers pay more only when the system demonstrably improves
efficiency or outcomes. While innovative applications of AI can momentarily com-
mand higher fees and signal competitive differentiation, vendors face mounting
pressure to demonstrate clear, measurable outcomes as the gen AI label becomes
ubiquitous. Consequently, enduring willingness to pay will depend not on the pres-
ence of generative AI alone, but on its distinct, value-adding contributions and the
supplier’s capacity to substantiate them. Even so, the generative AI innovation
signal still elevates perceived supplier credibility and can modestly lift WTP where
it aligns with strategic differentiation.
5.5 Pricing Generative AI
Continuing from the last section, this section addresses the last two research questions
RQ2.2 - How well do prevalent SaaS-derived pricing models capture generative-AI
value? and RQ2.3 -Under what conditions do alternative pricing architectures better
align with customer willingness-to-pay and supplier economics?.
When it comes to pricing generative AI products, a single preferred pricing model
has yet to emerge. This was confirmed by both the data analysis displayed in Table
6 and by the interviews. On the other hand, the interviews revealed a stronger
70
emphasis on moving towards value- and outcome-based pricing for gen AI products.
However, this shift has yet to be reflected in the data among the gen AI-classified Y
Combinator startups and scaleups.
The data in Table 6 clearly shows that traditional SaaS pricing models are still
frequently used in the gen AI landscape, with over one-third of the pricing models
being feature-, seat-, or flat-fee-based models. The data also reveals that the somewhat
newer pricing models, such as usage-based, are widely used. In addition, more gen AI-
specific pricing models with credit-based features are gaining momentum, particularly
among companies closely tied to the foundational LLMs.
Table 6: Distribution of Pricing Models used for generative AI products, based on
YC B2B startups and scaleups. *indicating traditionally SaaS pricing models.
Output No. of companies of total of pricing models
Feature Based*35 11% 26%
Usage Based 27 8% 20%
Credit Based 19 6% 14%
Feature + Credit Based 18 6% 13%
Open-Source 10 3% 7%
Feature + Usage Based 8 2% 6%
Seat Based*8 2% 6%
Flat Fee*5 2% 4%
Open-Source + Pricing 3 1% 2%
Flat Fee + Usage Based 2 1% 1%
No Information 189 58% -
Total 324 100% 100%
A combination of pricing models has also begun to emerge, including feature-based
models that utilise a credit or usage-based system. These models combine the well-
known pricing models of SaaS with a hedge for the supplier on usage, ensuring that
any usage exceeding expectations will be covered by the customer. This shifts the
unpredictability of product usage costs to the customers, rather than the supplier
bearing them and relying on the average usage to offset the cost unpredictability
among all customers.
5.5.1 Traditional SaaS Pricing
Some suppliers emphasised that a monthly subscription for generative AI would be a
preferred model. Particularly in the early stages of a product’s lifecycle, customer
71
usage and feedback are more important than having the correct monetisation strategy.
Thus, a known pricing model can encourage customers to return to the products,
providing a feedback loop for the company. In addition, a monthly subscription can
also help account for customer usage costs.
With a subscription model, it encourages the user to return to the service
because there’s a slight fear of missing out for what you’ve got
—Interviewee 8, Buyer & Supplier, Startup H
Monthly subscription is super easy to understand. ... it can also be priced
around your infrastructure costs.
—Interviewee 25, Buyer & Supplier, Startup C
Some buyers also emphasised that they would prefer paying according to a subscription
model, such as a flat fee or per-user, which makes it easy to comprehend and forecast
the total cost of the product. A flat fee model also ensures that customers can
take full advantage of the software product without worrying about unpredictable
expenses that could otherwise deter customers from using it.
Flat fee is good as you know how much it costs. . . . you can use the
service as much as you want. Otherwise, I think it might limit the usage.
If you have to think about it, okay, this search, for example, costs 25
cents.
—Interviewee 6, Buyer, Established F
If somebody could have come and told us that, hey, we’re going to do the
same thing for a fixed price, that would have been amazing.
—Interviewee 25, Buyer & Supplier Startup C
I would greatly prefer one single price ... either for a set number of
queries, or just upfront, and you [the supplier] figure out the economic
scale behind it.
—Interviewee 17, Buyer, Scaleup B
The pricing model should be based on the number of users that we have
on our end
—Interviewee 2, Buyer, Established B
Some suppliers also indicated a preference for flat fee-based pricing models. This
preference is generally driven by overall market conditions or based on the simplicity
that the pricing model provides to both the supplier and the customers.
72
With our product, we can easily show the value that we add. But we still
only charge a flat fee premium on top for the gen AI. ... Within our
market, everyone is moving towards flat fees, so it is very difficult to
capitalise on the upside of gen AI.
—Interviewee 20, Supplier, Scaleup D
We are going to use a yearly unlimited fixed-priced license.
—Interviewee 26, Supplier, Startup D
I want to have quite simple pricing models. So the flat fee structure is
simple enough. . . . it encourages experimentation with gen AI.
—Interviewee 24, Supplier, Startup B
Overall, some buyers and suppliers converge on traditional SaaS pricing models for
generative AI. They are driven by the practical way to price generative AI offerings,
especially in early product stages when usage insights matter more than precise
monetisation. Predictable monthly, per-user, or unlimited licences lower adoption
barriers, foster continual engagement, and simplify cost forecasting for customers
while allowing suppliers to align charges with infrastructure outlays.
5.5.2 LLM Cost Consideration
The cost of supplying a product or service sets the price floor, the minimum amount
that must be charged to ensure the price is not lower than the expenses. For generative
AI applications powered by underlying large language models, the computing cost to
train and update, as well as the inference cost variability between customers, can be
substantial. However, the challenge with traditional SaaS pricing models is that they
do not account for variable cost structures. Traditional SaaS pricing models were
developed for steady and predictable compute and customer costs. Consequently, to
better understand the pricing options available to companies, it is essential first to
identify the cost structure of the foundational models. Given that most products
and services are built on top of large language model providers, understanding the
pricing structure of these providers is crucial for comprehending how pricing models
should evolve in a gen AI setting.
The cost structure from different LLM providers also varies immensely. It is far more
widely distributed than typical cloud infrastructure costs from providers such as
Amazon Web Services (AWS), Google Cloud Platform (GCP), or Microsoft Azure.
Table 7 shows the price difference for comparable medium virtual machines among the
top three cloud providers. This example highlights the fact that the price difference
73
among the providers is less 1%.
Table 7: Hourly pricing for medium virtual machines from the top three cloud
providers. (Amazon Web Services, n.d. Google Cloud, n.d. Azure, n.d.)
Provider Instance Type Hourly Price (USD)
AWS m6i.2xlarge 0.384
GCP n2-standard-8 0.388
Azure D8s v3 0.384
Compare this number to LLMs cost variation for inference, which can be in the
tens of thousands of percentage points, depending on the provider and the model.
Table 8 provides an overview of the largest and most widely used foundational large
language model providers, along with their associated costs for the flagship model
and a lightweight version. Thus, the emphasis on cost structures and how they should
be factored into pricing is particularly more critical for generative AI products than
for traditional SaaS cloud offerings. Note, this is an example of the comparison for
cloud compute cost only and does not compare the compute capability of the virtual
machines to the compute requirements for gen AI applications.
Table 8: Pricing in US dollars of the top foundational large language model providers
for their lightweight and flagship models per 1 million input and output tokens.
Lightweight Flagship Price in USD/ 1M tokens
Provider Model Model Lightweight Model Flagship Model
Input Output Input Output
OpenAI 4o-mini o1 0.075 0.3 15 60
Anthropic Claude 3.5 Haiku Claude 3.7 Sonnet 0.8 4 3 15
Cohere Command R7B Command A 0.038 0.15 2.5 10
X - Grok - - 2 10
Mistral 3b-latest Large 24.11 0.04 0.04 2 6
Google Gemini 1.5 Flash-8B Gemini 2.0 Flash 0.038 0.075 0.1 0.4
Meta Llama 3.3 70B Llama 3.1 405B - - - -
Many suppliers also noted these implications. Highlighting that it is challenging to
estimate the cost variability of generative AI products accurately. This is driven by
the difference in customer product usage and how it translates to the underlying LLM
inference cost. Most interviews referred to inference cost, the cost related to each
query of the underlying LLMs, as simply the compute cost. However, in reality, there
74
is a difference between the inference and compute cost. Compute cost is generally
referred to as the underlying costs of training the models in the first place.
The underlying costs for our product are hard to estimate due to the
compute variance.
—Interviewee 31, Supplier, Startup H
We need to figure out how much a customer is going to cost in terms of
token cost.
—Interviewee 28, Supplier, Startup E
We are not able to estimate in any detail the AI cost per customer. At
a high level, we can use the number of queries expected to be run as a
proxy.
—Interviewee 16, Supplier, Scaleup B
Uneven usage leads to skewed cost distribution, with low power users
subsidising the high power users
—Interviewee 20, Supplier, Scaleup D
However, depending on the usage case, the cost considerations among suppliers are
minimal. Many suppliers highlighted that they are not concerned about inference
costs for their LLM product usage. Even today, many interviewees emphasised that
the cost of their gen AI use cases is already very small. Further, there is a belief
that, in the long term, the costs of LLMs will continue to decline and become more
commoditised. Driving down the overall price in the space.
In the short term, we need to understand the cost structure. In the long
term, it will become less relevant as the costs have historically been going
down.
—Interviewee 28, Supplier, Startup E
LLM costs are expected to continue to decline and become irrelevant.
—Interviewee 11, Buyer & Supplier, Established K
There is a cost associated with gen AI, but it is relatively small at the
moment. I haven’t thought about getting the investment back, as if you
can deploy it quickly, you recoup it basically immediately.
—Interviewee 9, Buyer & Supplier, Established I
The LLM compute cost we need to supply gen AI customer features is
going to be super small.
—Interviewee 30, Supplier, Startup G
75
So we’ve been using LLM and they’re not significant to us at all compared
to the cost that we actually pay for a human.
—Interviewee 29, Supplier, Startup F
In addition, one company highlighted that it is willing to bear the cost of supplying
generative AI features for now, believing that the long-term cost will continue to
decrease.
We are not factoring in any LLM costs because now the pattern has been
that it’s getting cheaper and cheaper. We are willing to bear the cost for
that. ... So, definitely out of our margin. Gen AI in any way is going to
increase our margin.
—Interviewee 12, Supplier, Established L
However, the cost consideration is heavily influenced by the product and usage
characteristics. Some suppliers took the opposite view, clearly highlighting that the
costs need to be considered and managed first before anything else.
We are forced to think about the cost first.
—Interviewee 16, Supplier, Scaleup B
There is a cost [associated with running gen AI] and the cost is not
marginal, every interaction basically costs. Pricing is changing due to AI,
and the SaaS market will need to adjust.
—Interviewee 23, Buyer & Supplier, Startup A
We are trying to maintain our margin levels, so that means we need to
consider the costs as well.
—Interviewee 18, Supplier, Scaleup C
There’s this huge cost associated with running AI. So I’m curious to see
how our gross margin will develop. But we are not actively looking at it,
even if we should be.
—Interviewee 21, Supplier, Scaleup E
Every customer interaction has a cost associated with it. But we decided
to roll it [gen AI feature] out to get some feedback.
—Interviewee 22, Supplier, Scaleup F
Given the cost uncertainty, companies still see opportunities to manage potential
cost rises. Two themes emerged: either switch to open-source models, which provide
better predictability and options for LLM costs, or move towards a pricing model
that factors in the cost unpredictability.
76
Suppliers emphasised the importance of having options to utilise pricing models
that more closely align with the underlying costs, or to establish usage limits for
the product. If a customer exceeds the usage limit, they can pay an additional
fee to continue accessing the service. This way, companies can hedge their costs,
particularly for high-power users who are expected to use the service more frequently
than the average user.
If costs spike, open source is a fallback for managing the cost.
—Interviewee 26, Supplier, Startup D
If the usage cost would surge, we would have to manage it and move to a
cost-plus model. ... We can have a contract with certain terms and certain
limits. And then, if the customer exceeds, we would have the opportunity
to invoice them for the overruns.
—Interviewee 21, Supplier, Scaleup E
We need to price for the next five years, and we do need to take the cost
into account, which is why most likely any AI features we implement will
need to have either strict usage limits or need to have a price against the
outcomes of your usage.
—Interviewee 16, Supplier, Scaleup B
Let’s say, everyone gets 1000 tokens for X amount and then they can
purchase additional tokens on top of that. ... Plans cannot be unlimited
for anyone.
—Interviewee 31, Supplier, Startup H
A Large part of our cost is the underlying LLM cost of processing data,
and what we’re doing or in our pricing, we just add a thin margin on top
of the cost.
—Interviewee 28, Supplier, Startup E
Some customers also noted that having a usage limit is fair, as it allows you to
pay more for additional usage when needed. Thus, a buyer is not overpaying for a
product or service that is not getting used, and if there is value created, they have
the option to buy additional access.
A fair pricing model is to have a flat fee that includes a specific usage
limit, and if you reach the limit, you can pay to get more.
—Interviewee 6, Buyer, Established F
77
LLM inference costs introduce substantial and unpredictable cost variance that
traditional SaaS pricing fails to capture, setting a volatile price floor for generative-
AI solutions. Suppliers report difficulty forecasting per-customer inference costs and
diverge on the strategic importance of these costs. Some regard them as negligible
and declining, while others treat them as margin-critical. To hedge the uncertainty,
some companies consider switching to open-source models, adopting cost-plus or
usage-capped models, and charging extra for high-consumption users. This approach
is one that buyers deem fair because it aligns fees with the actual value derived.
5.5.3 Moving Towards Value and Outcome-based Pricing
The majority of the interviewees emphasised that they are looking to move towards
an outcome or value-based pricing model for generative AI products and services.
Several viewpoints were highlighted on why a value or outcome-based pricing model
is the most appropriate for generative AI. For one, there is a widespread belief in
the market that generative AI will be a massive productivity boost and subsequently
create considerable long-term value. As a result, companies want to capitalise on the
upside, and the best way to capture the value increase is through an outcome- or
value-based pricing model.
If you’re subscribing to something, your tangibility is somewhat less than
if you’re charged by usage. Then, if you go to output and ROI [return
on investment], it’s even more tangible, which means that you can take a
higher fraction of the value from your customers.
—Interviewee 23, Buyer & Supplier, Startup A
Many suppliers highlighted that they are either transitioning towards value-based
pricing or outcome-based pricing. Outcome-based pricing models directly align
the price with the customer’s benefit. Particularly, if the expected benefit for the
customer is high, having an outcome-based pricing model can be very lucrative.
As a pricing professional, I like output. It is based on value, it is based
on ROI, and I will charge a fraction of the value that I deliver to you.
—Interviewee 23, Buyer & Supplier, Startup A
More and more, I see outcome-based pricing with AI.
—Interviewee 16, Supplier, Scaleup B
I could imagine that things move towards outcome-based or volume-based
pricing.
—Interviewee 15, Buyer, Scaleup A
78
We are trying to go to value-based pricing based on the customer production
increase. But it will be difficult and more work as a guide for our pricing.
—Interviewee 26, supplier, Startup D1
We are leaning towards value-based. But we would test it against our
expected costs, so we are also thinking cost-plus. ... However, it is difficult
to segment the AI value.
—Interviewee 18, Supplier, Scaleup C
Once we productise the Gen-AI offering better, I think there are opportu-
nities for much better models that are tied with value.
—Interviewee 22, Supplier, Scaleup F
However, some suppliers also highlighted concerns about a value-based pricing model.
Still, the expectation is that the market is going to transition towards value and
outcome-based, making the model easier to sell to customers.
Value-based pricing will be challenging to sell at first, but eventually it
will kickstart and there’ll be a transition towards it. Customers will be
adapted to it because, at the end of the day, they will generate more value
from this investment than from a normal subscription.
—Interviewee 12, Supplier, Established L
Some larger buyers also mentioned that they are prepared to pay according to
outcome and value. However, these models come with their own challenges for larger
companies, as discussed in Section 5.5.5.
If the outcome is very good, I am willing to pay more. If the results are
less good, I am willing to pay less. But I am absolutely convinced that it
should be output-based pricing.
—Interviewee 4, Supplier, Established D
We tend to like buying value-based. But at the same time, the supplier
can not get too much value out of it; it cannot be unrealistic.
—Interviewee 5, Buyer, Established E
Returning to literature, it also becomes clear why companies want to utilise outcome-
or value-based pricing. First, a price lies somewhere between the price ceiling
determined by the value and the price floor determined by the cost, as seen in
Figure 8. As discussed, the price floor is guarded by the underlying model costs.
Subsequently, the majority of companies cannot affect the price floor, and the only
way to capitalise is on the upside of the product. This can also be seen in Figure 10.
79
However, it remains unclear whether companies believe the value upside lies in the
reference or the differentiation element. Returning to the quotes about the value of
gen AI, it can be assumed that companies are adopting value-based pricing based on
the higher reference value. Driven by the gen AI hype cycle. However, given that
OpenAI and others have anchored the price for a very capable product remarkably
low, given the performance, the opposite view could also be argued. Still, some
suppliers noted the risk of low prices from publicly available language models, but
this has yet to pose a challenge to value-based pricing.
I feel like it’s annoying that ChatGPT and these companies have anchored
the price of AI very, very low. You’ve got these systems that are actually
very capable, but then they cost 20 bucks a month.
—Interviewee 25, Buyer & Supplier, Startup C
There is a risk that the low price of ChatGPT will affect the pricing
options. But at the moment, we are not experiencing it.
—Interviewee 26, Supplier, Startup D
I think that’s actually one of the dangers of the things that OpenAI has
done. People are immediately comparing, even if you had an LLM trained
on a particular set of data that creates a much better impact for you, you
might want it to be priced at the level [of OpenAI].
—Interviewee 17, Buyer, Scaleup B
Still, buyers also do not see the anchoring of ChatGPT and other providers as a
hurdle to paying much higher prices for vertical gen AI solutions.
I don’t think the price of OpenAI is going to affect the decision to buy
vertical gen AI solutions for tens of thousands [of euros], as long as it
provides real business value.
—Interviewee 14, Buyer, Established N
Even if interviewees emphasised a value- or outcome-based pricing model, it has yet
to be clearly adopted among companies. Among the YC companies in Table 6, not a
single company was directly classified as having an outcome or value-based pricing
model. Models such as usage- and credit-based pricing can be somewhat attributed
to outcome- and value-based pricing, as customer usage can be tied to the price.
However, this is mainly based on the consumption of tokens, meaning that the more
the product is used, the higher the price. On the other hand, this can be viewed
as a value-based model, as it is evident that the more a customer uses a product,
the more value it provides, assuming they continue to use it. However, it is still
80
not an absolute value-based model as it is not tied to any business KPIs or other
quantifiable value metrics.
Still, interviewees widely anticipate a shift toward outcome- or value-based pricing for
generative AI solutions. Aiming to capture the technology’s promised productivity
gains by tying fees to observable business results or ROI. Suppliers see this approach
as the principal means of monetising value above an essentially fixed cost floor,
though they acknowledge difficulties in quantifying outcomes and securing customer
acceptance at launch. Buyers say they are prepared to pay more when a measurable
impact justifies it, provided profit-sharing remains realistic, and regard token- or
usage-linked models only as partial proxies for genuine value pricing.
5.5.4 Pricing of Services
When it comes to the pricing of services that utilise gen AI as part of the service
offering, suppliers have started to see an opportunity to move away from hourly-
based pricing, which has had its scale disadvantages, and move towards a value-based
pricing strategy. Thus, for perhaps the first time, more time-based consultancies
see the opportunity to charge based on the value they provide, rather than just the
hours worked, plus a markup. This is particularly important for many consultancies,
as the speed advantage that Gen AI can add is massive. As a result, billable hours
would be reduced, or more customers would need to be acquired to maintain current
revenue levels. With outcome- or value-based pricing, consultancies do not have to
worry about the input hours; instead, they can focus only on the output.
In the services sector, we are seeing a shift away from hourly based towards
value-based pricing.
—Interviewee 9, Buyer & Supplier, Established I
The problem with time-based consultancy is that it doesn’t scale unless
you just hire more and more people. With gen AI, we can make those
hours count more and make more from one hour.
—Interviewee 11, Buyer & Supplier, Established K
We are now more willing to partner with our customers, reduce the price
and move over to base the price on value. You reap the benefits, we will
bear the cost, but we want a share of the upside.
—Interviewee 12, Supplier, Established L
Consultancies that embed gen-AI capabilities increasingly view value-based pricing
as a viable alternative to traditional hourly billing. This enables them to monetise
81
the substantial speed and productivity gains gen AI delivers without relying on
ever-expanding headcount. By linking fees to the measurable business outcomes, not
the labour inputs, firms can scale revenues more effectively, particularly given the
efficiency improvements of gen AI.
5.5.5 Challenges with Value and Outcome-based Pricing
Even if there appears to be a shift towards value- and outcome-based pricing models,
they do not come without their unique challenges. As noted previously, quantifying
and measuring the value impact remain challenging. In addition, when a human
is in the loop, attributing value to the gen AI versus the human is difficult. This
is particularly concerning for the early adoption of gen AI products that combine
humans and AI, and where the pricing model is based on value or outcome.
One challenge with value-based pricing will always be to quantify the real
impact on the customer side. Some of these factors may involve a mix of
human interaction and AI, making it difficult to attribute the contribution
of each.
—Interviewee 15, Buyer, Scaleup A
Additionally, suppliers face challenges with value- and outcome-based pricing. For
one, if the expected added value is low, suppliers are unlikely to take the risk of
supplying a service based on value. Further, the unpredictable cost for the consumer
is also a worry among some suppliers, particularly when dealing with established
and conservative companies.
Of course, some kind of result-based pricing could be one, but I really hate
that concept. I don’t want to take the risk on behalf of my client.
—Interviewee 24, Supplier, Startup B
In our market, it is a bit difficult to sell value-based solutions as the
customer is conservative and not used to value-based pricing.
—Interviewee 26, Supplier, Startup D
The main downside of usage-based pricing is the unpredictability for the
finance team. I think that is why companies have introduced the usage-
based pricing with a cap.
—Interviewee 28, Supplier, Startup E
CFOs are a bit afraid of value-based pricing.
—Interviewee 23, Buyer & Supplier, Startup A
82
Buyers also highlight these concerns. For one, CFOs do not feel comfortable when
they don’t know the full yearly cost for a product. In addition, if a company is
expected to become a power user, it knows it will be paying more for the product
or service than it would under another pricing model. Still, it offers the company
flexibility.
As a CFO, I need to ensure that everyone is following the limits and
impose restrictions to not go over the budget. ... Usage or value-based
pricing is super interesting. How on earth do I stay on top of my budget?
—Interviewee 21, Supplier, Scaleup E
As a CFO, I always feel a little bit funny about when I can’t project what
the cost will be. ... I require simplicity and transparency [on the pricing
model] so I can project what I will have to pay.
—Interviewee 15, Buyer, Scaleup A
Usage-based pricing is good to start with, as it enables you to try the
product cheaply. However, if you use the product a lot, it is not good as
you will be at the higher end of the usage spectrum.
—Interviewee 7, Buyer, Established G
Another challenging aspect of a value- and outcome-based model is aligning the
expected yearly cost with the budget. For one, companies tend to set budgets in
advance, and in many cases, managers want to use the entire budget to ensure that
they do not lose any of it in the following years. Both buyers and suppliers note this.
Pricing models such as value and outcome-based are challenging. We
have a budget we need to stick to, and we can not buy something that we
don’t know the end price of.
—Interviewee 2, Buyer, Established B
Our customers are conservative; they require predictability and have a
budget. We need to adapt our pricing model accordingly.
—Interviewee 26, Supplier, Startup D
There are people who want to use the entire budget within a year, not to
lose it the following year.
—Interviewee 15, Buyer, Scaleup A
We see that companies have a lump sum of money to use and they need
to use every penny of it, but not a dime more.
—Interviewee 25, Buyer & Supplier, Startup C
83
Although outcome- or value-based pricing models are conceptually appealing, their
wider adoption is constrained by the difficulty of disentangling the gen AI-specific
contribution when work is shared with humans, the reluctance of suppliers and
finance teams to accept revenue or cost volatility, and the tension between unknown
fees and strict annual budgets. These challenges are similar to those identified by
(Hinterhuber and Liozu, 2012), stressing the importance of understanding customer
value and finding data to quantify that value.
5.5.6 Indirect Cost Considerations
Apart from the product price, buyers also consider other costs associated with
procuring gen AI products. These are similar to other types of software products.
The indirect costs considered are mainly related to maintenance and competence
development. Still, these factors affect the perceived value of a product, as they
relate to a sacrifice, as discussed by Blut et al. (2024) and are visible in the customer-
perceived value formulas by Grönroos (1990). Even in the age of generative AI
products, buyers and suppliers should consider other cost factors associated with a
product’s procurement and operations.
We also need to consider other costs, such as upkeep and maintenance
costs. Most systems are quite similar; they all have some AI, but it doesn’t
affect our decision too much compared to the other factors.
—Interviewee 2, Buyer, Established B
The starting point [of gen AI investment] is how long is the journey, how
big is the change in the way we work, what is the return on investment,
and why do we want to do it.
—Interviewee 1, Buyer, Established A
You also have to consider quality, level of service, and sustainability. ...
also the willingness to innovate. What are the future plans? What does
the strategic innovation agenda look like? So that you work with the right
partners.
—Interviewee 4, Buyer, Established D
I’m used to using OpenAI for a lot. But if I’m going to change my habits
and behaviour, it takes a lot of time. So that is also a risk, a cost of
competence development.
—Interviewee 9, Buyer & Supplier, Established I
84
Beyond headline pricing, buyers evaluate the total cost of ownership, including
maintenance, integration effort, competence development, and the supplier’s long-
term innovation roadmap, when procuring generative AI solutions. These indirect
expenditures represent the sacrifice side of customer-perceived value, meaning that
even AI-enabled products must demonstrate favourable lifecycle economics and
strategic fit to justify adoption.
In addition, investment in generative AI is seen as normal research and development
(R&D) expenses to ensure that a company can stay competitive. The cost of
developing more advanced gen AI products can be substantial. However, customers
are unprepared to pay for part of the R&D expenses incurred in developing AI
products.
For us AI development costs are seen as normal R&D spend. ... We
are bearing the cost of gen AI development instead of pushing it onto
customers.
—Interviewee 18, Supplier, Scaleup C
We, as the company, are bearing the investment costs of gen AI. It is a
strategic decision from our side to invest in gen AI.
—Interviewee 12, Supplier, Established L
Investments into gen AI is seen as future proofing the product.
—Interviewee 16, Supplier, Scaleup B
The suppliers need to bear the development cost of gen AI. It can not be
pushed onto customers more than normal R&D spend.
—Interviewee 6, Buyer, Established F
On the other hand, one company mentioned that they partner with some of their
customers to co-fund generative AI investments. Thus, the supplier shares the
investment risk while keeping all the intellectual property.
We are partnering with some customers to co-fund or gen AI investments.
... our customers see us as a trusted innovation partner for AI.
—Interviewee 21, Supplier, Scaleup E
Suppliers uniformly regard gen AI development as part of their routine R&D spend
undertaken to maintain competitive relevance, absorbing these costs rather than
passing them to customers, who do not expect to subsidise such investment. While
isolated co-funding arrangements allow select clients to share risk in exchange for early
access, the prevailing norm is to treat gen-AI expenditure as an internal strategic
85
expenditure embedded within standard product-development budgets, not a separate
surcharge in pricing discussions.
86
6 Discussion
This discussion chapter is structured in two parts. First, it addresses the six sub-
research questions alongside the two overarching research questions, interpreting the
interview findings alongside existing literature. Subsequently, Section 6.2 introduces
the proposed pricing matrix framework, outlining its underlying drivers and explaining
how it emerges from and connects to the results.
6.1 Answering the Research Questions
Surprisingly, buyers and suppliers were largely aligned in both their assessment
of generative AI’s value and their views on how it should be priced. The main
differences arose from each side’s economic incentives. Suppliers aim to maximise
revenue, whereas buyers strive to contain costs. This was most clearly seen in the
willingness-to-pay and in buyers’ calls to place an upper limit on value-based pricing.
Even so, the two groups converged on most of the topics discussed in the preceding
chapter.
The interview findings can be linked to Figure 4 to address Research Question 1.1.
Respondents emphasised that generative AI adds value primarily by automating
tasks, increasing efficiency, and saving time. This reflects added value in the form of
convenience and efficiency. However, the value generated through the process itself
remains ambiguous. While some interviewees underscored that outcomes are the
primary driver of value, suggesting a high degree of industry maturity consistent
with Figure 5, others emphasised the importance of the process, pointing to value
drivers related to specific product attributes and performance.
The process can serve as both a value enhancer and a value detractor. For example,
generative AI may signal novelty and technological leadership, enhancing perceived
value. At the same time, in service-intensive industries, reduced human effort due
to automation may lead to perceptions of lower value, especially if pricing remains
unchanged.
Another dimension of RQ1.1 concerns the role of human oversight. Most intervie-
wees viewed human-in-the-loop systems as necessary, though their contribution to
additional value remains unclear. In general, human oversight does not appear to
reduce perceived value. However, in use cases involving highly predictable patterns
and where human error is a concern, fully automated systems may provide greater
value.
87
Turning to Research Question 1.2, quantifying the value added by generative AI
remains a challenge, a finding consistent with prior literature. Interviewees did not
offer concrete methods for measuring AI’s contribution. One issue is disentangling
the specific impact of generative AI from broader technological or human inputs.
Nevertheless, companies appear to use familiar performance indicators, such as time
saved or cost reductions, to estimate AI’s impact.
Regarding RQ1.3, interviewees expressed the belief that many generative AI features
will eventually become commoditised, making them essential for remaining competi-
tive. For now, however, generative AI still adds perceived value, primarily due to
its status and novelty, as illustrated in Figure 4. Even relatively simple AI features,
while marginal in absolute value, can currently enhance the overall offering.
Addressing Research Question 2.1, willingness to pay (WTP) for generative AI follows
traditional value-based logic. Firms are willing to pay when a solution contributes
to their perceived value dimensions. A key differentiator at present is the innovation
associated with generative AI, with interviewees noting that innovation remains a
central vendor selection criterion. This aligns with Figure 6, which suggests that
generative AI can contribute to customer delight. Currently, the perceived benefits
tend to outweigh the costs.
Figure 10 is therefore extended in Figure 15 to include generative AI’s contribution
to value. The dotted lines indicate the uncertainty of the actual value added of gen
AI going forward. Long term, the interviews revealed that gen AI will become a
commodity, meaning that long term, the current Gen AI added value will decrease.
The erosion results in lower total customer value, as the only value-adding factors are
the reference value plus the differentiation value. This ultimately leads to a decrease
in the delta between the price and value difference. On the other hand, it can be
that this is acceptable as the initial reference value is going to be high enough, which
will drive up the total value and, subsequently, keep prices stable.
However, it can also be that the current processes that Gen AI is trying to solve
have a very low reference value, meaning that any solution that can improve the
process is going to have a very high differentiation value. Thus, even if the generative
AI-specific value erodes, the overall value will remain very high, driven by the process
efficiency it drives.
With respect to professional services, Homburg, Koschate, and Hoyer (2005) found
that willingness to pay is highest and lowest at the extremes of customer satisfaction
and relatively flat in the middle. For generative AI, this could lead to two outcomes.
88
Figure 15: Value add of gen AI in value-based pricing
First, AI-driven improvements in speed and cost may increase customer satisfaction
and thereby WTP. Second, there is a risk of reduced satisfaction when efficiency
gains are not matched by price reductions, as reflected in interview feedback.
Concerning Research Question 2.2, both interview data and insights from YC com-
panies suggest that traditional SaaS pricing models, such as subscriptions or flat fees,
can be suitable for generative AI. These models benefit from customer familiarity and
simplicity. Furthermore, they are often anchored in predictable infrastructure costs.
However, challenges remain in aligning usage-based cost variability with pricing.
Yet, as Piercy, Cravens, and Lane (2010) cautions, pricing novel solutions like
generative AI in the same way as conventional software may erode perceived value.
This concern was echoed in interviews, where some noted that using standard SaaS
pricing might limit long-term value capture. Moreover, shifting away from this model
could prove tricky once customers become accustomed to it, potentially locking the
industry into suboptimal pricing practices.
Another aspect of traditional SaaS based pricing is indicated in Figure 11. These
pricing models tend to have low value contribution in addition to being an augmen-
tation layer for humans. Consequently, there can be situations where buyers have
the option to choose between several vendors. Being a vendor that offers traditional
SaaS-based pricing can be a negative aspect, as it indicates that the company does
not believe that its product adds enough value to use a value-based pricing model.
On the other hand, as suggested earlier, SaaS-based model has the advantage of
being easy to understand and predictable.
89
Lastly, regarding Research Question 2.3, interviews indicate an overall preference for
value-based and cost-based pricing for gen AI. Both strategies offer advantages and
disadvantages and can work well in different settings. On the other hand, there was
no indication of any more exotic or advanced pricing models being considered at this
stage.
The interviews also indicated a preference towards cost-based and value-based pricing
strategies. No indication was made that competition-based pricing would be a
consideration. These findings align with Ingenbleek, Frambach, and Verhallen (2013)
research displayed in Table 3, where competition-based pricing for new product
launches leads to subpar pricing and product performance. On the other hand,
cost-based and value-based strategies, as are considered by the interviews, lead to
better pricing and product performance in highly competitive industries regardless
of the underlying product cost being high or low. An essential factor is that there
is still some uncertainty and product-specific differences in the cost implications of
generative AI solutions.
Still, the challenge of quantifying gen AI makes moving towards a value and outcome-
based pricing model tricky. Some industries can have an advantage if product delivery
can be tied more easily to a normal KPI metric that the customer uses. For example,
this could be related to click-through rates in marketing and direct savings, such as
preventing waste material or replacing a person. However, as indicated by suppliers,
the aim is not yet to replace a human.
Moreover, many service firms are moving away from traditional hourly billing models
toward more value- or outcome-oriented pricing, in line with findings from Biermann
and Petersen (2024a). These new models aim to capture the efficiency gains that
generative AI enables. Without such a shift, firms may face revenue pressure,
especially as some industries have already begun to see declining revenues.
Despite these efforts, service-based firms still face the same fundamental challenge
as product companies: accurately measuring and pricing the value delivered by
generative AI. Nonetheless, existing experience with value-based pricing provides a
potential advantage.
6.2
Capturing Generative AI Value with New Pricing Frame-
work
To enable more effective pricing of generative AI solutions in B2B contexts, a pricing
matrix was developed based on key generative AI characteristics. This matrix aims
90
to optimise value capture from generative AI offerings and is presented in Figure 16.
Figure 16: Double-dimensional pricing matrix for pricing generative AI solutions in
a B2B setting.
The matrix is based on a double-dimensional matrix structure. On the x-axis, value
measurability and expected value deliver are assessed on a scale from difficult to easy
and low to high, respectively. Consequently, the x-axis measures the ability of a
supplier to calculate the value of gen AI solution and how much value is added. This
sets the initial understanding of how value should be captured and consequently
priced.
On the y-axis, inference cost and inference variability are measured on a similar
high-to-low scale. These measurements are gen AI-specific and account for the
inference cost of LLM-based solutions. In addition, the inference variability between
customers should be accounted for, which the y-axis also considers.
The purpose of the double-dimensional matrix is to support the prioritisation of the
included dimensions. Dimensional hierarchy is conveyed through spatial proximity
91
to the matrix centre. Dimensions positioned closer to the core are assigned higher
priority than those placed further out, ensuring that tiebreakers can be determined
in edge cases.
The pricing matrix extends classical value-based pricing theory in two ways. First, it
explicitly integrates stochastic cost floors, where earlier work assumes relatively stable
marginal costs. LLM-powered gen AI solutions are built on top of token pricing, tying
the cost directly to the usage, which introduces cost volatility that reposition the
cost–value corridor. Both in terms of unpredictable usage patterns between customer
profiles and the significant difference in cost between LLM providers. Second, it adds
a measurability axis that conceptualises the value delivered, showcasing the breadth
of customer value as indicated by Woodruff (1997) and others. By showing how
measurability moderates the translation of customer-perceived value into price, the
framework bridges customer-value literature and modern challenges of generative AI.
Value measurability has priority over expected value delivered, as a supplier can not
price based on value delivered if that value can not be measured and quantified.
Similarly, inference cost has priority over inference variability as the LLM costs
determine the floor price of any pricing strategy. Once the floor price is set, the
inference variability between customers should be considered.
For example, an edge case that creates a contradiction would be difficulty in measuring
the value and a high expected value delivered. This scenario would contradict the
framework. Yet, the dimensional hierarchy ensures the elimination of contradictions.
In this example, the supplier chooses the x-dimension based on the value measurability.
The framework results in a 2x2 matrix with four quadrants, each with the optimal
pricing model that suits the specific requirements measured for each of the four
dimensions discussed. The pricing matrix is based on the gathered interview data,
desk research and adapted to suit the specific characteristics, value drivers and
challenges in each pricing position.
Quadrant one in the upper right corner represents value-based pricing. Value-based
pricing is the ultimate goal for many companies supplying gen AI solutions. Value-
based pricing is suitable in situations where value measurability is easy and expected
value delivered is particularly high. This is supported by the fact that to use value-
based pricing, companies need to be able to quantify the value they are offering to
customers. In addition, it only makes sense to use value-based pricing if the expected
customer value is high. Otherwise, the supplier can not capitalise on the upside of
value-based pricing.
92
Value-based pricing should be used when the underlying inference cost and inference
variability are low. The drivers are based on the fact that value-based pricing is
not tied to any underlying cost structure. In most cases, as also seen in Figure 4,
the price ceiling determined by the value is higher than the price floor governed by
the underlying cost structure. However, there is no direct linkage between value
and costs. Consequently, to ensure success with a value-based pricing model, low
underlying cost and cost variability are desired.
As highlighted by some interviews, a traditional SaaS-based pricing model, such as
subscription-based or flat fee pricing, works for gen AI offerings, particularly for new
product launches. Quadrant two, the upper left corner, represents traditional SaaS
pricing.
SaaS based pricing works particularly well when value measurability is difficult and
expected value delivered is low. When it is uncertain how to assess the value provided
to customers or when the anticipated value is insufficient for value-based pricing, it
is advisable to adopt traditional SaaS pricing. Particularly for new product launches,
SaaS-based pricing can be a simple yet effective strategy for gaining initial learning
and stable revenues.
Suppliers should be careful not to set SaaS prices too low, especially when their
product includes cutting-edge gen AI capabilities that justify a premium. They must
also watch out for pricing models that is tied to seat-based pricing. If the software
delivers significant efficiency gains, customers may end up reducing headcount,
shrinking the number of seats and, in turn, the supplier’s revenue.
Similarly to value-based pricing, the traditional SaaS-based pricing model works well
where the inference cost and inference variability are low. Thus, if the underlying
cost structure of the gen AI solution resembles traditional cloud cost structures, SaaS
pricing can be very effective. SaaS pricing also enables suppliers to tie the price to
the underlying cost better. However, accounting for any variability is difficult with
very traditional SaaS pricing models.
Moving down on the y-axis to account for high inference cost and inference variability
leads to a credit-based pricing model. A credit-based pricing model involves customers
pre-purchasing credits that are then consumed at a pre-determined rate depending
on the feature and extent of the customer’s usage.
Consequently, if the underlying inference cost is high, the credit can either be priced
higher or customers consume it at a higher rate. In addition, when the credits can
be tied to different product features, it makes it easier to account for high inference
93
variability both between different product features and between different customer
groups. The model also enables buyers to purchase additional credit if they experience
high usage. As such, the power user of the product, who also has a higher cost
associated with them, is accounted for.
Credit-based pricing models are also preferred when value measurability is difficult
and expected value delivered is low. With credit-based, there is no need to measure
the expected value delivered. The model also ensures that the supplier can gain
revenue, even in situations when the customer is not using the product a lot, as the
credits are prepaid.
Lastly, quadrant four in the lower right corner represents usage-based pricing. This
is in a situation when the value measurability is easy and expected value delivered
is high. As usage-based pricing is tied to customer usage, it is the optimal pricing
model in situations where high customer value is expected, as it leads to higher usage
of the product. It is also a pricing model that requires some quantification of the
value delivered to set the base price that then scales with the usage.
Usage-based pricing works particularly well when inference cost and inference vari-
ability are high, as the model is tied to the underlying cost structure, such as LLM
tokens consumed. Similarly, as to credit-based pricing, usage-based pricing also
enables the supplier to have different consumption patterns for various product
features
Some buyers and suppliers highlighted the budget constraint requirements that a
product must adhere to. The pricing matrix can also account for budget constraints
among buyers. This adapted pricing framework is displayed in Figure 17 with the
grey areas indicating the pricing models available to budget-constrained buyers. The
pricing matrix only needs minor updates to account for budget-conscious customers.
In simple terms, the left-hand side accounted for any budget constraints, given
that both SaaS and credit-based pricing are based on known prices that can be
extrapolated to ensure budget adherence.
In addition, the adapted pricing framework incorporates mechanisms to cap the
supplier’s upside, addressing buyer concerns about disproportionate value capture.
Specifically, the cap ensures that suppliers do not retain an excessively large share
of the value delivered, thereby promoting a more balanced distribution of profits.
This feature enhances the framework’s applicability by reducing the risk of buyers
switching to alternative suppliers who accept a smaller share of the upside.
However, this limits the upside potential for high-value products. On the other hand,
94
Figure 17: Adapted pricing matrix to account for budget constraints
the right-hand side of the matrix can also be adapted to comply with a budget by
introducing a cap. The cap works as an upper barrier to set a limit on the maximum
upside of the product. Consequently, it enables the buyer to set the maximum cap
in the beginning, and if the cap is reached, access to the product can be limited or
the cap can be raised if the budget is increased.
Lastly, the measurement scale is discussed. The dimensions are intentionally kept
as ranging from low to high or, in the case of value measurability, from difficult
to easy, as there is currently no quantitative indication of the scale and when it
becomes optimal to move from one quadrant to the next. Given that pricing is seen
as inherently dynamic in nature, and considering the current fast-paced environment
of generative AI, it is difficult to tie the dimensions to a single numeric value. On
the other hand, some initial views on the scaling of dimensions can be formulated.
When it comes to value measurability, initial guidance can be provided. The scale
can be conceptualized as: can we as a company quantify the value we are delivering,
and can we measure the value in practice? These are two distinct questions that
95
should be evaluated independently. It can be relatively easy to quantify value in
a vacuum; however, estimating that value in a customer context is very different
and can pose unforeseen challenges. Consequently, decision-makers must be realistic
and transparent about their internal capabilities while allowing room for informed
judgment.
For expected value delivered, the guidance would be to assess the solution relative to
the current status quo. If the company’s solution is significantly better—referred
to in startup terminology as “10x better” than the existing market offering—the
value delivered will be high. In general terms, value delivery should be assessed in
comparison to current offerings. If it is believed that the generative AI solution will
not represent a substantial improvement, or if there are uncertainties surrounding its
impact, then pricing toward the lower-value side of the matrix would be advisable.
This also requires careful judgment, as in many cases it may not be possible to
directly compare a generative AI solution to the existing state, as was also noted in
the interviews.
Inference cost is also a judgment-based assessment, but it can be benchmarked
against a customer’s current cloud costs. If the costs are proportionally much higher,
it could indicate that pricing models characterized by high inference cost should be
considered. Otherwise, prices may become significantly higher than existing SaaS
offerings, which could potentially lead buyers to question the justification behind the
high list price.
For inference variability, one approach to assessment is to use the coefficient of
variation across the full spectrum of customer usage. Another method is to define
a baseline threshold that no more than a specified number of usage instances can
exceed. Additionally, modeling techniques can be used to simulate the effect on
margins by testing different variables across usage levels and determining which
category the solution aligns with.
As the company gathers more data, it becomes increasingly feasible to assess where
its solution falls along the four dimensions. In addition, external benchmarks can
be used to evaluate the solution and more accurately determine the current pricing
position. However, as with any important decision, several factors should be weighed
against one another. Even if the pricing matrix does not explicitly incorporate a
competitive perspective, monitoring competitors’ price levels remains an important
consideration for establishing long-term, justifiable pricing.
96
7 Conclusion
The final section of this thesis outlines the managerial implications of the findings
and the proposed pricing framework. It also highlights the study’s contributions
to academic literature and identifies areas for future research. Finally, the study’s
limitations are acknowledged, followed by a short concluding discussion tying back
to the research questions.
7.1 Managerial Implications
Generative AI solutions are reshaping B2B software economics faster than any prior
technology. Each year, investors channel tens of billions of euros into generative
AI startups, with hundreds of billions more going into infrastructure and in-house
projects aimed at harnessing the technology’s potential. However, these investments
have yet to translate into meaningful revenue and return on investment.
Given the investments, return expectations, and development costs, managers must
now convert the high interest and usage of gen AI into cash flow. The findings of this
thesis offer an initial guide to capturing generative-AI value and pricing strategies
managers can use given their specific scenarios.
The results indicate that gen AI solutions can be a competitive advantage today.
Further, the long-term expectations are that gen AI features will become a commodity
and an expectation of the market. Given this, managers need to invest in generative
AI capabilities today. For one, it ensures the company can stay competitive in the long
term. Second, investing today can help the company recapture some investment costs
while price premiums for gen AI solutions still exist. Over time, the commoditisation,
competition, and general market dynamics will exert downward pressure on prices.
Further, managers should focus on capturing value where customers perceive it the
most. Interview data show that buyers place the highest value on generative AI
features that can automate tasks, leading to increased efficiency and time savings.
Consequently, when developing new generative-AI features, managers should prioritise
automating process tasks and tailor those features to their customers’ workflow
priorities. Aligning product features with customer value, leading to gaining customer
satisfaction, ensures that the company can maximise the willingness to pay.
Lastly, managers need to apply the appropriate value capture strategy depending on
the specific circumstances in which the company operates to ensure that generative
AI features can become profitable. The pricing matrix introduced in Section 6.2
97
provides an entry point for managers to set the initial price strategy. It accounts for
the expected customer value against the distinctive cost structure of large-language-
model-powered products. It helps managers select the pricing strategy, whether
value-based, usage-based, credit-based, or traditional SaaS-based pricing strategies.
The pricing framework helps managers ensure that the bottom line is guarded while
guaranteeing that the maximum upside of the product can be achieved.
7.2 Contribution to Literature and Future Research
First, this thesis contributes to academic literature by providing an initial under-
standing of the requirements and characteristics regarding the value and pricing of
generative AI solutions by introducing the double-dimensional pricing framework for
generative AI. As pricing has been under-researched, given its financial importance
for companies, this research adds to the already extensive pricing research and extends
into the world of generative AI. Further, this research is among the first to explore
the intersection of value and price for B2B generative AI solutions.
Second, the findings provide early evidence that market actors anticipate a rapid
commodification of generative-AI solutions. In contrast, prior literature tends to
portray generative AI as a novel and durable source of competitive advantage, the
interviewees framed it as a forthcoming baseline expectation. This has the implication
that even if the generative AI today can convey a premium and a higher willingness-
to-pay, long-term, this is not going to be the case. Consequently, it also indicates
that investments in the field should be made today to recoup the investments quickly.
The study further demonstrates that generative AI technologies may be treated
analogously to earlier technological innovations, particularly with respect to R&D
investments and indirect-cost allocation. It also highlights that conventional value
and pricing theories still apply in many instances, but can fail when it comes to
specific cost considerations of LLM-powered generative AI applications. In addition,
similar challenges persist regarding value-based pricing, such as the difficulty of
measuring the value.
This thesis adds an important first view on how companies developing and buying
generative AI solutions are currently considering harnessing the technology’s value.
As such, it gives researchers an initial starting point to continue research as it evolves
and matures over the coming years.
Future research should focus on three aspects. First, quantifying the current findings
and the pricing framework would add further depth to understanding the true
98
implications of the proposed framework. To do this, an in-depth case study should
be conducted among several companies that fall into each of the four quadrants.
This would enable the researchers to quantify the revenue and bottom-line impact
difference for the different quadrants. Particularly, research should focus on finding
the optimal threshold values along each matrix axis to make it more tangible for
managers to decide when to move to a new quadrant in the pricing framework.
In addition, future research should study a particular industry vertical. This research
aimed to give a broader yet in-depth understanding of the current generative AI
value and pricing. Further research should add additional depth to a specific vertical
to make the findings even more targeted to different companies. However, this should
only be done once the industry becomes more mature to ensure that the findings
remain applicable for longer.
Lastly, future research can explore potential differences between pricing a new
generative AI solution versus pricing a generative AI product feature on top of a
traditional cloud-based offering. It can further explore the relation of human-in-
the-loop affect and find a way to ideally segment the difference between the human
contribution and the generative AI value contribution. Further, research could
explore the interlinkages between humans and the gen AI, particularly when AI
solutions move towards AI agents and replace human work to a greater extent.
This additional research would enable the development of an entirely new pricing
model based on a qualitative understanding of gen AI product pricing and quantified
data gathered from companies. This will lead to even higher value capture for
suppliers that also align with the specific characteristics of the buyers while taking
into consideration the market factors that are going to shape generative AI solutions’
pricing in the long term.
7.3 Limitations
A key limitation of this study is its broad scope. As the field of generative AI is
still in its early stages, a wide focus was necessary to ensure sufficient participant
recruitment, particularly given the sensitivity and strategic importance of pricing
discussions. While narrowing the scope to a single vertical could have yielded deeper
insights into that context, it would have limited the broader applicability of the
findings at a time when the field is rapidly evolving.
The choice to interview stakeholders from startups, scaleups, and established compa-
nies was deliberate. However, it added additional breadth to the study. On the other
99
hand, including perspectives from across these segments was essential to building a
comprehensive understanding of the emerging generative AI pricing landscape.
Another limitation relates to the maturity of gen AI adoption. Many organisations
have not yet implemented generative AI at scale or used multiple products for
comparison, which restricts the depth of experience among participants. Although
all interviewees had some exposure to generative AI tools, several noted that it was
still early days and that they had not previously reflected on many of the specific
questions posed in the interview.
Technical challenges were also encountered in a small number of interviews, including
occasional connectivity issues and the loss of one of the two audio recordings. As
this technical problem had been anticipated, each interview was recorded using both
a phone and a computer, ensuring redundancy. The Speech2Text application also
struggled with specific transcripts, requiring manual corrections. However, these
issues were minor and did not materially affect the quality or integrity of the findings.
Another limitation is the absence of quantitative data. While a qualitative, interview-
based approach was well-suited for the exploratory nature of the research question,
quantitative data could have added depth by measuring differences and validating
insights. Particularly quantifying the developed pricing framework would have helped
to answer how the framework would perform in a real-life setting under different
conditions and assumptions. However, pricing data is rarely shared due to its strategic
value, and collecting meaningful data would have required extensive access to multiple
companies, something not feasible within the scope of this study.
While qualitative interviews are a robust method for exploratory research, they come
with well-documented challenges (Myers and M. Newman, 2007). To mitigate these,
interviews followed a structured protocol: interviewees were briefed on the research,
provided with key questions in advance, and granted confidentiality. Audio recordings
(without video) were used with consent to enhance comfort and transparency. All
recordings were used solely for transcription and subsequently deleted. The identities
of participants remain confidential throughout the research and in the final report.
7.4 Conclusion
This thesis examined how B2B buyers and suppliers perceive the evolving value of
generative AI solutions (RQ1). It also identifies which technological and market
factors shape suppliers’ pricing decisions to align with buyers’ value perception
and willingness-to-pay (RQ2). The research does so through 31 semi-structured
100
interviews and Gioia-style qualitative analysis to develop a pricing strategy framework
for generative AI solutions. The result is a 2×2 pricing matrix that helps managers
choose the appropriate model under different conditions by mapping cost predictability
against value clarity. This pricing framework extends pricing literature into the
generative-AI era and provides an actionable starting point for managers to set robust
pricing strategies that ensure long-term revenue and profitability.
101
References
ADOBE (2023). Adobe Unveils Firefly, a Family of new Creative Generative AI.
Accessed: 13 February 2025. url:
https://news.adobe.com/news/news-
details / 2023 / adobe - unveils - firefly - a - family - of - new - creative -
generative-ai.
AKTER, Shahriar et al. (2023). “A framework for AI-powered service innovation
capability: Review and agenda for future research”. In: Technovation 125,
p. 102768.
ALEXANDER, Melanie and Matthew HIGGINS (2024). Negotiate AI and Generative
AI Pricing to Avoid Skyrocketing Costs. Tech. rep. Accessed: 24 February
2025. Gartner. url:
https://www.gartner.com/document/5440063?ref=
algobottomrec&refval=6147291.
ALTMAN, Sam (2025). [Twitter] 5 January. Accessed: 9 February 2025. url:
https://x.com/sama/status/1876104315296968813.
AMARAL, Juliana Ventura and Reinaldo GUERREIRO (2019). “Factors explaining
a cost-based pricing essence”. In: Journal of Business & Industrial Marketing
34.8, pp. 1850–1865.
AMAZON WEB SERVICES (n.d.). Amazon EC2 On-Demand Pricing. Accessed:
28 April 2025. url:https://aws.amazon.com/ec2/pricing/on-demand/.
AZURE, Microsofit (n.d.). Windows Virtual Machines Pricing. Accessed: 28 April
2025. url:
https : / / azure . microsoft . com / en - us / pricing / details /
virtual-machines/windows/.
BAKER, Walter L, Michael V MARN, and Craig C ZAWADA (2010). The Price
Advantage. 2nd. Hoboken, New Jersey: John Wiley & Sons.
BELLAICHE, Lucas et al. (2023). “Humans versus AI: whether and why we
prefer human-created compared to AI-created artwork”. In: Cognitive Research:
Principles and Implications 8.1, p. 42.
BICK, Alexander, Adam BLANDIN, and David J DEMING (2024). The rapid
adoption of generative ai. Tech. rep. National Bureau of Economic Research.
BIERMANN, Phil and Torben PETERSEN (2024a). Generative AI and the price
model revolution in professional services – think golf not tennis. Accessed: 14
February 2025. url:
https : / / www . simon - kucher . com / en / insights /
generative- ai- and - price- model- revolution - professional- services-
think-golf-not-tennis.
—
(2024b). The “AI-pocalypse” in professional services? More Roman Empire
than dinosaurs. Accessed: 14 February 2025. url:
https : / / www . simon -
102
kucher.com/en/insights/ai- pocalypse- professional- services- more-
roman-empire-dinosaurs.
BLUT, Markus et al. (2024). “Customer perceived value: a comprehensive meta-
analysis”. In: Journal of service research 27.4, pp. 501–524.
BREIDERT, Christoph, Michael HAHSLER, and Thomas REUTTERER (2006).
“A review of methods for measuring willingness-to-pay”. In: Innovative marketing
2.4.
BRYNJOLFSSON, Erik, Danielle LI, and Lindsey RAYMOND (2025). “Generative
AI at work”. In: The Quarterly Journal of Economics, qjae044.
CHAN, David (2024). AI’s $600B Question. Accessed: 10 February 2025. url:
https://www.sequoiacap.com/article/ais-600b-question/.
CLARK, Dorie, Alisa COHN, and Marshall GOLDSMITH (2019). A Short Guide
to Pricing Your Services as a Consultant or Coach. Accessed: 21 February
2025. url:
https://hbr.org/2019/08/a-short-guide-to-pricing-your-
services-as-a-consultant-or-coach.
COOK, Jan et al. (2024). Quick Answer: How Is ServiceNow Pricing and Packaging
Its GenAI Products? Tech. rep. Accessed: 24 February 2025. Gartner. url:
https://www.gartner.com/document/6183955?ref=algobottomrec&refval=
6147291.
DELAND, Ashley (2022). Understanding Pricing Strategies, Price Points And Maxi-
mizing Revenue. Accessed: 21 February 2025. url:
https://www.forbes.com/
councils/forbesbusinesscouncil/2022/08/22 /understanding - pricing-
strategies-price-points-and-maximizing-revenue/.
DELARUE, Marie-Laure and Marc JESCHONNECK (n.d.). Audit innovation.
Accessed: 14 February 2025. url:
https://www.ey.com/en_gl/services/
audit/innovation.
ETIKAN, Ilker, Sulaiman Abubakar MUSA, Rukayya Sunusi ALKASSIM, et al.
(2016). “Comparison of convenience sampling and purposive sampling”. In:
American journal of theoretical and applied statistics 5.1, pp. 1–4.
EULERICH, Marc et al. (2024). “Is it all hype? ChatGPT’s performance and
disruptive potential in the accounting and auditing industries”. In: Review of
Accounting Studies 29.3, pp. 2318–2349.
FLINT, Daniel J, Robert B WOODRUFF, and Sarah Fisher GARDIAL (1997).
“Customer value change in industrial marketing relationships: a call for new
strategies and research”. In: Industrial marketing management 26.2, pp. 163–175.
FOUNTAINE, Tim, Brian MCCARTHY, and Tamim SALEH (2019). “Building the
AI-powered organization”. In: Harvard Business Review 97.4, pp. 62–73.
103
FRIEDMAN, Hershey and Miriam GERSTEIN (2014). “Innovative Pricing Strategies:
A Primer”. In: Available at SSRN 2480205.
FRIEDMAN, Nat (2021). Introducing GitHub Copilot: your AI pair programmer.
Accessed: 13 February 2025. url:
https://github.blog/news- insights/
product-news/introducing-github-copilot-ai-pair-programmer/.
GALE, Bradley (1994). Managing customer value: Creating quality and service that
customers can see. New York: The Free Press.
GERON, Tomio (2013). Top Startup Incubators And Accelerators: Y Combinator
Tops With $7.8 Billion In Value. Accessed: 13 February 2025. url:
https:
//www.forbes.com/sites/tomiogeron/2012/04/30/top-tech-incubators-
as-ranked-by-forbes-y-combinator-tops-with-7-billion-in-value/.
GIOIA, Dennis A, Kevin G CORLEY, and Aimee L HAMILTON (2013). “Seeking
qualitative rigor in inductive research: Notes on the Gioia methodology”. In:
Organizational research methods 16.1, pp. 15–31.
GOLDMAN SACHS (2023). AI Investment Forecast to Approach $200 Billion
Globally by 2025. Accessed: 6 February 2025. url:
https://www.goldmansachs.
com / insights / articles / ai - investment - forecast - to - approach - 200 -
billion-globally-by-2025.
GOODMAN, Leo A (1961). “Snowball sampling”. In: The annals of mathematical
statistics, pp. 148–170.
GOOGLE CLOUD (2024). 321 real-world gen AI use cases from the world’s leading
organizations. Accessed: 8 February 2025. url:
https://cloud.google.com/
transform/101- real- world- generative- ai- use- cases- from- industry-
leaders.
—
(n.d.). VM instance pricing. Accessed: 28 April 2025. url:
https://cloud.
google.com/compute/vm-instance-pricing?hl=en.
GOTTS, Ian (2024). GPT Is Disrupting The Consulting Industry. Accessed: 15
February 2025. url:
https://www.forbes.com/councils/forbestechcouncil/
2024/01/12/gpt-is-disrupting-the-consulting-industry/.
GRAF, Albert and Peter MAAS (2008). “Customer value from a customer perspective:
a comprehensive review”. In: Journal für Betriebswirtschaft 58, pp. 1–20.
GRIFFITH, David E and Roland T RUST (1997). “The price of competitiveness
in competitive pricing”. In: Journal of the Academy of Marketing Science 25,
pp. 109–116.
GRÖNROOS, Christian (1990). Service management and marketing. Vol. 27. Lex-
ington books Lexington, MA.
104
GUERREIRO, Reinaldo and Juliana Ventura AMARAL (2018). “Cost-based price
and value-based price: are they conflicting approaches?” In: Journal of Business
& Industrial Marketing 33.3, pp. 390–404.
HINTERHUBER, Andreas (2004). “Towards value-based pricing—An integrative
framework for decision making”. In: Industrial marketing management 8, pp. 765–
778.
—
(2008). “Customer value-based pricing strategies: why companies resist”. In:
Journal of business strategy 29.4, pp. 41–50.
HINTERHUBER, Andreas and Stephan LIOZU (2012). “Is It Time to Rethink Your
Pricing Strategy?” In: MIT Sloan Management Review 53.4, pp. 68–77.
HINTERHUBER, Andreas and Todd C SNELGROVE (2022). Value First, Then
Price: Building Value-based Pricing Strategies. 2nd. New York, NY: Taylor &
Francis.
HOMBURG, Christian, Nicole KOSCHATE, and Wayne D HOYER (2005). “Do
satisfied customers really pay more? A study of the relationship between customer
satisfaction and willingness to pay”. In: Journal of marketing 69.2, pp. 84–96.
HONG, Joo-Wha (2021). “Artificial intelligence (AI), Don’t surprise me and stay in
your lane: An experimental testing of perceiving humanlike performances of AI”.
In: Human Behavior and Emerging Technologies 3.5, pp. 1023–1032.
HOWARD, Tom (2025). Robots could speed up audits — but they’re unlikely to cut
fees. Accessed: 21 February 2025. url:
https://www.thetimes.com/business-
money/technology/article/robots-could-speed-up-audits-but-theyre-
unlikely - to - cut - fees - 5b9hz5vsn ? utm _ source = chatgpt . com & region =
global.
IBM (n.d.). IBM’s computer checkmated a human chess champion in a computing
tour de force. Accessed: 12 February 2025. url:
https:/ / www.ibm .com/
history/deep-blue.
INGENBLEEK, Paul, Marion DEBRUYNE, et al. (2003). “Successful new product
pricing practices: a contingency approach”. In: Marketing letters 14, pp. 289–305.
INGENBLEEK, Paul, Ruud FRAMBACH, and Theo VERHALLEN (2013). “Best
practices for new product pricing: Impact on market performance and price level
under different conditions”. In: Journal of Product Innovation Management 30.3,
pp. 560–573.
INGENBLEEK, Paul and Ivo VAN DER LANS (2013). “Relating price strategies and
price-setting practices”. In: European Journal of Marketing 47.1/2, pp. 27–48.
KAPLAN, Soren (2024). The Business Consulting Industry Is Being Disrupted, and
Nothing Can Stop It. Accessed: 15 February 2025. url:
https://www.inc.com/
105
soren-kaplan/the-business-consulting-industry-is-being-disrupted-
nothing-can-stop-it.html.
KÖBIS, Nils and Luca D MOSSINK (2021). “Artificial intelligence versus Maya
Angelou: Experimental evidence that people cannot differentiate AI-generated
from human-written poetry”. In: Computers in Human Behavior 114, p. 106553.
LANGE, Ruben de, Joep van den NIEUWENHOFF, and Gregor BILJARDT (2025).
The key to AI monetization for tech companies. Accessed: 26 February 2025.
url:
https://www.simon-kucher.com/en/insights/key-ai-monetization-
tech-companies.
LECUN, Yann et al. (1998). “Gradient-based learning applied to document recogni-
tion”. In: Proceedings of the IEEE 86.11, pp. 2278–2324.
LERMANN HENESTROSA, Angelica and Joachim KIMMERLE (2024). “The
effects of assumed AI vs. human authorship on the perception of a gpt-generated
text”. In: Journalism and Media 5.3, pp. 1085–1097.
LESZINSKI, Ralf and Michael V MARN (Feb. 1997). “Setting value, not price”. In:
McKinsey Quarterly.
LIOZU, Stephan (2017). “State of value-based-pricing survey: Perceptions, challenges,
and impact”. In: Journal of Revenue and Pricing Management 16, pp. 18–29.
MAHMOOD, Rafid (2024). “Pricing and competition for generative ai”. In: Advances
in Neural Information Processing Systems 37, pp. 75727–75748.
MARR, Bernard (2022). What Does ChatGPT Really Mean For Businesses? Accessed:
13 February 2025. url:https://www.forbes.com/sites/bernardmarr/2022/
12/28/what-does-chatgpt-really-mean-for-businesses/.
—
(2023). The Most Thought Provoking Generative Artificial Intelligence Quotes
Of 2023. Accessed: 12 February 2025. url:
https://www.forbes.com/sites/
bernardmarr / 2023 / 11 / 29 / the - most - thought - provoking - generative -
artificial-intelligence-quotes-of-2023/.
MARTINA, Rebecca, Jo LIVERSIDGE, and Hannah DECKER (2024). Quick An-
swer: Salesforce Einstein 1 Pricing and Packaging Options. Tech. rep. Accessed:
24 February 2025. Gartner. url:
https:// www .gartner. com /document/
5523595?ref=authbottomrec&refval=6147291.
MAYER, Hannah et al. (2025). Superagency in the workplace: Empowering peo-
ple to unlock AI’s full potential. Accessed: 13 February 2025. url:
https:
/ / www . mckinsey . com / capabilities / mckinsey - digital / our - insights /
superagency - in - the - workplace - empowering - people - to - unlock - ais -
full-potential-at-work.
106
MICROSOFT (n.d.). Microsoft 365 Copilot. Accessed: 26 February 2025. url:
https://www.microsoft.com/en-us/microsoft-365/copilot/enterprise.
MONROE, Kent B. (2003). Pricing: Making Profitable Decisions. 3rd. New York:
McGraw-Hill/Irwin.
MYERS, Michael D and Michael NEWMAN (2007). “The qualitative interview in IS
research: Examining the craft”. In: Information and organization 17.1, pp. 2–26.
NAGLE, Thomas and R HOLDEN (2018). The strategy and tactics of pricing. 6th.
New York: Routledge.
NOBLE, Peter M and Thomas S GRUCA (1999). “Industrial pricing: Theory and
managerial practice”. In: Marketing Science 18.3, pp. 435–454.
OPENAI (2022). Introducing ChatGPT. Accessed: 6 February 2025. url:
https:
//openai.com/index/chatgpt/.
PARASURAMAN, Ananthanarayanan (1997). “Reflections on gaining competitive
advantage through customer value”. In: Journal of the Academy of Marketing
Science 25, pp. 154–161.
PARKER, Charlie, Sam SCOTT, and Alistair GEDDES (2019). “Snowball sampling”.
In: SAGE research methods foundations.
PATTANAYAK, S (2020). “Generative AI in Business Consulting: Analyzing its
Impact on Client Engagement and Service Delivery Models”. In: International
Journal of Enhanced Research in Management & Computer Applications 9.3,
pp. 5–11.
PIERCY, Nigel F, David W CRAVENS, and Nikala LANE (2010). “Thinking
strategically about pricing decisions”. In: Journal of Business Strategy 31.5,
pp. 38–48.
PIGOU, Arthur Cecil (2017). The Economics of Welfare. Originally published in
1920. London: Macmillan and Co.
PINEDA, John et al. (2024). GenAI Needs Pricing Strategies to Match Its Potential.
Accessed: 25 February 2025. url:
https://www.bcg.com/publications/2024/
genai-needs-pricing-strategies-to-match-its-potential.
POWELL, Emma (2025). “AI is biggest shift of our lifetimes, says Google boss
Sundar Pichai”. In: The Times. Accessed: 12 February 2025. url:
https://
www.thetimes.com/business-money/technology/article/ai-is-biggest-
shift-of-our-lifetimes-says-google-boss-sundar-pichai-75n3d8d0n.
RAPOPORT, Gene et al. (2024). AI Survey: Four Themes Emerging. Accessed:
10 February 2025. url:
https://www.bain.com/insights/ai-survey-four-
themes-emerging/.
107
RAUTAKOURA, Annika (2024). Value-Based Pricing Requires a Fundamental Mind-
set Shift in Companies – “It Is Time for a Change”. Accessed: 21 February 2025.
url:
https://www.aaltoee.fi/en/aalto-leaders-insight/2024/value-
based- pricing-requires-a-fundamental-mindset- shift- in- companies-
it-is-time-for-a-change.
RAVALD, Annika and Christian GRÖNROOS (1996). “The value concept and
relationship marketing”. In: European Journal of Marketing 30.2, pp. 19–30.
ROSÁRIO, Albérico Travassos (2024). “Generative ai and generative pre-trained
transformer applications: Challenges and opportunities”. In: Making Art With
Generative AI Tools, pp. 45–71.
SALEM KHALIFA, Azaddin (2004). “Customer value: a review of recent literature
and an integrative configuration”. In: Management decision 42.5, pp. 645–666.
SALESFORCE (2023). Salesforce Announces Einstein GPT, the World’s First
Generative AI for CRM. Accessed: 14 February 2025. url:
https: //www.
salesforce.com/news/press-releases/2023/03/07/einstein-generative-
ai/.
SANDMO, Agnar (2011). Economics evolving: A history of economic thought.
Princeton University Press.
SANTORO, John and Jim HARE (2025). How to Package and Price Your AI-
Enhanced Product. Tech. rep. Accessed: 24 February 2025. Gartner. url:
https:
//www.gartner.com/document/6147291?ref=solrAll&refval=454058487&.
SAP (2024). 5 Major Generative AI Innovations Unveiled at SAP TechEd 2024. Ac-
cessed: 13 February 2025. url:
https://community.sap.com/t5/artificial-
intelligence - and - machine - learning - blogs / 5 - major - generative - ai -
innovations-unveiled-at-sap-teched-2024/ba-p/13898803.
SENGAR, Sandeep Singh et al. (2024). “Generative artificial intelligence: a sys-
tematic review and applications”. In: Multimedia Tools and Applications, pp. 1–
40.
SHETH, Jagdish N, Bruce I NEWMAN, and Barbara L GROSS (1991). “Why we
buy what we buy: A theory of consumption values”. In: Journal of business
research 22.2, pp. 159–170.
SIMON, Hermann, Stephan A BUTSCHER, and Karl-Heinz SEBASTIAN (2003).
“Better pricing processes for higher profits”. In: Business Strategy Review 14.2,
pp. 63–67.
SINGLA, Alex et al. (2024). The State of AI in Early 2024: Gen AI Adoption
Spikes and Starts to Generate Value. Accessed: 10 February 2025. url:
https:
108
/ / www . mckinsey . com / capabilities / quantumblack / our - insights / the -
state-of-ai.
STOBIERSKI, Tim (2022). A BEGINNER’S GUIDE TO VALUE-BASED STRAT-
EGY. Accessed: 17 February 2025. url:
https://online.hbs.edu/blog/
post/value-based-strategy.
TELLIS, Gerard J (1986). “Beyond the many faces of price: an integration of pricing
strategies”. In: Journal of marketing 50.4, pp. 146–160.
THE ECONOMIST (2015). Y Combinator, the X Factor of tech. Accessed: 13
February 2025. url:
https://www.economist.com/business/2015/11/05/y-
combinator-the-x-factor-of-tech.
TIGRE MOURA, Francisco, Chiara CASTRUCCI, and Clare HINDLEY (2023).
“Artificial intelligence creates art? An experimental investigation of value and
creativity perceptions”. In: The Journal of Creative Behavior 57.4, pp. 534–549.
TURING, Alan M (1950 [2009]). Computing machinery and intelligence. Springer,
Dordrecht.
VARIAN, Hal R (1989). “Price discrimination”. In: Handbook of industrial organi-
zation 1, pp. 597–654.
VIEIRA, Valter Afonso (2013). “Antecedents and consequences of perceived value: A
meta-analytical perspective”. In: Journal of Customer Behaviour 12.2-3, pp. 111–
133.
WARREN, Zach et al. (2024). 2024 Generative AI in Professional Services. Accessed:
13 February 2025. url:
https://www.thomsonreuters.com/content/dam/ewp-
m/documents/thomsonreuters/en/pdf/reports/tr4322226_rgb.pdf.
WASWANI, A et al. (2017). “Attention is all you need”. In: Advances in Neural
Information Processing Systems.
WOODALL, Tony (2003). “Conceptualising ‘value for the customer’: an attributional,
structural and dispositional analysis”. In: Academy of Marketing Science review
12.1, pp. 1–42.
WOODRUFF, Robert B (1997). “Customer value: the next source for competitive
advantage”. In: Journal of the academy of marketing science 25, pp. 139–153.
Y COMBINATOR (n.d.). Startup Directory. Accessed: 13 February 2025. url:
https://www.ycombinator.com/companies.
ZEITHAML, Valarie A (1988). “Consumer perceptions of price, quality, and value:
a means-end model and synthesis of evidence”. In: Journal of marketing 52.3,
pp. 2–22.
ZHANG, Duzhen et al. (2024). “Mm-llms: Recent advances in multimodal large
language models”. In: arXiv preprint arXiv:2401.13601.
109
ZHANG, Emma Yann et al. (2023). “From turing to transformers: A comprehensive
review and tutorial on the evolution and applications of generative transformer
models”. In: Sci 5.4, p. 46.
ZHANG, Hao, Xiaofei BAI, and Zengguang MA (2022). “Consumer reactions to
AI design: Exploring consumer willingness to pay for AI-designed products”. In:
Psychology & Marketing 39.11, pp. 2171–2183.
110
A Interview Guide and Questions
Interview Template
The following interview template was used to guide the interview, ensuring that all
relevant aspects were covered before, during, and after the interview.
1. Introduction, background, and aim of the thesis
2. Sharing the results when the thesis is finalised
3. Stating that all interviews are fully anonymous
4. Asking for permission to record the audio for transcription, deleted after
5. If permission granted: Start recording (both phone and computer), otherwise
take notes by hand
6. Ensuring that it is OK to use the transcription for other research purposes
7.
Clarifying which area the interviewee feels most comfortable speaking from, if
not agreed beforehand
8. Interview questions according to the scope
9. Asking if the interviewee has anything else to share related to the thesis
10. Stop audio recording
11. Ask if the interviewee has anyone in their network that I should speak with
12. End the interview
Interview Questions - Buying
1. Background
(a) Could you briefly describe the company and your role?
(b)
Has your organisation integrated or procured any generative AI (gen AI)
services or solutions?
i.
If yes, could you describe the type of services and their intended
benefits?
2. Value
(a) Criteria for Value Assessment
i. When deciding to purchase a service (AI-based or otherwise), which
factors or drivers are most important in determining its value to your
organisation?
ii.
How do you evaluate where you should incorporate gen AI products
and services?
A. Who decides, and what are the measurements?
111
(b) Comparing Human vs. AI-Integrated Services
i.
In your experience, how does the quality or effectiveness of generative
AI-produced output (fully AI and partially AI) compare to content
created by humans?
A.
What specific differences in value (if any) do you notice? Higher,
lower, same?
B.
What factors make one option more appealing, and when would
you use an AI vs. a human?
ii.
Do you compare AI products to other software products or human-
generated services?
iii.
Does it matter to you (or your organisation) whether a deliverable
was generated by AI or by a human expert?
A.
For instance, if you know the content was AI-generated, does that
change your trust in it or how much you’re willing to pay for it?
(c) KPI and Success Metrics
i.
If you have experience evaluating gen AI offerings, what key perfor-
mance indicators (KPIs) or success metrics do you use to measure
their value or effectiveness of AI solutions vs human solutions?
A. Why is it different?
(d) Desired Value Outcomes
i.
What kind of evidence or communication from a vendor helps convince
you of a GenAI product’s value?
A.
For example, do you look for case studies, ROI calculations, pilot
project results, etc.?
ii.
What primary advantages or value benefits do you seek when exploring
or purchasing services enhanced by gen AI?
iii.
How would you describe the quality difference (if any) between AI-
generated vs. human-generated outputs in your experience?
A. Do you measure quality differently when AI is involved?
3. Pricing
(a) Observed Pricing Models
i. How are you currently paying for services?
A.
What if you knew that 10%, 50%, or 99% of the work was done
with AI — how would it change your view of a fair price?
ii.
Which pricing models for gen AI-enhanced services have you encoun-
112
tered? (e.g., seat-based, pay-per-use, outcome-based, token, etc.)
(b) Most Appealing Pricing Approach
i.
Do you find any specific pricing model for gen AI services more
compelling or fair than others?
A. Could you explain the reasons behind your preference?
B.
Which model best aligns with how you derive value from the
service?
C.
How does the pricing model factor into your ability to pay for it?
(known cost vs. adaptive cost)?
ii.
When you think of fair or best-value pricing, do you compare AI-based
offerings primarily to other AI providers, to human-based services, or
to internal benchmarks?
A. Which reference points matter most?
(c) Fair Pricing Characteristics
i.
Do you feel the prices you’re seeing for AI-enabled services are fair
relative to the value they provide?
A.
Should price reflect lower costs or the high value/ROI it delivers?
ii.
What elements constitute a “fair” pricing model for AI-enhanced
services, from both the client’s and provider’s perspectives?
iii.
Do you think it’s fair for vendors to charge different customers different
prices for essentially the same generative AI capability?
A.
How would you feel if you discovered that another client is paying
a different rate for the same AI-driven service?
iv.
What other cost factors do you consider when purchasing new products
and services?
A. How will these change with gen AI?
v.
Do you foresee scenarios where you’d pay different prices for AI-based
services depending on the type of project, volume of work, or level of
customisation?
A. How does it differ from current offerings?
4. Willingness to Pay (WTP)
(a) AI-Generated Service Disclosure
i.
Does knowing that part of the service is generated by AI influence
your willingness to pay? If so, how?
(b) Production Cost vs. Willingness to Pay
113
i.
Are you willing to pay a premium for a service because it incorporates
generative AI, or do you expect it to cost less than a traditional one?
A. Why — outcomes, novelty, or reduced labour?
ii.
How does learning that generative AI might reduce the provider’s
production cost affect your willingness to pay?
5. Other
(a) Risks
i.
Are there any risks or reservations that make you hesitant to adopt
or pay for AI-driven services (e.g., reliability, data security, hidden
costs)?
Interview Questions - Selling
1. Background
(a) Could you briefly describe the company and your role?
(b)
How is your company currently using generative AI in products or services?
2. Value
(a) How do you assess the value your product delivers to customers?
i. What key factors and KPIs do you use?
ii.
How do you determine which aspects of value are most important to
customers?
(b)
From your interactions with customers, how do they perceive AI-generated
outputs compared to human-created outputs?
i. Have clients commented on quality, creativity, or value differences?
(c) How do you expect generative AI to impact perceived customer value?
i. What factors might increase or decrease value?
(d) How do you demonstrate or quantify the value of your GenAI solution?
i. What proof points do you use to justify the price?
ii. Do you use specific ROI metrics?
iii. Have perceptions changed after communicating the AI value?
(e)
Are there changes in customer understanding of value when gen AI is
used?
i. Differences between customer groups?
A. Who is supportive, more negative, etc.?
(f) When should humans be used, and when should AI?
(g)
SERVICES: Do you discuss with clients whether the service includes
114
AI-generated elements?
i. Why or why not?
3. Willingness to Pay
(a)
How do you anticipate customer willingness to pay (WTP) will change
with AI integration?
i. What key factors would influence it?
ii. What factors could limit or increase willingness to pay?
(b) Have you observed changes in willingness to pay with gen AI features?
i.
Are some clients willing to pay more or less based on cost savings or
speed?
4. Pricing
(a) What is your familiarity with pricing strategies and models?
(b) What is your current pricing strategy?
i. Cost-based, competition-based, value-based?
(c) What is your current pricing model?
i.
PRODUCT: outcome-based, usage-based, seat-based, tier-based, flat-
rate, credit/token, freemium?
ii.
SERVICES: hourly billing, hybrid (software + hourly), output-based,
outcome-based?
(d)
Does your current pricing model align with how you believe AI value
should be captured?
(e) How do you expect gen AI to influence your pricing model?
(f) In what ways (if any) have you adjusted pricing since adding gen AI?
(g) How well do current models work for AI-enhanced offerings?
(h) Have you tested new models for gen AI?
i. Why does this model best capture the GenAI value?
ii. What has your experience been?
iii. How do you measure the impact of a pricing model change?
(i) How have customers responded to your GenAI pricing?
i. Any challenges with consumption-based models?
(j) How do you ensure pricing is perceived as fair?
i. If AI lowers your costs, do you adjust the price or reinvest in value?
(k) How does cost (e.g., compute, overhead) factor into pricing?
i. Best ways to link gen AI cost into pricing?
115
A. How tightly should pricing and cost be linked?
ii. Do you tie pricing to compute intensity or usage?
iii. Do you pass on savings from reduced labour?
iv. How do you handle cost spikes like GPU/infrastructure surges?
v.
Have you turned down customers due to unprofitable usage patterns?
(l) Does gen AI enable more personalised offerings?
(m) Do you use segmented or customised pricing for gen AI?
(n) Do you see potential to differentiate pricing by customer segment?
i. For example: speed, quality, variety delivered by gen AI.
