IN DEGREE PROJECT INDUSTRIAL ENGINEERING AND MANAGEMENT, SECOND CYCLE, 30 CREDITS , STOCKHOLM SWEDEN 2018 A Pricing Model for AIaaS An analysis of a new AI personalization product within the edtech space ZENJA JEFIMOVA SOFIE NABSETH KTH ROYAL INSTITUTE OF TECHNOLOGY SCHOOL OF INDUSTRIAL ENGINEERING AND MANAGEMENT
96
Embed
A Pricing Model for AIaaS - Divakth.diva-portal.org/smash/get/diva2:1259306/FULLTEXT01.pdfKTH Industriell teknik och management Industriell ekonomi och organisation SE-100 44 STOCKHOLM
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
IN DEGREE PROJECT INDUSTRIAL ENGINEERING AND MANAGEMENT,SECOND CYCLE, 30 CREDITS
, STOCKHOLM SWEDEN 2018
A Pricing Model for AIaaSAn analysis of a new AI personalization product within the edtech space
ZENJA JEFIMOVA
SOFIE NABSETH
KTH ROYAL INSTITUTE OF TECHNOLOGYSCHOOL OF INDUSTRIAL ENGINEERING AND MANAGEMENT
A Pricing Model for AIaaS
An analysis of a new AI personalization product within the edtech space
by
Zenja Jefimova and Sofie Nabseth
Master of Science Thesis INDEK 2018:331
KTH Industrial Engineering and Management
Industrial Management
SE-100 44 STOCKHOLM
En Prismodell för AIaaS
En analys av en ny AI-baserad personifieringsprodukt inom edtech
av
Zenja Jefimova och Sofie Nabseth
Examensarbete INDEK 2018:331
KTH Industriell teknik och management
Industriell ekonomi och organisation
SE-100 44 STOCKHOLM
Master of Science Thesis INDEK 2018:331
A Pricing Model for AiaaS
An analysis of a new AI personalization
product within the edtech space
Zenja Jefimova
Sofie Nabseth
Approved
2018-06-01
Examiner
Gregg Vanourek
Supervisor
Terrence Brown
Commissioner
Sana Labs AB
Contact person
Joel Hellermark
Abstract
As pricing is vital for an organization’s marketing strategy, it is a significant area to consider for companies offering new products where Artificial Intelligence as a service (AIaaS) is provided. The purpose of this study was to investigate possible pricing models for an AIaaS product. The study was delimited to the edtech industry. The main research question to be investigated was “What pricing model should an AI-company have for its B2B personalization product to correspond to the value delivered by it?”. Sub-research questions consisted of how perceived value can be related to price, what factors the organization should consider for the pricing model and the implications of implementation.
The exploratory research was carried out through a literature review, a survey where Van Westendorp’s Price Sensitivity Meter was applied, as well as in-depth interviews to gather qualitative data. The quantitative results showed that the price sensitivity depends on the number of monthly active users a platform has, where there is a negative relationship between the number of monthly active users and the price willing to pay per learner. The qualitative results showed that the perceived value depends, amongst other factors, on which segment the buyer belongs to. The primary results were discussed with the findings from the literature review, which mainly consisted of pricing model for Software as a Service (SaaS) products, that resulted in a designed pricing model for AIaaS providers.
The conclusion of the study is that pricing an entirely new product is complicated as the buyer does not know the value of the product. Also, there does not exist one single value which can be quantified and translated into price; the price must be adjusted according to the segment’s perceived value. The pricing model presented accounts for adjustable variables needed to be considered by an AIaaS provider before determining a price.
Key-words: AI, AIaaS, edtech, education technology, personalization, price, pricing, pricing strategies, pricing models, pricing tools, pricing AIaaS, software pricing, pricing for new products, value, value adding
Examensarbete INDEK 2018:331
En Prismodell för AIaaS
En analys av en ny AI-baserad
personifieringsprodukt inom edtech
Zenja Jefimova
Sofie Nabseth
Godkänt
2018-06-01
Examinator
Gregg Vanourek
Handledare
Terrence Brown
Uppdragsgivare
Sana Labs AB
Kontaktperson
Joel Hellermark
Sammanfattning
Då prissättning är viktigt för en organisations marknadsstrategi, är det ett betydelsefullt område att ta hänsyn till för organisationer som erbjuder nya produkter baserade på Artificiell Intelligens as a Service (AIaaS). Syftet med denna studie var att undersöka möjliga prismodeller för AIaaS produkter. Studien var begränsad till edtech industrin. Den huvudsakliga forskningsfrågan var “Vilken prismodell borde ett AI-bolag ha för att dess B2B personifieringsprodukt ska motsvara det levererade värdet ?”. Delforskningsfrågor bestod av hur uppfattat värde kan relateras till pris, vilka faktorer en organisation bör ta hänsyn till för en prismodell samt implikationerna av att implementera den.
Den utforskande studien genomfördes genom en litteraturstudie, en webbenkät där Van Westendorps priskänslighetsmätare applicerades, såväl som fördjupningsintervjuer för att samla kvalitativ data. De kvantitativa resultaten visade att priskänsligheten beror av antalet månatliga aktiva användare som plattformen har, vilket visar på ett negativt samband mellan antalet månatliga aktiva användare och priset en är villig att betala per elev. De kvalitativa resultaten visade att det uppfattade värdet beror av vilket segment köparen tillhör. De primära resultaten diskuterades mot resultaten från litteraturstudien, vilka främst bestod av prismodeller för SaaS produkter, och slutade i en framtagen prismodell för AIaaS leverantörer.
Slutsatsen av studien är att prissättning av en ny produkt är komplicerat eftersom köparen inte vet värdet av produkten. Det finns heller inte ett enskilt värde som kan kvantifieras och översättas till ett pris; priset måste anpassas enligt segmentets uppfattade värde. Prismodellen som presenteras tar hänsyn till justerbara variabler som en AIaaS leverantör måste utvärdera innan ett pris bestäms.
Nyckelord: AI, AIaaS, edtech, online-utbildning, personifiering, pris, prissättning, prisstrategi, prismodell, prisverktyg, prissättning för AIaaS, prissättning för mjukvara, prissättning för nya produkter, värdeskapande
Abbreviations
AIaaS Artificial Intelligence as a Service
AIP Artificial Intelligence Provider
B2S Business to School
Edtech Educational Technology
MAU Monthly Active User
OCP Online Course Provider
Definitions
Edtech: “the study and ethical practice of facilitating learning and improving performance by
creating, using, and managing appropriate technological processes and resources” (Richey, et
al., 2008).
AI: “the study of how to make computers do things at which, at the moment, people do
better” (Rich & Knight, 1991). This is a definition that will always be relevant, not only in the
1.5 RESEARCH QUESTIONS .............................................................................................................................. 3
2.2 THE IMPACT OF PRICE ............................................................................................................................... 6
2.3 STRATEGY, MODEL AND TOOL AS A FUNNEL OF PRICING .................................................................. 7
2.5.2 Pricing for New Products ........................................................................................................................ 14
2.5.3 Performance Based Pricing ...................................................................................................................... 14
3.1 RESEARCH DESIGN ................................................................................................................................... 18
3.2 DATA COLLECTION .................................................................................................................................. 19
3.2.1 Quantitative Sampling through a Survey .................................................................................................. 19
3.2.2 Qualitative Sampling through In-Depth Interviews ................................................................................... 20
3.3 APPLICATION OF LITERATURE AND THEORY ....................................................................................... 21
3.3.1 Theory Bits ............................................................................................................................................ 22
3.3.2 Van Westendorp Price Sensitivity Meter.................................................................................................. 22
3.3.3 Value Determination Inspired by CBC ................................................................................................... 23
3.6 AI FOR EDUCATION.................................................................................................................................. 28
4. RESULTS AND ANALYSIS ........................................................................................................... 30
4.3.3 Cloud and IT costs ................................................................................................................................. 41
4.3.4 In-house Development Costs .................................................................................................................... 42
4.5 VALUE PERCEPTION ................................................................................................................................. 44
4.6.1 Price Based on Segment........................................................................................................................... 47
4.6.2 Issues with AI in Education ................................................................................................................... 49
5.2 VALUE PERCEPTION ................................................................................................................................. 52
5.3.2 Issues with AI in Education ................................................................................................................... 58
5.4 PROPOSAL OF A PRICING MODEL FOR AIAAS ...................................................................................... 59
6.1 MAIN FINDINGS......................................................................................................................................... 66
6.4 FUTURE RESEARCH ................................................................................................................................... 68
As it is not of importance to examine the product choices per se, the product choices are
presented and later discussed in section 5.1 with a focus on the attributes that the selected
products consist of. The most apparent and significant results from the ten product choice
questions are presented in Table 7.
Table 7. The most apparent results from respondents’ product choices
Question # Findings
1 When the price remained constant at USD 10,000, six out of the 16 respondents would
not choose any of the three product bundling alternatives presented. Five out of 16
would prefer the option with 2x learning efficacy and dedicated solutions engineer for the price
of USD 10,000
2 No conclusion regarding respondents’ preferences can be drawn as all four alternatives
were chosen equally amount of times
3 Seven out of 16 respondents chose the alternative with 2x learning efficacy, USD 7,500
and weekly performance report
4 The equally preferred choices with six scores each are the combination 12% improved churn rate, USD 7,500 and dedicated solutions engineer and the combination 12% improved churn rate, USD 5,000 and no additional services
5 Eight out of 16 respondents chose 2x learning efficacy, USD 10,000 and weekly performance report. Five respondents would not choose any of the products presented
6 Ten out of 16 respondents chose 2x learning efficacy, USD 5,000 and no additional services
7 The service dedicated solutions engineer appears to be more appealing than weekly performance report
8 Eight out of 16 respondents chose 2x learning efficacy, USD 5,000 and no additional services
9 Nine out of 16 respondents chose 2x learning efficacy, USD 5,000 and no additional services
10 Eight out of 16 respondents chose 12% improved churn rate, USD 7,500 and executive briefings
44
4.5 Value Perception
This section presents the most common and significant themes highlighted during the semi-
structured part of the in-depth interviews. These results are related to value perception and
price for a B2B personalization product within edtech.
4.5.1 KPIs
The 12% improved churn rate KPI is important or the most important KPI for some
organizations (R27; R30; R34; R38; R42; R43). One organization expresses that it
accomplishes other KPIs (R27) and provides a holistic view of an organization’s performance
since the KPI can be translated into an increase in revenue for the organization (R30; R42)
which makes it easier to estimate the product’s actual worth for the organization (R42). One
of the organizations did not consider this KPI to be important (R36). An organization who
provides support and courseware for educational institutions expressed that it is not
concerned with churn rate since it is the schools’ concern (R33). It is also expressed that this
KPI is more applicable to higher education than to K12 (R40). An issue with measuring the
churn rate within the K12 segment is that the students usually get assigned to use the platform
as part of the course work and might not have a choice whether to use it or not which makes
the KPI not as useful. For that reason, it is suggested that a better KPI would be engagement
expressed in daily, weekly or monthly terms (R24).
The KPI 2x learning efficacy is important or the most important KPI for organizations (R24;
R28; R34; R35; R36; R43; R44) and claimed that when this KPI is reached, the other two KPIs
will automatically be reached as well (R28). An organization expressed that they would be
willing to pay millions to have a tool that guarantees 100% increase in efficacy (R24) while
another organization is willing to pay anything for a product that brings this increase in
efficacy KPI (R44).
One organization expressed that this KPI is more important than 4.5x more problems solved
(R30). Another organization that practices subscription per student per school as a pricing
model to sell their product towards K12, expresses that for their business model, it does not
matter how long time the learning takes (R42). Another organization emphasized the
ambiguity in promising 12% improved churn rate while at the same time delivering 2x learning
efficacy since it could be translated to that it requires a rather “low” effort (12% improved churn
rate) for duplicating the learning efficacy (R40). Further, one organization expressed that
learning efficacy is the goal of the organization which is reached without this kind of AI-
personalization product (R33). Another respondent highlighted that learning efficacy is a
highly complex KPI which takes long time to measure (R35) and can be reached by asking the
right questions which is enabled through ML (R39). A respondent argued for that reason that
using this kind of KPI would equal to simplifying something that is of high complexity (R35).
A B2B-organization within the edtech space expresses that, specifically for them, it is not
about increasing the end-learners efficacy but rather equip the teachers to be as efficient as
they can be when teaching. For that reason, it is more important for them to increase the
teachers’ efficacy by providing solutions that counter the problem of not having enough time
to teach. The same respondent added that it does not matter if the learners use their product
more efficiently since the K12-segment is dedicating a certain amount of teaching hours
45
regardless of how efficient the learning is. For that reason, since a teacher is deciding on the
teaching content and supports the learning, it is beneficial to have all students at the same
level. What the organization does to personalize their offering is to provide tips and
recommendations on what to re-read. For that reason, it is claimed that efficacy for students is
not important as the learning can take more or less time and the schools are not striving for
making it more efficient (R42). An e-learning organization within the enterprise segment
questions the concept of improved learning efficacy as “faster training” is not their intention.
It is important that the learners do not speed through the courses as the content is important
for their work (R34).
The 4.5x more problems solved KPI is the most important KPI for some organizations (R29, R43)
and not as important for other organizations (R34; R36). For a software vendor within higher
education, this KPI is considered as very important (R43). It is further formulated that this
KPI is not useful since it focuses on quantity and does not reveal anything about the actual
quality for the end learners (R35; R38). Another organization suggests that rather than
measuring the amount of problems solved, one could put a number on the level of
engagement connected to the learning outcomes (R39). An issue with this KPI is whether it
actually measures how well a learner is learning. The issue is not about how many problems a
learner is facing but rather how appropriate they are (R24). It is desired to have a KPI that
would pay regard to the organization’s return of investment (ROI) (R40; R46) and thereby
base it on the value for the customer (R46).
In addition to the KPI-specific comments above, it was expressed that it must be proven that
the KPIs are working for specific customers and not limited to a certain customer segment
(R35). A respondent means that 100% increase in efficacy or 4.5x more problems solved are
unrealistic while 12% improve in churn rate is more realistic. The organization would for that
reason want independent institutions to guarantee these KPIs for that specific organization
(R44). From one respondent, there is a perception that the KPIs used are reasonable (R36).
However, it is expressed that having a student perspective is important in order to deliver
value to actors within the education industry and to think in terms of “what delivers value to
the learners?”. For that reason, it is important to be able to measure the amount of knowledge
and the intended learning outcomes the learners achieve and fulfill. (R36) Lastly, another
organization expressed the difficulty in quantifying these values into KPIs without considering
the volume and profile of users (R39). One interviewee expressed that different learning
segments care about different KPIs (R43) and that the KPIs must be proved within each
specific segment as it is not good enough to prove KPIs from other case studies (R35).
4.5.2 Price Span
As already mentioned, two organizations expressed that they would be willing to pay millions
or even anything for this kind of product (R24; R44). Three other organizations expressed that
a price of USD 10,000 per month is “incredibly cheap” (R26; R28, R29), especially if it is not
linked to the amount of users connected to the platform (R26). It is supported by other
respondents that the suggested prices are low (R40, R46) and the lowest amount could be
raised to USD 8,000 per month (R40). Another organization expressed that USD 10,000 per
month seemed reasonable (R35). Other organizations expressed that USD 10,000 per month
46
is a pricey investment relative to the organizations’ current goals (R33, R39) since it would
correspond to the cost of 1.5 of their team’s computer engineers (R33).
It was formulated that the current prices used in the structured interview part only served the
upper end of the market (R34). To make this investment appealing, the product must increase
the customers’ revenue. A price of USD 10,000 is not too high if the product would legitimate
the customer to charge a higher price per learner. This could be done by delivering desirable
KPIs for the curriculum partners and for the market, for example a significantly higher level of
learner engagement (R39). It is further stated that larger organizations would be able to pay
USD 10,000 per month but schools would find it expensive. However, if the technology was
proven, it is possible that schools would be willing to pay this amount as well. The willingness
to pay for this product is dependent on the size of the organization (R43). Several
organizations prefer a price setting that is based on the organization’s volume of MAUs since
a price’s impact is different depending on the size of the organization that is buying the
product (R34; R35; R36). On the other hand, it is also mentioned that a fixed price is more
appealing than a volume-based fee (R26) since cost that depends on variables tend to become
much more expensive and uncertain for the organization than what fixed prices are (R39). A
B2B actor expressed that in their position, it is difficult to set a price based on user volume.
The respondent would rather see it formulated in relation to the size of the solution or API
instead (R29). Another desired way to price this kind of offering is by royalty shares in which
both risks and rewards are shared between the two organizations (R39). Other respondents
highlight that the price setting should be per student per year rather than a monthly fee (R36;
R40; R41; R43; R44). If the price is set per student, one would possibly need to use price
discrimination per geography since, for example, schools in China have a lot more students
than schools in Finland which would make the price per student fee too expensive in China
(R40).
A respondent from a company that is scaling rapidly expresses that for companies in this
stage, it is preferred to use a fixed monthly price. On the other hand, for earlier stage start-ups,
the most optimal price model would be “per student price” that scales with the startup before
leveling off to a more beneficial fixed monthly price (R33). It is stated that a subscription fee is
a decent method for pricing the product but another preferred price setting would be a one-
time fee which would allow the organization to own the technology and evolve it like they
want (R35).
The costs associated with content creation should be an additional variable to consider when
evaluating the price for this product (R21). Moreover, the price should not be fixed but rather
related to the price of a textbook which varies greatly in different countries. For that reason,
the price should not be expressed in absolute numbers but rather compared to prices in the
geographic market (R44). Specifically for start-ups, a respondent highlighted the importance of
reducing the entrance price point as much as possible. In this stage, it is critical to show value
very fast, for example by showing the student engagement and performance before and after
the product. However, it is added that some of the factors affecting the sales are intangible
and can be hard to track back to the product provider. A higher level of engagement can
however be quantified into improvement in the customer’s business in long term. When the
customer has become dependent on the product, they will be willing to pay for it since the
organization’s offering will not be as good without it (R45).
47
4.5.3 Additional Services
For an actor within the enterprise segment, it is considered that the most valuable additional
services are executive briefings and weekly performance reports (R34). Other organizations mentioned
weekly performance reports as the most important additional service (R35; R36; R42; R43) and
added that they would even need daily performance reports (R35; R42) since “without
reporting, it means nothing” (R42). On the other hand, another respondent mentioned
executive briefings as the single most important additional service and did not consider the other
two to be valuable (R38). It is expressed that what is considered as the most valuable service
might change over time (R39) as dedicated solutions engineer might be needed during the
implementation and initial phase or occasionally rather than on an ongoing basis like weekly
performance reports (R34, R39). Not to mention, executive briefings may not be needed more
often than once in a quarter (R43). For one organization, dedicated solutions engineer was not
considered as an important additional service since they, within the higher education segment,
rather exclude this service and pay less for the whole product since they can have their own
internal solutions engineer instead (R40). However, for the K12 segment, dedicated solutions
engineer might be an important service as this is something that is too expensive for the actors
to have internally (R43). For another organization, both executive briefings and dedicated
solutions engineer were considered as unimportant (R36). It was also expressed that additional
services are secondary to the choice as the KPIs are the most important factor to consider
(R44). For an actor within the enterprise segment, it is important that the additional services
are adapted to the specific customer’s needs and type of organization (R34).
4.6 Emerged Themes
Besides the areas that affect the value perception of an AI-personalization product, it was
revealed from the survey as well as during the in-depth interviews that there are additional
factors to consider when a price for this kind of product is formulated. The main category for
these factors is segment-based pricing. Aside from this, the respondents highlighted some
issues related to AI as a service in education which could affect the product’s value as well as
price perception.
4.6.1 Price Based on Segment
An actor within K12 tells that “If you can teach the content to 40 students together it will be
cheaper than personalizing each one" (R47). In line with this, it is expressed that schools are
economically restricted (R38) due to price pressure in the market that schools must adapt to
(R33), which makes price an important factor to consider (R38). Another actor mentions that
they are constrained by the budget as well as the fact that a product like this must be refined
continuously (R35). In line with this, it was also expressed that the K12 segment cannot afford
to pay as much as, for example, the enterprise segment (R43). Furthermore, it is said that
schools are very conservative institutions which makes it hard to implement new ideas and
technologies (R44). One interviewee within the K12 segment expressed that, each pupil has a
certain school capitation allowance in which school rent, teachers’ salaries, food and teaching
material are some of the areas that are included. If a learning institution would increase its
costs for teaching material, it would be at the expense of a lower amount of the school
48
capitation allowance to other areas that are needed for a K12 pupil. However, the same
respondent added that if a product can replace teachers or education centers, it is possible to
pay up to 50% more for that product (R36). Moreover, it is mentioned that the price should
be related to what the different school districts are receiving in monetary amount per student
annually (R40). Another organization within this segment would rather formulate the monthly
price expressed per student and per course. Additionally, the same respondent expressed that
the price should be related to the, per student, monetary amount received by the state in which
a participatory share would be better suited than a certain fixed sum (R36). A non-profit
organization towards professional development for teachers expresses that they are not at the
point of personalizing their offering due to lack of funding (R9).
A B2B-organization expressed that the Van Westendorp question was more relevant to answer
for organizations that provide B2C products rather than B2B products. Also, the same
organization stated that their own pricing depends on which country they sell to as they
consider the country-specific market conditions. Examples of that could be the different ways
software is being acquired by schools as this differs widely across countries, for example in
US, every school can make these decisions by itself. An actor in the edtech space that provides
learning that is not yet a part of the standard curriculum for K12, explains that having early
adopters as their customers prohibits them from being able to set a fixed price for their
product. For that reason, to target different schools in different countries, the organization
always needs to negotiate its price within K12 as well as consider competition and their
customers’ size when setting the price. (R42)
The issue of retaining learners is bigger in higher education than in K12. In the K12 segment,
where students are more or less forced to be on a certain platform, actors within the K12-
segment rather want to retain schools to the learning platform, not students (R33; R35; R40;
R42; R44). Other organizations mention that they work with keeping the teachers and parents
engaged to improve the customer retention (R35; R42). One organization within the K12
segment reveals that it is up to the teachers to plan and organize the courses. For that reason,
OCP is not trying to make the students spend more time on their platform (R36). A
respondent tells that it is the teachers that sign up the students to their learning platform and
support them in challenges as they are progressing (R42). It is expressed that one organization
within K12 is interested in any product that would assist the teachers and lighten their
workload in for example marking written assignments (R25). Since the students, at the request
of schools, buy textbooks from e-learning providers, these providers retain the customers by
working with schools and instructors to adopt the courses. After the end of a course they do
not work with retaining the students since it is the schools’ task (R33). An organization
mentions that they work with retention by providing content which is attractive, fun and
engaging for kids since the odds for them staying on the platform increases if it is fun to use
(R27). Another organization expresses their interest in implementing personalized learning and
emphasizes that they would need to have an external provider to realize it (R23).
A respondent within higher education reveals that there are high variations in volume of
students since their customers are colleges and other learning institutions with different focus
depending on if it is spring or fall season (R33). Some material is used continuously and some
is used on specific periods (R33). Another respondent highlighted the term disruptive
education which refers to that people are, as alternative learning rises, starting to question the
49
traditional learning paths (R45). With the rise of disruptive education, universities will partner
with new edtech-companies as more and more people realize that there are situations where
people need to learn faster to get job faster (R45).
Similar to R33, an actor who provides online classes for self-paced learning suggested to
consider peak- as well as low seasons and thereby price differently based on that (R26).
Another organization within this segment expressed that they do not work actively with
retaining their users (R42).
A provider of corporate training reveals that they have a varying client base which makes it
hard for them to offer a fixed fee that would suit all the enterprises’ different needs and
possibilities. For that reason, when talking about price, a fee based per user is more
appropriate. The amount of users decides how much an enterprise is willing to pay per user
and economies of scale is a relevant term in this situation since a company with a lower
amount of users probably do not have the same possibility to pay the same price as a company
with a significantly higher amount of users (R34). It is expressed by an organization that they
are not in charge of retaining the end learners to their platform but rather their customers
which are the enterprises (R29). It is also expressed that within the enterprise segment, it is less
common with personalization in the learning. The reason for that is the wide range of learning
content which makes it hard to interpret what the learners need and when they need it. For
that reason, personalization through AI would be valuable within this segment (R34).
4.6.2 Issues with AI in Education
It is argued that it is difficult to estimate the product’s value and price due to the limited
tangibility of it (R29) and that the value proposition of this kind of product is difficult to
quantify into commercial benefit (R39). A worry is also expressed towards the unproven
commercial viability of AI offers in the marketplace (R34). One respondent expresses that
these kinds of products are far from market reality and points out that digitalization is a huge
investment for publishers as they have to create more material than before (R44). Worries are
for that reason connected to the cost for the solution associated with feeding enough content
into the adaptive engine to support auto generated question types (R39). In line with this, an
organization also expresses that they would need to produce a lot more content for enabling
personalized feedback (R25). Another organization mentions that they are skeptical towards
these products due to that there are so many other publishers of learning content and courses
(R34). In UK, it is common to publish the learning material with a number of different exam
boards which makes it harder to reach a large enough volume for AI and ML engines (R33).
For a publisher, introducing different levels of adaptivity and difficulty would make all
different alternatives to mount up very quickly which imply significant overhead costs (R39).
For that reason, schools and colleges are unsure of the value and emphasizes that it is not only
the cost of the technology that needs to be considered but also the overhead costs associated
with generating metadata and content for the algorithms (R2). One respondent commented
that most of the work for personalization is done in higher education since computer based
learning is more established there than for younger students that do not have laptops (R24).
A provider of corporate training is in contact with an AI organization but they are skeptical as
it seems extremely expensive (R43). Skepticism is also raised towards having an external
provider of this kind of personalization product and it is expressed that the organization
50
would rather want to develop it in-house. The reason for that is because the product is
strongly connected to the core of the organization and they want to have the control of and
decide what problems and learning content that the learners receive (R35). Moreover, it is
expressed that an issue with using AI for personalization in learning is that there are no
international standards for expressing these adaptive question types and no equivalent on how
to interpret the content (R39). It is stressed that grading is an issue when personalized learning
paths are used. The respondent asked “how will teachers set grades on students that took
different paths?” (R44). There are lack of standards for adaptive assessments which is key to
scalability (R39). It is also expressed that personalization works well for self-study but not for
schools and institutions (R44). Another organization expresses that the organization’s goals
are fulfilled without AI and ML and therefore sees no reason to focus on that (R33).
Another organization highlights their restrictions in implementing this kind of product as they
strictly adhere to the Children's Online Privacy Protection Act (COPPA) standards which
prohibit them from collecting any personally identifiable information. The respondent added
that “If we were to personalize our lessons, we would have to reconsider our business model
and our COPPA compliance strategies”. Furthermore, it is important that the tools they
provide teachers with to reinforce the lessons are completely safe to use for their learners, the
pupils. (R5)
Further issues connected to personalized learning through AI are that it takes away the social
aspects of learning in a classroom setting since conversations amongst learners might get lost,
jealousy can be created and the feeling of group cohesion could be threatened (R35). Another
issue highlighted by an organization is that it is not always desired, from the schools’ point of
view, to create too individual learning paths as the pupils should be able to interact with not
only other pupils but also the teacher in a classroom setting. One organization expresses that
they have not been able to focus on personalization through AI yet but could see themselves
do it in the future, depending on how their platform evolves (R20).
51
5. Discussion
In this section, a discussion of the results is presented. It is initiated with the discussion of the product
preferences collected from the structured interviews. The semi-structured part of the in-depth interviews is then
discussed based on the three already set subject areas. Thereafter, the emerged themes; segments and issues with
AI in education, are discussed. This is then related and discussed to eventually reach a proposal of a pricing
model for companies offering AIaaS. After a discussion on sustainability and ethics, the chapter is finalized by
summarizing the findings.
5.1 Product Preferences
Table 6 shows that among the attributes presented, 2x learning efficacy was chosen the highest
amount of times relative to how many times it was available for choice. This number
amounted to 38%. From the semi-structured part of the in-depth interviews, 2x learning efficacy
was considered as important or the most important KPI by seven respondents, which was the
largest category of organizations expressing a KPI as important or the most important. Hence,
since both the structured and the semi-structured part of the in-depth interviews showed the
same result, it is fair to say that 2x learning efficacy seems to be the most important KPI for the
OCPs that were interviewed. The second most important KPI, based on results from the
structured as well as the semi-structured parts of the in-depth interviews, is 12% improved churn
rate, which was chosen 26% of the times it was available and regarded by 6 of the 16
respondents as important or the most important KPI. The least important KPI among the
three was 4.5x more problems solved which was chosen 16% of the times it was available and
regarded by two out of 16 as important whilst no respondent considered it to be the most
important. A possible reason for why the attribute level none was chosen the highest amount of
times relative to how many times it was available is because it only appeared, except from one
time in Q9, together with the cheapest price of USD 5,000. Hence, it is reasonable to assume
that this affected the none additional service attribute level to a high extent.
From the respondents’ product choices which are presented briefly in Table 7 and to the full
in Appendix III, five out of 16 chose none of the products in Q1 when the choice was
between three product combinations which all had a price of USD 10,000. This implies that,
although seven out of 24 respondents found the price to be too cheap, the price does seem to
have an impact on the product choice. Another interesting result found when comparing Q3
and Q5 is that; in the former question, seven respondents chose 2x learning efficacy, USD 7,500
and weekly performance report while in the latter question, eight respondents chose the same KPI
and service combination but for the price of USD 10,000. The choices are however dependent
on the different bundlings of attribute level combinations which limits the possibility to draw
conclusions without ceteris paribus. However, some conclusions are more obvious and can be
drawn. It is revealed from Q4 that six respondents chose 12% improved churn rate, USD 7,500
and dedicated solutions engineer, which is the same amount that chose 12% improved churn rate, USD
5,000 and no additional services. Based on this, it can be said that it lies in some organizations’
interest to pay extra for dedicated solutions engineer while it for other organizations is more
valuable to exclude the service and purchase AIaaS at a cheaper price instead. Further, as an
additional argument for why 4.5x more problems solved has revealed to be the least valuable KPI
among the three presented, the combination 4.5x more problems solved, USD 5,000 and no
52
additional services did not get chosen by any respondent while the same combination with the
KPI 2x learning efficacy got chosen ten times in Q6. Even when 4.5x more problems solved, USD
5,000 and no additional services was presented among twice as expensive alternatives, it did not
get chosen a significant amount of times. Contradicting to this are the responses for Q7 where
it is revealed that 4.5x more problems solved is as attractive as 12% improved churn rate. However, in
Q10 it is clear that 12% improved churn rate is considered more attractive than 4.5x more problems
solved as the other two attributes remained constant and 12% improved churn rate received nine
selections while 4.5x more problems solved received one selection. This emphasizes that the
context in which the products are presented may affect the product choices made by the
respondent group.
Continuing with the results from Q6 and the dominating ten responses for 2x learning efficacy,
USD 5,000 and no additional services, another four respondents chose 2x learning efficacy, USD
7,500 and executive briefings. This shows that even when cheaper alternatives exist, some
respondents are willing to pay extra for adding additional services to the AIaaS. When the
attribute level combination 2x learning efficacy, USD 5,000 and no additional services appeared a
second time, it received once again the double amount of selections as the second most
selected product in Q8. Further, in Q9, the same product combination as in Q8 received three
folded as many selections as the second most selected product which emphasizes that, at the
price of USD 5,000, an AIaaS with this KPI is highly desired. It is also revealed that there are
as many that chose 12% improved churn rate, USD 7,500 and no additional services as the ones that
chose 12% improved churn rate, USD 10,000 and weekly performance report. This emphasizes that
additional services should be offered as optional alternatives that can be added to the main
AIaaS.
5.2 Value Perception
This part of the discussion is divided into the three subject areas that were investigated
through in-depth interviews; KPIs, Price and Additional Services.
5.2.1 KPIs
Six out of 16 respondents expressed 12% improved churn rate as important or the most
important KPI to consider. Four out of these six respondents are actors within K12. To make
this product an appealing investment, three out of these four actors explicitly expressed that
the most desirable KPI is the one that can be quantified into value expressed as long term
revenue increase for the OCP. Based on these results and the consideration that K12 is a
conservative and slow market, a conclusion for this segment is that it demands guarantees in
the form of quantifiable KPIs to trust new products. It is also expressed that a price of USD
10,000 is not too high if the product would legitimate the OCP to charge a higher price per
learner. This could be done by delivering desirable KPIs for the curriculum partners and for
the market, for instance a significantly higher level of learner engagement. In contrast, it was
also revealed that, within the K12 segment, it is common that students are assigned to use a
certain platform as part of the curriculum which makes the KPI misleading in the K12
segment and better suited for higher education where mandatory participation in platforms are
not as common. An alternative to the KPI improved churn rate that could be more useful within
53
K12 is to measure daily, weekly or monthly engagement. Yet, it can be argued that improved churn rate
is particularly important as the AI-product and its corresponding value proposition is new and
somewhat unclear. By analyzing responses from in-depth interviews, one can conclude that
this KPI may be more relevant for a B2C OCP than a B2B or B2S OCP. This is because
churn rates usually are a concern for B2C actors where an individual learner seeks to be kept at
the platform. For a B2B or B2S OCP the retention would concern an entire school or
business, and not the individual learner. In addition to this, it is expressed that it could be
more appropriate to express the price based on user volume for a B2C actor rather than for a
B2B or B2S OCP. Factors to base the pricing on towards B2B or B2S actors can therefore be
expressed in relation to the size of the solution or the number of recommendations, which will
be considered in the proposed pricing model.
Seven out of 16 respondents expressed that 2x learning efficacy is important or the most
important KPI. Similar to what has been expressed about 12% improved churn rate, there is a
perception that it can be connected to overall improved revenue for the OCP. There is a
perception that if the KPI was to be proven, this would be invaluable for actors within
education. Hence, to increase the price, case studies that prove this KPI across segments
should be provided by the AIP itself or preferably by an independent institution. However,
criticism is also raised for this KPI as a respondent emphasizes that, specifically for their
organization, it does not matter how long the learning time is. In addition to this, the in-depth
interview results revealed that actors within K12, self-paced online learning and education
towards enterprise are not striving to make the learning faster but would rather want to
provide personalized recommendation tools to reinforce the learning. Another argument that
questions this KPI is that teachers within the K12 segment are dedicating a certain amount of
hours on a particular lesson or course regardless of how efficient the learning is. Hence, they
would rather have all students at the same level to be able to teach as many students as
possible during the same time. On the contrary, this relies on that all students are learning
efficiently and reaching the assessment criteria. Therefore, learning efficiency may be more
relevant for schools or teachers where there is a challenge in reaching the learning criteria for
all students. This argument is in line with that the types of KPIs used should be adjusted
depending on if the customer is a B2B or B2C actor while it must pay attention to the
organization’s specific interests. For instance, one of the respondents expressed that they work
with improving teacher’s efficacy and therefore would rather see KPIs corresponding to
teacher specific issues. However, one can argue that the increase in student efficacy is highly
related to the efficacy of teachers as their job is related to the hours spent per student.
Therefore, the authors believe the KPI 2x learning efficacy still to be highly relevant for the
considered pricing model.
For the KPI 4.5x more problems solved, there are two organizations that consider this important
and two organizations that do not consider it important. When analyzing the results, there is a
perception that 4.5x more problems solved is the least appreciated KPI among the three presented
as none of the 16 interviewed organizations did consider it the most important KPI. There is a
higher appreciation for this KPI within the higher education segment than in the K12 or
enterprise segment. The reason for this could be that K12 and enterprise learning is more
standardized and does not see the value in speeding through the learning process as they value
quality above quantity. However, the number of problems solved must not always be solely
54
related to quantity rather than quality, for instance; if a student is not at its optimal level of
difficulty, the student may spend unnecessary time for one specific question. Therefore, the
student does not solve enough problems in a given time frame, compared to if the level of
difficulty is optimized. Hence, the number of problems solved may still be an appropriate
KPI.
In addition to the discussion about the three KPIs, it was revealed during the interviews that
other possible KPIs are also important to discuss. Among these, a KPI that was mentioned
twice by different respondents was one that pays regard to OCP’s return of investment (ROI).
To be able to measure this, one would need to be able to derive increased revenue or profit to
the specific effect from the product by the AIP. Following discussions with the commissioner,
this can be done through A-B testing but is not investigated in this study due to time
constrictions. Also, to measure how well a learner is actually learning, rather than measuring,
for instance, the number of problems solved, the AIP can provide a KPI that involves
engagement connected to the learning outcomes achieved as well as how appropriate the
content is. However, to measure whether the student has reached the learning criteria or not is
up to each OCP rather than the AIP. Within higher education, it is possibly more common
and encouraged to have individual learning paths, which makes a KPI based on engagement
more relevant for this segment.
5.2.2 Price
The results from the PSM showed that an edtech actor’s willingness to pay for an AIaaS
product is a monthly fixed fee of USD 5,000 - USD 10,000. Seven out of all the 24 in-depth
interview respondents expressed this range to be very cheap or cheap and that organizations
“would be willing to pay anything if the product outcome is what it promises”. Two of these
organizations are however in the enterprise segment where prices generally tend to be less
restricted than for K12. Nevertheless, the two other respondents are within the segments K12
and higher education. Additionally, a respondent within the K12 segment has expressed this to
be an acceptable price but too cheap if the platform has “for instance 5,000,000 users”. Four
other organizations have however expressed this to be “pricey”, “above what the organization
can afford” or “only serve the upper end of the market”. Often, this depends on to which
segment the organization belongs where online platforms aimed towards the K12 segment
must adapt its prices to what the school can afford. In addition to this, the geography and the
difference in how financing works for each school in different countries also affect the
perception of the price span.
One interview revealed that there is a perception of larger organizations would be able to pay
USD 10,000 per month for this product while schools would find it expensive and that the
willingness to pay for this product is dependent on the size of the organization. However, no
clear correlation between the tier group and the perception of whether USD 10,000 is
expensive or cheap could be found from analyzing the in-depth interviews alone. Four out of
the 16 respondents that answered the structured in-depth interview expressed that USD
10,000 is a cheap or reasonable price while two respondents expressed that USD 10,000 is
pricey compared to the organization’s current goals. These former four respondents are within
different tiers and no conclusion can thereby be drawn, from in-depth interviews, between tier
group and if USD 10,000 is considered as cheap. The two respondents who expressed that
55
USD 10,000 is expensive belongs to the two middle tiers; Tier 3 and 4, which does not reveal
any particular connection between tier and if USD 10,000 is considered as expensive.
Regarding the way of pricing, there are two different alternatives that have been mentioned by
the interviewees and literature review; recurring payments or one-time fee. It can be argued
that the latter would make little sense as the personalization product would possibly need
maintenance periodically and require certain expertise that may not be present internally,
which is likely to be the case within the K12 segment where internal resources are strained. By
several actors, this also served as an argument for why they would like an external AI provider
to deliver this product.
The subscription based pricing can further be fixed or variable in which the latter allows the
price to depend on a certain factor. One of 16 respondents preferred a fixed fee over a
volume-based fee while five out of 16 rather would see a volume-based annual price for the
product. These respondents belong to different tier groups and no correlation between
volume-based fee preference and a specific tier group can thereby be identified, which makes
it more generalizable. As stated in section 4.3.1, the number of MAUs, which also was
revealed during the in-depth interviews, is the single most important factor that the variable
fee should consider. Another base for the variable subscription fee, specifically towards K12-
actors, would be to express the price as a proportion of the annual monetary amount per
student received by the state. However, one respondent revealed that there is a perception that
volume-based prices tend to become more expensive than fixed fees. On the other hand, the
interview results also showed that what is considered as the most favorable price setting
among fixed and variable depends on the current stage of the company. For that reason, for a
pricing model to be scalable, it should consider the growth stage of the OCP. A company in
an early stage with users in Tier 1 would benefit from paying per student rather than having a
fixed fee, which is supported from the Van Westendorp analysis. It is further argued that the
entrance price point should be as low as possible to be appealing for these companies. As the
company is growing and it reaches the tipping point of a large amount of users, the pricing
would level off to a more beneficial fixed monthly price instead. On the other hand, a
company that is scaling rapidly would benefit from a fixed fee which is more predictable and
do not escalate significantly when the number of MAUs increases. For companies that grow
beyond the highest tier group, a variable pricing based on price per student could be added to
the fixed price. This variable should be less than what the smaller OCPs in Tier 1 would pay
per learner as it is displayed in Table 5 that tier groups ≥ 2 are generally not willing to pay as
much per learner as actors in Tier 1. This is also related to the theory of Chao (2013) with
different tariffs depending on the tier.
In addition, the respondents are willing to not only base the price on an organization’s MAUs
but also express it as a sum of all students connected to the platform, and therefore consider a
school’s total amount of students rather than just the MAUs. However, if the price is set per
student, there are reasons for using price discrimination to be able to attract actors in more
economically developed countries as well as in less economically developed countries. It was
suggested that the price of a textbook within a geographic market can serve as an index for the
price discrimination, which is also a reason to not express the prices in absolute numbers.
Content creation costs is an aspect that an interviewee regards as an issue but does not
necessarily have to be related positively to personalized learning.
56
A common theme from interview responses is that the OCPs value the importance of having
guarantees for that this product actually works for their specific segment and company. The
need for this kind of warranty can be originated to the newness of AIaaS within edtech and
the limited research within the field. In addition to this, the in-depth interviews revealed that
there are different issues related to AIaaS which make OCPs cautious; this is further discussed
in 5.3.2. One way to counteract this product uncertainty is by providing a pricing that is based
on shared risks and rewards, referred to as royalty shares by one respondent, between the AI-
provider and the OCP. Similar to PBP, shared risks and rewards can be a way to increase trust
between the actors, which is important for these kinds of new offerings.
5.2.3 Additional Services
There is dispersion in which additional services that are the most important. However, weekly
performance reports was considered by four interviewees out of 16 as the most important
additional service and two respondents added that they would even appreciate daily reports. It
was expressed that it is desired to have additional services that change over time as the needs
and requirements of the OCPs change. Moreover, depending on what the OCP desires, some
of the services can be offered periodically, such as executive briefings and the dedicated solutions
engineer, while other services might be offered continuously, such as performance reports.
Furthermore, it was revealed that the need for additional services is dependent on what
segment the OCP belongs to as actors within higher education tend to have more resources
and believe that they might provide some of the services themselves. On the contrary, actors
within the K12 segment do not have the same possibilities due to the limit of resources. This
suggests that product bundling should be tailored towards specific actors depending on the
segment and the OCP’s level of internal resources. To have the possibility of adapting the
additional services towards the needs of the OCPs seem to be more important within the
enterprise segment since organizations have a wide range of needs and interests than schools
or universities. It is also expressed that KPI is the most critical factor for choosing the product
and additional services are secondary to that.
5.3 Discussion on Emerged Themes
This section discusses the most common focus areas connected to pricing discussed in the in-
depth interviews as well as from the survey answers.
5.3.1 Segments
What segment the respondent belongs to has showed to be important to consider when
examining the price setting for an AI-personalization product within education. It is revealed
that actors from different segments have different needs, restrictions, interests and budgets
which are highly relevant for the price determination of a new recommendation product for
education. Specifically for K12, the actors strive to use as little resources as possible to reach
the largest volume of students possible, which makes the K12 segment a hard market to
penetrate for providers of new innovations. This limited amount of resources derives from
price pressures in the market which makes it hard for organizations that have schools as their
customers to invest in new kinds of products. However, there is a will to drive the adoption
57
higher through AIaaS within the segment but it is hard for OCPs to drive this adoption
through price increases for schools. Such economical restrictions within a segment should for
that reason be considered and reflected in the pricing unless it is proven that the product
replaces learning material or teaching hours. In this scenario, when the effects of AIaaS can be
directly derived to decreases in costs, it is possible to free resources which in turn enable B2S
OCPs’ customers, schools, to pay for personalization through AI. In relation to the discussion
of a direct connection of KPIs to the revenue increase, it is as relevant to examine the
possibility of connecting the product to a potential decrease in current costs for schools. This
is however potentially more suitable for the K12 segment. Further, one respondent expressed
that, for their organization, it was not a very complicated task to create their personalization
tool. However, not all actors have enough data or MAUs to enable a personalization engine
and it is not possible for all actors within the edtech space to develop the product by
themselves, especially not for schools within the K12 segment where both resources and
capabilities are lacking. In line with this, several actors expressed that they would like an
external AI provider to deliver such a product.
Similar to relating the AIaaS pricing to the price of a textbook within the specific geographical
market, one could extend this to include other geographically specific conditions as well. For
instance, in some geographical locations, the decision to acquire software is made by the
individual schools themselves while in other geographies the decision is made at a higher level
and applies to several schools, for instance per district. A deeper way of customizing the price
towards OCPs is by considering how integrated their content is in the standard curriculum as
for example OCPs within programming are struggling more to sell their offering to schools
than what OCPs in mathematics might do. The former aims to target early adopters and it is
not unusual to sell such products through negotiation to K12. For this reason, it is reasonable
to assume that these actors also have different possibilities to pay for AIaaS. However, since
this factor is so specific, it will not be considered in the proposed pricing model as the authors’
aim to propose a generalized model.
In addition to considering the amount of MAUs, one may also consider high and low seasons
during a year and therefore price differently. As MAUs in K12 and higher education peaks
during spring and fall, other OCPs might have their peak season during the summer instead.
On the contrary, other organizations that provide testprep or enterprise education may peak
during the summer and winter holidays as people take time from the regular work for
education. For this reason, the consideration of peaks in seasons might be a good complement
to the number of MAUs used for pricing since the number of users might differ slightly from
the MAU average within a year or a semester. However, since the number of MAUs will be
considered in the proposed pricing model, the decision is made to not include peak seasons in
the model since it is not believed to have a major impact on the MAU average.
Another area that differs between different actors and segments is how much they are working
with retaining the end learners. As this is highly important for schools, it is less significant for
OCPs towards K12 to measure how well the students are retained since participation often is
mandatory and it is in the teachers’ interest to retain students. For that reason, as it is not the
OCP’s mission to make the students spend more time on their platform, five respondents
expressed that what they instead focus on is to retain the schools. An OCP within corporate
training also expressed that they work with retaining the enterprises rather than the end
58
learners, which are the enterprises’ employees. This highlights that the KPIs presented towards
B2S and B2B actors within the K12 and enterprise segment should capture how well their
customers will be retained through the AIaaS. However, expressing the AIaaS’s KPIs in terms
of OCPs’ customers’ interests is an indirect way of expressing the OCP’s direct interests. An
OCP for self-paced learning also expressed that they do not work actively with retaining their
users which is surprising as many of the actors within the segment are profit-based. For that
reason, by providing learning content that is more interactive, it is possible that such
organizations would experience a higher level of user retention and thereby increase in
revenue.
The enterprise segment varies from the K12 segment in the sense that it is not as restricted
when it comes to budget since OCP’s customers (the enterprises) might be willing to pay extra
for the personalization offering towards its employees. Additionally, within the corporate
training segment, it is expressed that personalization is difficult to develop internally as the
OCP has a wide range of content. For that reason, one can legitimate a higher price within the
enterprise segment since it would be a valuable feature that enterprises might be willing to
finance themselves.
5.3.2 Issues with AI in Education
From the in-depth interviews, some issues with purchasing AIaaS in general as well as for the
edtech sector were raised. These are similar to the ones with SaaS where the biggest concern is
that the actual value is not visible until after the purchase is made and the product has been
used. To counter this issue, as already mentioned, it is of great importance to express the value
in terms of relevant long term business KPIs. Also, there seem to be a division of opinions
regarding how far from the market reality AIaaS is. Some respondents expressed that this kind
of personalization product is a costly and an unnecessary investment which in turn requires a
lot of resources to produce large enough data to analyze. In contrast, others pointed out that
personalization tools are vital to stay relevant in today’s digitalized society. Moreover, it is
expressed that AIaaS for personalization is more applicable to higher education than to K12
since the latter segment is not as digitalized. However, ever younger pupils are using laptops,
tablets and smartphones which puts pressure on schools to adapt and invest in new
technologies (Manning, 2017).
Some actors are worried about losing control of the content by having an external AIP.
However, the buying party is still the owner of its content and the content provided to the end
users is the same as without the personalization product with the exception that the learning
path may be different for the learners. For this reason, the concern regarding control should
not have a major impact on the pricing. Another issue that may however be more significant is
the lack of standards to express and interpret adaptive learning content. This is particularly for
learning institutions, where it is important that adaptive learning and assessments are marked
as fairly as non-adaptive content. In addition, current standards such as COPPA might have to
adjust its current criteria in order to enable personalized education while not excluding
involved actors from complying with the standards.
Lastly, there is an issue of depriving the social aspects of classroom learning by introducing
digital personalization tools in education. In order to minimize this impact, the personalization
can be restricted to only provide content from a certain part of the curriculum. In this way, the
59
students might be exposed to slightly different problems but the subject area studied will
remain the same for all learners to ensure group cohesion and encourage classroom
discussions.
5.4 Proposal of a Pricing Model for AIaaS
The model presented in Figure 7 is a pricing model which can be applied to an AI-
organization delivering products based on AI and ML, where AI is offered aaS. The model is a
result of theory bits since it is based upon the model presented by Lehmann & Buxmann
(2009) where a specific variable is chosen within each parameter to fit an AIaaS provider. In
addition to this theory, the model also considers the SaaS pricing identified by Chao (2013).
Hence, the model combines parts of existing pricing models found in the performed literature
review as well as the primary results gathered through the performed survey and in-depth
interviews to fulfill the needs of an AI-organization. The model consists of free
implementation to create a lock-in effect for the customer as this was shown through the
literature review to be of high importance to strengthen the network effects of a software
provider. This is then combined with a monthly subscription fee based on the number of
MAUs. The model is also related to the customer’s perceived value through the price bundling
parameter, where additional services can be added on to the fixed fee for MAUs.
The model does not include strategy as one of the parameters as the results in the literature
study showed that a strategy can contain several models - it is therefore irrational if the model
itself contained a strategy. Hence, the strategy is displayed as an umbrella unit at a level above
the parameters of the pricing model.
Figure 7. Proposed pricing model for AI personalization products
60
The strategy an organization should decide upon is the first step in the proposed pricing
model. There are two paths to choose in between where skimming was present as a third
choice in the model presented by Lehmann & Buxmann (2009). Similar to a software offering,
it is reasonable to assume that the initial price should not be set too high as it is more difficult
to win dominant market share. With an initial low price, an organization can grow quickly as
more customers adopt to the product. With existing customers who become dependent on the
software, the price can be increased step-wise to maximize revenue and profits, this is also
supported by respondents in the in-depth interviews. The skimming strategy has therefore
been excluded in this model as the interview results have shown it to be difficult to carry
through and be successful with such a strategy within the edtech space, based on the
experience from the interviewees. Furthermore, literature suggests that freemium and pricing
penetration are more suitable for software tools (Lehmann & Buxmann, 2009) as such
strategies seek to create a lock-in effect, which is important for organizations benefiting from
network effects. In order to create strong network effects, the provider must follow a strategy
which allows it to lock in as many suitable users as possible which will create a CLV (Farris, et
al., 2010) high enough for the organization to possibly provide free or discounted integration.
This also promotes the choice of excluding the skimming strategy.
For the parameters, specific variables have been chosen to suit companies delivering AIaaS
products. The formation of price is value based as cost based does not make any sense for AI
companies considering the CLV. The degree of interaction can be unilateral or interactive at
an initial point; the more unilateral and more automized the price determination becomes, the
less resources are needed to be spent by the provider. Nevertheless, the degree of interaction
may be needed to be more interactive at the initial growth stage of an organization along with
the level of market acceptance of the product. When providing a new product, the degree of
interaction might also be higher for the vendor to have a dialogue with the buyer in order to
agree upon a price to secure the customer and create a lock-in effect. As a product matures
and an organization grows, the degree of interaction will decrease to eventually become
entirely unilateral. This should also be what an AIP aims for.
The structure of payment flow is primarily based on recurring payments, either yearly,
quarterly or monthly depending on the buyer’s wish. A yearly signup could come with a small
discount compared to monthly signups. Integration is for free as the organization seeks to
create a lock-in effect for the users where a free integration perks the buyer to connect the AI
product into its existing system and discover the need of it; therefore becoming locked in. This
is more important for startups than for more mature companies as AIPs are highly dependent
on vast amounts of data to create strong network effects. Therefore, startups should provide
free integration in order to collect a sufficient amount of data. As an organization becomes
more mature, the integration may be charged for as the organization already possesses the data
needed to train its AI product. The recurring fee can be either fixed, variable or a mix of both.
A variable price will depend on the exact number of MAUs whilst the fixed price will depend
on a range of MAUs. The purpose of this is that there is a limit where the variable price
becomes too expensive and a fixed price is more beneficial. Moreover, one can create a
combination of the two where the OCP pays a fixed price up to a certain number of MAUs
and if the OCP exceeds this tier, a variable price is paid for the number of MAUs exceeding
the ones included in the fixed fee.
61
The assessment base, in the commissioner’s case, is usage dependent based on the number of
MAUs whilst for other AI-recommendation companies it can be based on the number of
recommendations provided by the algorithm. The in-depth interviews have shown that the
recommendation base could be more appealing for some OCPs since it may be more relevant
for their business. If choosing the usage dependent assessment base, the AIP must then select
a linear pricing model, 2PT or a 3PT (Chao, 2013). From the survey and in-depth interviews,
results have shown that the level of tariff depends on the number of MAUs. OCPs with the
lowest number of users should be provided a linear model whilst the OCPs with the largest
number of MAUs should be allowed for a 3PT. The assessment base can also be usage
independent, which is performance based, where specific KPIs can be provided to measure
the improved performance. The KPIs must alter depending on which segment the customer
belongs to or whether it is a B2C or B2B actor. Such a KPI can be ROI, increased revenue,
customers may want to perform an A-B testing to ensure the improved performance, use
cases will increase the assurance for future customers as the product becomes proved and
market accepted. For a new product, A-B testing may however be the most suitable way to
prove the increase in performance. An extension of this logic would be to base the pricing on
a percentage of the customer’s revenue, or to benchmark it to other comparable expenses.
What this percentage would be has however not been investigated in this research as the
respondents have not shared information on its revenue.
Price discrimination may, in this case, be one of the most important factors as it must be
present in order to reach a customer base which is as wide as possible. Price discrimination
will be present to a second and third degree. For the second discrimination degree is built on
the principle of self-selection for the product combination. The second degree also involves
quantity-based price discrimination where the number of MAU’s can be used to place the
buyers in different tier groups. The number of MAU’s relates to the buyer’s revenue,
profitability and hence its ability to pay. Therefore, organizations with a larger number of
MAUs will be placed in a tier group of a higher price per user. This also includes time
discrimination; this is where actors pay different prices depending on the point in time the
product is purchased. For instance, early buyers may receive a lower price as the AIP is in
greater need of data compared to when the company is more mature and does not have to be
as selective; such buyers will pay a higher price. The third degree of discrimination involves
market discrimination such as geographies and segments. One measurement of determining
geographical discrimination can be through a textbook index where the price of a textbook
acts as a base for the price. This would however mostly be relevant for the K12 segment and
possibly higher education, as prices of textbooks vary significantly depending on which
country the book is bought in.
Price bundling is based on customized bundling which is individual for each buyer. The price
bundling is based on the software itself, integration, maintenance and additional services
which makes up the product. The degree of integration is complimentary as the products are
independent of one another. The price level of the product bundling is then subadditive as the
buyer should benefit from combining the products rather than purchasing them individually.
62
5.5 Sustainability
This study investigated pricing models and its relation to value for AIaaS companies within the
edtech space. In order for the proposed solution to be valid, one must consider its
sustainability effects and implications. The concept of sustainability was in 1987 defined by the
UN to consist of three pillars; economic, social and environmental sustainability (UN, 1987).
Within the economic sustainability pillar there are several aspects to be discussed; one must
consider the sustainability of the provider as well as the buyer and the end user; the learner.
Beginning with the provider, which in this case is Sana Labs but may be any AIaaS provider,
the solution has been investigated to meet profitable revenue targets based on the
development costs of the product. The individuals involved in the organization must be
compensated for the time they put in, investors are expecting to receive a return on their
capital and the organization itself seeks to contribute to a growing regional economic
development, which is measured through the contribution to the GDP- a goal generally
accepted by the public in order to keep economic sustainability and has been the most
important policy goal for the five most recent decades (Moldan, et al., 2012). Sustainable
development also considers the usage of non-renewable resources in a manner that does not
eliminate easy access to them for future generations (Moldan, et al., 2012). Since a pricing
model only is an idea rather than a physical product it is up to each and every organization to
implement a policy along with the pricing model that meet the global economic sustainability
goals. Sana’s product which seeks to make adaptive learning available for everyone is by itself
economically sustainable as education should foster more innovation, job creations and
economic growth (Bughin, et al., 2017). UNESCO states that 24.4 million primary school
teachers will need to be recruited and trained globally for the world to reach total access to
primary education by 2030 (Bughin, et al., 2017). In addition to this, another 44.4 million
teachers will need to be recruited for secondary schools (Bughin, et al., 2017). AI products
could contribute in solving the demand of teachers where, for instance, Sana’s AI product
personalizes the education which saves the teacher time and effort. At a wider scale, online
education in general may also be a possible contributor to the solution in meeting the rising
demand of teachers where students will be able to, to a certain extent, learn through AI rather
than a physical teacher and be assessed by a computer rather than a teacher (Bughin, et al.,
2017).
The social pillar of sustainability is most likely the one of highest significance for the human
survival (Moldan, et al., 2012). From previous literature, it is not entirely clear what is meant by
social sustainability - whether it relates to health or to the long term survival of human
mankind (Moldan, et al., 2012). Within this study, social sustainability relates to the people
which are affected by the product. There are two sides to this aspect; the organizations which
seek to implement the pricing model for its AIaaS product and the learners within the edtech
space that are the final consumers being affected by the use of Sana’s AI product. As the goal
of the product essentially is to make adaptive learning available for everyone, the purpose of
the product’s existence meets the sustainability goal of maintaining good health; as this can be
spread through education. In a report by the consulting firm McKinsey & Co (2017) it is
stated that many developed countries suffer from mismatches between education and
employment as the educations to fail meet the demand of the employers and educated
individuals also feel that the labor market fails to match the education with the employment.
63
At a more general level than the specific AI that Sana works with, McKinsey believes that AI
in education will help to minimize these mismatches between academia and the labor market.
AI in education will also help individuals define their level of competency; improve learning
outcome and the quality of education which then can be used to match for an applicable
employer. AI in education also aims to reduce teachers’ administrative tasks and focus on what
teachers are meant to do; teach (Bughin, et al., 2017). This will further contribute to social
sustainability as students and teachers achieve a better working environment, a factor
important for the health and therefore social sustainability.
The third pillar of sustainability, the environmental one, was formerly a part of the social and
economic which sought to both include environmental sustainability (Moldan, et al., 2012). A
definition of environmental sustainability introduced by Goodland (1995) is that it “seeks to
improve human welfare by protecting the sources of raw materials used for human needs and
ensuring that the sinks for human wastes are not exceeded, in order to prevent harm to
humans” (Moldan, et al., 2012, p.6). With relation to pricing models of AIaaS products, the
major resource used is the raw materials needed for computers used by the AI researchers and
developers. This is an aspect every organization using a computer needs to consider today, as
materials such as metals and plastics are processed and used. One important action is to
recycle or make up for the environmental footprint the organization causes which can be done
by planting trees which emit the amount of CO2 used by the processes of the organization.
Another aspect for an AI provider is the use of data storage. The storage of data, specifically
remote storage which is stored in clouds or platforms, requires vast amounts of energy to
continuously be kept running. In the report The Cloud Begins with Coal (Mills, 2013) it is
explained that data traffic used to be the data which flows to and from the user and the data
storage. However, nowadays data traffic is to a larger extent associated with intra-data-center
traffic due to the increase use of IT services such as remote data storage and real time
processing. In the same report, Mills (2013) also shows that nearly seven zettabytes (1ZB =
10^21 bytes or 1Bn Terabytes) per year was needed in 2016 for global data center traffic. In
comparison to a hard drive in a laptop, in which 1 Watt per user is needed to access one photo
twice a day or a hundred, cloud storage can consume 10 times more energy than storing and
accessing it through a laptop. In cloud storage the energy consumption increases with the
amount of data, which is not the case for a hard drive (Mills, 2013). Another researcher states
that data storage will be one of the largest energy consumers and will stand for one fifth of the
global energy consumption by 2025 (Andrae, 2017). This is something that organizations
providing AI solutions need to consider with great seriousness as their products depend on
vast amounts of data, which in Sana’s case is stored in a remote server through a cloud service
(Sana, 2018).
5.6 Ethical Implications
As for sustainability, ethical implications in this study can be divided into and applied to the
area of AI and its relation to education as well as the subject of pricing. Below, a discussion of
ethics related to AI and education will be followed on ethics within pricing.
For AI as a subject, there is a lot of discussion within ethics. The first distinction that needs to
be made when discussing machine ethics is the difference between developing ethics for a
64
machine and developing ethics for the humans who use the machine. In the first case, the
humans who develop the machine, or more specifically AI-product, the machine must learn
ethical principles or have a process of discovering and solving unethical dilemmas through
their own decision making. The second case involves the insurance of that a machine is not
used by a human in an unethical manner. This is however related to the decision making of the
human rather than the machine’s. (Anderson & Leigh Anderson, 2011)
Another ethical implication related specifically to AI in education is who owns the data on
students; who has access to it, what is it used for and by whom (Bughin, et al., 2017). In
Europe, in which the commissioner operates, the General Data Protection Regulation (EU,
2016) is particularly important when collecting data. For organizations operating within the AI
segment, there exists guidelines on automated decision making (Jacobs & Ritzer, 2017). In the
article on how AI is influenced by the GDPR Jacobs and Ritzer (2017) emphasize that the
regulation states decision making “solely based on automated processing”, meaning there is a
complete absence of human involved decision making. These aspects must be considered by
the organizations sharing and collecting the data; Sana in particular but also its customers who
have primary access to the student data.
Ethical implications related to marketing and pricing, are generally discussed within the field of
targeted marketing. Targeted marketing has almost become synonymous with competitive
strategies, where ethics commonly is discussed for harmful products such as alcohol and
cigarettes. The opposite of this is price discrimination - where some potential buyers are
excluded from the marketing as certain groups are denied access to a specific price as they
belong to a specific customer segment (Cui & Choudhury, 2003). In this study, the use of
price discrimination is one of the main features in the pricing model. The reason for not
providing the same price for all segments is that the vendor, Sana in this case, would not
receive enough profits related to the resources and the increase in profit and efficiency
estimated to be improved for the buyer. In addition to this, an AI company that benefits from
network effects needs the maximum amount of users in order to reach maximum value
creation through the enhancement of network effects. As larger platforms have access to a
larger amount of MAUs, the AIP would target larger platforms to maximize its data gathering
and therefore network effects. Due to this, prices would also be adjusted for larger platforms,
which are likely to have higher revenue and profit, making the prices above the range of what
smaller platforms are able to pay. Therefore, by allowing for price discrimination the smaller
platforms, such as startups, are not excluded by high prices, but can afford to integrate the AI
product into their platform and benefit from price discrimination.
5.7 Summary of Findings
The results and analysis of the Van Westendorp PSM showed that respondents considered
USD 5,000 to USD 10,000 as an acceptable price range as a fixed fee per month for the AIaaS
personalization product. When expressing the willingness to pay per learner instead, the
accepted price range turned out to lie in between USD 0.01 and USD 1 per learner per month,
with different spans in between these numbers depending on the end of the range. When this
fixed monthly fee was utilized as an area of discussion in the in-depth interviews, it was
65
discovered that seven out of all the 24 in-depth interview respondents expressed this range to
be very cheap or cheap.
Further, it was discovered that the willingness to pay is dependent on several factors in which
the OCP’s number of MAUs and segment has showed to be the most important. However, it
is difficult to relate all the tier groups to a certain monthly fee entirely based on the collected
results. Nevertheless, a general suggestion is to base the price on a combination of a fixed and
a variable fee for larger tier groups while OCP’s with fewer MAUs only are suggested a
variable fee based on a linear model. In addition to this, the MAUs that exceed a certain tier
group should be priced according to a variable fee above the fixed monthly fee, where the
initial fixed monthly fee is higher for tier groups with more MAUs. Moreover, additional
factors to consider were categorized under price discrimination as it was revealed that the
price could depend on geography, a textbook index as well as the maturity stage of the
company. Other findings to consider for a price model are high or low season, if the OCP is a
B2B or B2C actor and if the product bundling should be tailored or not.
In the structured part of the interviews, the price range from Van Westendorp was combined
with two other attributes; KPI and Additional services in which three to four attribute levels were
defined for each attribute. The respondents’ evaluation of these attribute levels resulted in that
2x learning efficacy was considered as the most important KPI, supported both from the
structured as well as semi-structured parts of the interviews while 4.5x more problems solved was
criticized for measuring quantity rather than quality and therefore considered to be the least
valuable. Other types of KPIs that emerged as a result of the empirical data gathering are KPIs
that are connected to learners’ engagement as well as ones that relate to the OCP’s ROI,
increase in revenue or decrease in costs. In general, the most important finding connected to
KPIs is that these should reflect the OCP’s long term business improvement expressed in
monetary value. Further, weekly performance reports appeared to be the most appealing of the
three presented additional services. However, the results emphasize the importance of being
able to choose to include additional services or not.
Lastly, as a result of the discussion of secondary and primary sources is the proposed pricing
model which visualizes the main decisions to be made as part of determining a price model for
an AIaaS product.
66
6. Conclusion
This section presents the main findings from the research as well as the answers to the research questions. The
answers to the sub- and the main research question hence fulfill the purpose of the study. The conclusion also
includes the contribution within the marketing field of pricing, and the limitations of the study are presented.
The chapter is finalized by a recommendation of future research within pricing models for AI products.
6.1 Main findings
The purpose of this study was to investigate the possible pricing models for an AIaaS product
delimited to the edtech industry. It also sought to create a suitable pricing model for AI
companies offering a new personalization technology based on AI and deep learning
algorithms. To fulfill the purpose of the study, one main research question and two sub-
research questions sought to be answered. The questions and answers to the sub-research
questions are presented below.
1. What is the perceived value delivered by AIaaS and how can it be determined and
mirrored to price?
2. What factors should an AIaaS providing organization consider when determining a
pricing model?
3. What are the main implications of implementing AIaaS within the edtech industry?
This research has shown that the perceived value depends on which segment the customer
belongs to. Specific to this study and the edtech industry, the majority of the respondents
belong to the K12 segment where KPIs such as learning efficacy are highly valued as it enables
more efficient teaching for teachers. For the other segments, it is difficult to draw a conclusion
based on the results as the respondents within these segments were few. In general, the study
has shown that the perceived value is what a potential customer can interpret as increased
performance, which can be measured and determined through KPIs such as decreased costs,
increase in revenue or improved churn rate. This should thereby act as the assessment base of
the price. In addition to this, the perceived value has shown to vary with the interpretation of
the new product as different organizations are at various levels of maturity of acceptance for a
new AIaaS product; some perceive very low value whilst others find it invaluable. The
perceived value becomes highly related to the willingness to pay, which affects the price.
Therefore, mirroring the perceived value to a price must be altered depending on variables
such as segment, size, maturity or geography.
Factors that an AIaaS providing company should consider when determining a pricing model
is the formation of price, structure of payment flow, assessment base, price discrimination as
well as price bundling. Specific to this research, the number of MAUs, size of buying
organization as well as its stage of maturity have shown to be particularly important. An AIP
should pay attention to the assessment base, where a volume-based price rather than a fixed
price justifies the corresponding value and price for the buyer. This can depend on the
number of MAUs, recommendations or another variable that is suitable for the AIP’s
67
customer. In relation to this, the AIP must consider the maturity of the purchasing
organization as the variable price will vary for different tier groups.
The main implications for implementing AIaaS are the general skepticism towards a new
product where one pays for the product without a guarantee of performance. Specifically for
AI in education, there is limited research in how personalized education will affect the social
interaction is classrooms as well as the level of workload for the teacher. Also, standard
procedures for how to deal with adaptive learning are not yet established. In addition to this,
there exists a general perception that more content needs to be created in order to achieve
adaptive learning which means more work and higher creation costs.
Together, the answers to the sub-research questions seek to answer the main research question
which is presented again below.
“What pricing model should an AI-company have for its B2B personalization
product?”
For a B2B personalization product, an AI company should have a pricing model which
corresponds to the value delivered by the product based on the pricing model presented in 5.4.
The pricing model should account for the formation of price, structure of payment flow,
assessment base, price discrimination and price bundling in order to be generalizable for any
AIP. In order to guide the AIP, variables are presented for each parameter which makes the
pricing model adaptable for AIPs outside the edtech industry.
6.2 Contribution
Within industrial management, this study has contributed to the field of marketing research of
pricing models for new AIaaS products. Based on data collected within the edtech space, the
research has contributed through a proposal of a scalable and generalizable pricing model for
AIaaS products. Therefore, this study has contributed to research in the intersection of AI and
pricing; pricing models for AIaaS.
6.3 Limitations
The study was limited to the number of interviews which depended on the primary email
addresses that were accessible to the authors, which in turn affected the response rate and
possibly the results. Further, the study was limited to the people who were interviewed and
respondents in the survey as the ones who did respond may not have possessed the
knowledge needed to entirely cover what was required for the survey or interview.
With limited previous research in pricing for AI products, the study became limited with the
conclusions that could be drawn from research in related areas such as software pricing,
although AI pricing can be argued to be closely related to SaaS due to its similarities such as
network effects.
68
This research was also limited to a time frame of 17 weeks, which likely affected the number
of responses as more reminders and more interviews could have been held with a larger time
span, also a possible effect on the results.
6.4 Future Research
To continue the development and application of AIaaS, pricing models must be further
researched. In addition to this, pricing models must be tested and proved in practice by
organizations in order to ensure the function of the model.
To extend the proposed model and make it more of a model for determining the actual price
(and not a model of how one determines the price), a more extensive investigation of a linear,
2PT and 3PT tariff should be carried out. Future researchers should examine the relationship
for the limit of variable to a fixed price in order to reach a generalizable model for AIaaS
providers, which could be done through pricing optimization. This could rather serve as input
for adjusting the parameters and variables used in the proposed pricing model. Although this
cannot be proven until tested, such pricing optimization may be very market specific and
difficult to create for all AIaaS providers.
This research aimed to reveal customers’ value perception through qualitative data gathering
conducted through structured as well as semi structured interviews. However, to obtain a
more significant statistical perception of customers’ value perception in which preferred
attribute levels are put in numbers, a recommendation for future research that aims to quantify
value is to conduct a full conjoint analysis. With the results from an analysis which contains
the five complete stages of the conjoint analysis, one will be able to statistically determine what
the population perceives as the most attractive product attributes in relation to the presented
price levels. Results from such research does however become very market specific.
Further research within the field specifically for edtech might discover other or completely
new pricing models which potentially could be more suitable for pricing an AIaaS product.
Moreover, similar research to the one in this study but with other KPIs, price levels and
additional services could be conducted to test if the willingness to pay or the price model
could be formulated differently when other areas or attribute levels are considered.
Furthermore, this research was delimited to examine willingness to pay as well as value
perception of AIP’s customers within the edtech industry. Similar research can be done within
other industries that would benefit from AIaaS for personalization as well. An example of an
industry that is changing rapidly as a result of digitalization is medical technology (medtech),
which would be an interesting field to investigate for evaluation of the proposed pricing
model. In addition, an investigation of value perception and willingness to pay in other
industries would serve as a tool to test if the model has a broader field of application, hence
testing the generalizability of the proposed pricing model for AIaaS. As this kind of testing of
the model’s usefulness in other industries was not within the scope of this study, the identified
gap in the field of pricing models for new AI products is not completely closed. For that reason,
future research within other industries to test the pricing model would contribute to closing
the identified gap in the literature.
69
References
Accenture, 2018. Making it personal. Accenture Interactive.
Anderson, C., 2009. Free: The future of a radical price. 1st ed. London: Random House.
Anderson, M. & Anderson Leigh, S., 2011. Machine Ethics. New York: Cambridge University Press.
Andrae, A., 2017. Total Consumer Power Consumption Forecast. Huawei Technologies.
Baker, R., 2011. Implementing Value Pricing: A Radical Business Model for Professional Firms. Hoboken, New Jersey: John Wiley & Sons.
Blomkvist, P. & Hallin, A., 2014. Method for Technology Students: Degree Project Using the 4-Phase Model. First Edition ed. Lund: Studentlitteratur.
Brenner, R., 2016. Spreadsheet Models for Managers. Boston: Fajada Butte Press.
Bughin, J., Hazan, E., Ramaswamy, S., Chui, M., Allas, T., Dahlström, P., Henke, N. & Trench, M., 2017. Artificial intelligence the next digital frontier?. McKinsey Global Institute.
Chao, Y., 2013. Strategic Effects of Three-Part Tariffs Under Oligopoly. International Economic Review, 54(3), pp.977-1015.
Collis, J. & Hussey, R., 2013. Business research: A practical guide for undergraduate and postgraduate students. 4th ed. Basingstoke: Palgrave macmillan.
Cui, G. & Choudhury, P., 2003. Consumer Interests and the Ethical Implications of Marketing: A Contingency Framework. The Journal of Consumer Affairs, 37(2)., pp. 364-387
Deneckere, R., & Kovenock, D., 1992. Price Leadership. The Review of Economic Studies, 59, pp. 143-162
Dholakia, M., 2016. A quick guide to value-based pricing. Harvard Business Review.
Dixit, et al., 2008. A taxonomy of information technology-enhanced pricing strategies. Journal of Business Research, 61(4), pp. 275-283
Dodds, W., Monroe, K. & Grewal, D., 1991. Effects of Price, Brand, and Store Information on Buyers' Product Evaluations. Journal of Marketing Research, 28(3)., pp. 307-3019
Dolgui, A. & Proth, J.-M., 2010. Pricing strategies and models. Annual Reviews in Control, 34(1).
Doyle, P., 2000. Value Based Marketing: Marketing strategies for corporate growth and shareholder value. Wiley.
Ertel, W., 2009. Introduction to Artificial Intelligence. Wiesbaden: Springer.
Esomar, 2015. A New Approach to Study Consumer Perception of Price. RW Connect. [Online] Available at: https://rwconnect.esomar.org/a-new-approach-to-study-consumer-perception-of-price/ [Accessed 8 May 2018]
EU, 2016. REGULATION (EU) 2016/679 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation). Official Journal of the European Union.
Farris, P., Bendle, N., Pfeifer, P. & Reibstein, D., 2010. Marketing Metrics: The Definitive Guide to Measuring Marketing Performance. New Jersey: Pearson Education
Fink, A. & Kosecoff, J., 1998. How to Conduct Surveys: A Step by Step Guide. Thousand Oaks, California: Sage Publications
Fortune, 2012. Amazon’s Recommendation Secret. [Online] Available at: http://fortune.com/2012/07/30/amazons-recommendation-secret/ [Accessed 15 May 2018]
Gallo, A., 2015. A Refresher on Price Elasticity. Harvard Business Review, 21 August.
Green, P. E. & Srinivasan, V., 1978. Conjoint Analysis in Consumer Research: Issues and Outlook. Journal of Consumer Research, 5(2), pp. 103-123.
Green, P. E. & Srinivasan, V., 1990. Conjoint Analysis in Marketing: New Developments with Implications for Research and Practice. Journal of Marketing, 54(4), pp. 3-19.
Greener, S., 2008. Business research methods. 1st ed. London: BookBoon.
Guest, G., MacQueen, K. & Namey, E., 2012. Introduction to Applied Thematic Analysis. Thousand Oaks: SAGE Publications Inc.
Hager, MA., Wilson, S., Pollak, TH., Rooney, PR., 2003. Response rates for mail surveys of nonprofit organizations: a review and empirical test. Nonprofit Voluntary Sector Q, 32, pp. 252–267.
Harmon, R., Raffo, D. & Faulk, S., 2004. Value-Based Pricing For New Software Products: Strategy Insights for Developers. Portland International Conference Proceedings.
Hinterhuber, A., 2008. Customer value-based pricing strategies: why companies resist. Journal of Business Strategy, 29(4), pp. 41-50.
Holton, J., et al., 2010. The Grounded Theory Review: An international journal. The Grounded Theory Review, 9(2), pp. 1-59.
Ingenbleek, P., Debruyne, M., Frambach, R. T. & Verhallen, T. M. M., 2003. Successful New Product Pricing Practices: A Contingency Approach. Marketing Letters, 14(4), pp. 289-305.
Ishibashi, I., 2008. Collusive price leadership with capacity constraints. International Journal of Industrial Organization, 26(3), pp. 704-715
Iveroth, et al., 2013. How to differentiate by price: Proposal for a five dimensional model. European Management Journal, 31(2), pp. 109-123
Jacobs, S. & Ritzer, C., 2017. Data Privacy: AI and the GDPR. Norton Rose Fullbright, 2nd of
November 2017.
Kienzler, M. & Kowalkowski, C., 2017. Pricing strategy: A review of 22 years of marketing
research. Journal of Business Research, 78, pp. 101-110.
Laatikainen, G. & Ojala, A., 2014. SaaS architecture and pricing models. Jyväskylä: Department of Computer Science and Information Systems, University of Jyväskylä.
LeCun, Y., Bengio, Y. & Hinton, G., 2015. Deep learning. Nature, 521(7553), pp. 436-444.
Lehmann, S. & Buxmann, P., 2009. Pricing Strategies of Software Vendors. Darmstadt: Business & Information Systems Engineering, 1:452.
Lipovetsky, S., 2006. Van Westendorp price sensitivity in statistical modeling. International Journal of Operations and Quantitative Management, 12(2), pp.141.
Maital, S., 1994. Executive economics: ten tools for business decision makers. 1st ed. New York: Simon and Schuster.
Manning, E., 2017. Out with the old school? The rise of ed tech in the classroom. The Guardian. 1st of August 2017.
Marr, B., 2016. What Is The Difference Between Artificial Intelligence And Machine Learning?. Forbes, 6th of December 2016.
Mills, M., 2013. The Cloud Begins with Coal: Big Data, Big Networks, Big Infrastructure, and Big Power. Digital Power Group.
Mintzberg, H., Lampel, J., Quinn, J. & Ghoshal, S., 2003. The Strategy Process - Concepts, Context, Cases. Essex: Pearson.
Moldan, B., Janouskova, S. & Hak, T., 2012. How to understand and measure environmental sustainability: Indicators and targets. Ecological Indicators, 17, pp. 4-13.
Nagle, T. T., Hogan, J. & Zale, J., 2013. The Strategy and Tactics of Pricing A Guide to Growing More Profitability. Fifth ed. Harlow: Pearson Education Limited.
Nulty, D., 2008. The adequacy of response rates to online and paper surveys: what can be done? Assessment & Evaluation in Higher Education, 33(3), pp.301-314
Orme, B. K., 2010. Getting started with conjoint analysis: strategies for product design and pricing research. 2nd red. Madison: Research Publishers LLC.
O'Sullivan, A. & Sheffrin, S. M., 2003. Economics: Principles in Action. 1st ed. Upper Saddle River: Pearson Prentice Hall.
Parker, G. G., Van Alstyne, M. W. & Choudary, S. P., 2016. Platform Revolution: How Networked Markets Are Transforming the Economy and How to Make Them Work for You. 1st ed. New York: WW Norton & Company.
Pettey, C., 2015. Moving to a software subscription model. Gartner. [Online] Available at: https://www.gartner.com/smarterwithgartner/moving-to-a-software-subscription-model/ [Accessed 8 May 2018]
Planing, P., 2014. Innovation acceptance: the case of advanced driver-assistance systems. 1st ed. Wiesbaden: Springer Gabler.
Poyar, K., 2018. The Key to SaaS Pricing, Data Driven Sales - How the best B2B sales leaders use data to grow faster, pp. 6-17. Berkley, California: Clearbit.
Rachev, S., Höchstötter, M., Fabozzi, F. & Focardi, S., 2011. Introduction to Regression Analysis. John Wiley & Sons.
Rich, E. & Knight, K., 1991. Artificial Intelligence. New York: McGraw-Hill.
Richey, R. C., Silber , K. H. & Ely, D. P., 2008. Reflections on the 2008 AECT Definitions of the Field. TechTrends, 52(1), pp. 24-25.
Rouse, M., 2017. Artificial Intelligence as a Service (AIaaS). TechTarget. [Online] Available at: https://searchenterpriseai.techtarget.com/definition/Artificial-Intelligence-as-a-Service-AIaaS [Accessed 8 May 2018]
Ryan, M., 1996. Using Consumer PReferences in Health Care Decision Making - The application of conjoint analysis. London: White Crescent Press.
Sana Labs AB, 2018. Interviews with CEO Joel Hellermark. Stockholm. (January - April 2018)
Sánchez-Fernández, R. & Iniesta-Bonillo M., 2007. The concept of perceived value: a systematic review of the research. University of Almería, Spain: Marketing Theory, 7(4).
Sawtooth Software, 2017. The CBC System for Choice-Based Conjoint Analysis. Sawtooth Software Technical Paper Series, 9. Orem, Utah.
Silverstein, D., Samuel, P. & Decarlo, N., 2008. 51. Conjoint Analysis. in: The Innovator's ToolKit: 50 Techniques for Predictable and Sustainable Organic Growth. Hoboken: John Wiley & Sons, pp. 312-317.
Simon, H., Butscher, S. A. & Sebastian, K.-H., 2003. Better pricing processes for higher profits. Business Strategy Review, 14(2), pp. 63-67.
Shapiro, B., 2002. Is Performance-Based Pricing the Right Price for You? Business Research for Business Leaders.
Shapira, B., Rokach, L. & Freilikhman, S., 2013. Facebook single and cross domain data for recommendation systems. User Model User-Adap Inter, 23(2), pp. 211-247.
Spencer, R. W., 2009. How Do Climate Models Work?. [Online] Available at: http://www.drroyspencer.com/2009/07/how-do-climate-models-work/ [Accessed 16 February 2018].
UN, 1987. General Assembly Resolution on Implementation of Agenda 21, the Programme for the Further Implementation of Agenda 21 and the outcomes of the World Summit on Sustainable Development. (A/RES/64/236)
Van Westendorp, P., 1976. NSS-Pricesensitivity-Meter (PSM) – A New Approach to Study. Consumer Perception of Prices. Proceedings of the 29th ESOMAR Congress, pp. 139-167.
Verbrugge, B., 2016. Best Practice, Model, Framework, Method, Guidance, Standard: towards a consistent use of terminology – revised. [Online] Available at: https://www.vanharen.net/blog/van-haren-publishing/best-practice-model-framework-method-guidance-standard-towards-consistent-use-terminology/ [Accessed 16 February 2018]
Zimmermann, N., 2017. Network effects helped Facebook win. Deutsche Welle, 8 September 2017.