A critical assessment

of the proposed
Data Network Effect Theory

Ricardo COSTA CLIMENT,

Dirección y Gestión de Negocios Digitales

1
 Key Message

• AI success: empirical evidence shows that various uses of Machine Learning (ML)
technologies enables some organizations to create and capture value
and thereby positively influence their performance and competitive advantage.

• The Data Network Effects (DNE) theory (Gregory et al, 2020 *)

has been proposed recently to account for the factors
that co-condition value creation from the use of ML technologies

• Limited empirical support:

while very promising, the DNE theory is thus far only supported empirical
by some case studies

• A critical assessment of the DNE theory is provided here

which articulates several of its limitations.

• Further research is also proposed to advance the DNE theory

* Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2020). The role of artificial intelligence and data network effects for
creating user value. Academy of Management Review, (ja).
2
 Background
• For the last 1-2 decades,
New AI technologies, collectively called "Machine Learning", have been
developed and used.

• Machine learning uses represent significant investments by organisations

motivated by anticipated value creation

• Therefore, a key question here is:

– How does the use of machine learning techniques
give rise to value creation?

• The theory of the Data Network Effect, a recent and highly relevant concept in
this context, is worth exploring. It provides the most up-to-date answer to that
question;

– we assess that theory here!

* Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2020). The role of artificial intelligence and data network effects for
creating user value. Academy of Management Review, (ja).
3
 Content

1. Summary of the Data Network Effects theory

2. Critical assessment

3. Next steps: Research approach

4
 1. Summary of the
Data Network Effects theory
Data Network Effects*

DNE shows a direct positive relationship between a platform's AI capability and the

perceived value of the platform by its users; a relationship that is moderated by*:

• platform legitimation,

• data stewardship,

• user-centric design.

* Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2020). The role of artificial intelligence and data network effects for
creating user value. Academy of Management Review, (ja).
5
 1. Summary of the
Data Network Effects theory
• In essence, DNE says that:

– the more a platform, (or a supplier) learns from the data it collects about users,
the more valuable the platform, or product, becomes to each user.

– Active DNE generates increased functionality of offerings (product / service),

with higher quality and improved user experience.

The model extends the theory of Network Effects:

• The Network Effects theory focuses on the value creation conditioned by the size of
the network.
• The recent DNE theory goes beyond the size of the network. The scope of data-driven
learning is the key mechanism for the realisation of DNEs.
• The key to improve the value perceived by users is by improving predictions based on
the identification of patterns in data offers previous uses of the offering.

6
 1. Summary of the Illustration
Data Network Effects theory case
ILLUSTRATIVE CASE STUDY ON THE USE OF DNE:
UBER *

– Uber uses ML algorithms to analyse data collected from users in real-time.

– This improves algorithmic matching and the information offered to each user.

– ML algorithms predict the arrival time of a driver to pick up the rider.

– The rider can plan their time more effectively.

– The better prediction, the more value is offered to the rider, the driver and thus to Uber,
and so, the user decides to continue using Uber.

* Rosenblat, A. (2018). Uberland: How algorithms are

rewriting the rules of work. Univ of California Press.

** Image: 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. WW Norton & Company.
A napkin sketch created by David Sacks, co-founder of
Yammer and veteran of PayPal

Uber DNE Example** 7

 1. Summary of the DNE
Data Network Effects theory Virtuous circle

Use of offers

generates continuous

data points

Users are encouraged to These data points can be

continue using the continuously analyzed for
offers provided. pattern identification.

These predictions can be This pattern identification can

used to develop and adjust be followed by continuous
offerings to create additional learning to make more
value for users. accurate predictions.

8
 Content

1. Summary of the Data Network Effects theory

2. Critical assessment

3. Next steps: Research approach

We have the following key critical reflections here:

• 2.1 Strengths
• 2.2 Criticism in literature
2.3 Clarifications on the criticisms
• 2.4 Our critical analysis
• 2.5 Identified gaps
• 2.6 Research Question.
These are detailed in the following
9
 2.1 Strengths
2. Critical assessment of the DNE theory

• Offers a non-trivial theoretical account of real-life active data network

that are utilized by organizations and cause their success
(e.g. Apple, Amazon, AirBnB, Google, uber, Facebook, etc).

• Thereby, DNE theory accounts for the unique characteristics of ML technologies

and their value creation for its users

• DNE theory advances in the conventional network effects theory,

by moving beyond simple network size as the key factor for value creation

• It offers practical guidance to practitioners that intend to invest in the use of ML

technologies

10
 2.2 Criticism in
2. Critical assessment literature

Critical Point The model does not take into account the difference between:
(external) database (centralized) and user base (decentralized)
I
Differences:
Database User base and add-ons

o Inside the company o Outside the company

o Centrally managed o It is managed in a decentralized way
o Ownership of the company o Free and autonomous to leave
o The size of the base: it's o The size of the base: it has at a given
created over a period of time.
time.

Consecuence:
✓ The tension between the parties for the capture of value is ignored.
✓ It is not a network effect.

* Clough, D. R., & Wu, A. (2020). Artificial Intelligence, Data-Driven Learning, and the Decentralized Structure of Platform
Ecosystems. Academy of Management Review, (ja).
11
 2.2 Criticism in
2. Critical assessment literature

Critical Point The model focuses on the value perceived by the user, and
(external) ignores the capture of value through DNEs*
II

• Model focuses on creating value for the user through DNEs.

• Model ignores how the owner captures value
• It does not take into account the decentralized structure of the platform...
...which creates tensions between the parties over the capture of value.

How the owner can capture It means a value creation reduction for
value… the customer:

Raise prices or lowers quality User sees a decrease in value.

Price discrimination Opportunistic or predatory models.

Captures value at the cost of Having to choose between creating or

value creation capturing value

* Clough, D. R., & Wu, A. (2020). Artificial Intelligence, Data-Driven Learning, and the Decentralized Structure of Platform
Ecosystems. Academy of Management Review, (ja).
12
 2.2 Criticism in
2. Critical assessment literature

Critical Point
(external) The consideration of DNE as a network effect is weak*
III

• Network effects are derived from the size and structure of the user base

• DNE implies that there is a close link between data-driven learning and

the user base

• The differences (slide 11) between the database and the user base mean that:

The coupling between

data-driven learning and the size of the installed base

is weakly coupled so, as to be considered a network effect.

* Clough, D. R., & Wu, A. (2020). Artificial Intelligence, Data-Driven Learning, and the Decentralized Structure of Platform
Ecosystems. Academy of Management Review, (ja).
13
 2.3 Clarifications
2. Critical assessment on the criticisms
(by Gregory et al, 2021)*

CLARIFICATIONS BY GREGORY ET AL (2021) TO CRITICISM

Clough and Wo, 2020 Gregory et al, 2021

• Differences between database • Data does not have to be the

and user base organisation's property.

• Value capture is ignored • Value creation and capture must be

considered holistically in terms of a
duality.

• Doubts about DNE as a

• The conditions for DNE as a network
network effect
effect are fulfilled by AI:
– Product improvement has an impact
on all users, not just one.
– The improvement of the product has
an impact on its current value.

* Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2021). Data Network Effects: Key Conditions, Shared Data, and
the Data Value Duality. Academy of Management Review, (ja).
14
 2.4 Our critical
2. Critical assessment analysis
After analyzing:
- The theory of DNE
- Criticism in literature
- Gregory et al.’s clarifications (2021)

WE IDENTIFY our 4 critical points:

1. The DNE theory do not account specifically for value capture

with DNEs.

2. There are other characteristics of data that affect value creation

and value capture, which are not taken into account.

3. The DNE theory considers "the user" of ML-enabled offerings to be

a single actor.

4. The model has a static conception of value creation, does not take
into account dynamic and temporal factors that affect DNEs

15
 2.4 Our critical
2. Critical assessment analysis

Critical Point The DNE theory does not account specifically for
(own) value capture with DNEs.
1

• DNE theory only focuses on the value created for the user by the DNEs

• Value creation and value capture are distinct and interconnected processes.

• Differences between the interests of parts generate tensions in the

capture of value.
• Therefore, we disagree with the holistic conception of the balance between
creation and capture, as a dual process (Gregory et al, 2021).

• DNE does not address:

1. The value capture mechanism (barriers to entry, talent retention,

bargaining power...)
2. The identification of moderators of the capture (interests of the parties,
data thresholds, type of business model...

16
 2.4 Our critical
2. Critical assessment analysis

Critical Point There are other characteristics of data that affect

(own) value creation and value capture, which are not taken into account
2

• The DNE theory does take into account the QUALITY and QUANTITY of the
data for learning and creating value for the user.

• But...
– there are other aspects related to data that influence the creation
and capture of value.

• Examples of ignored factors include:

• Data expiration
• Existence of critical mass
• Existence of data thresholds
• Data source (interacting users or from sensors?)
• Other factors that do affect network effects:
(social bias, network saturation...).
17
 2.4 Our critical
2. Critical assessment analysis

Critical Point DNE's model considers the user as a single actor,

(own) and ignores the possibility of multi-agent.
3

• The DNE theory considers "the user" of ML-enabled offerings to be a

single actor.

• But casual empirical observations show that the user may be a multi-agent
actor.
• a service or product can be relevant for different agents (health
app)
• The user can be a group of users (users who are companies,
formed by different agents, with different levels of ability to
decide...)
• Each agent has different interests, different purposes

• This implies a different legitimacy for each agent

• The DNE moderating factor, the legitimization in "user-centered design”, is

limited by:
– Values of legality, morality and transparency.
– But there are other individual values what depend on the user's
purpose: personal, ecological, professional, etc. 18
 2.4 Our critical
2. Critical assessment analysis

Critical Point The model has a static conception of value creation,

(own) does not take into account dynamic and temporal factors that affect DNEs
4

• The model has a static conception of value creation.

– the value for the user that has the size of the network at a given time.

– This perspective does not take into account dynamic and temporal
factors that affect DNEs:
• External factors that condition the dynamism of the size of the network
(regulations, fashions, competitors, etc.).
• Expiration of the accumulated data.
• Decrease in product novelty by habituation through use.
• Critical mass
• Network saturation
• Data Thresholds
• Imitation of the competition

– The implementation of IT, the use of ML, does not mean immediate
returns.
– There is a time lapse between the value created by DNE and the
value capture that the owner can get. 19
 2.5 Identified
2. Critical assessment gaps

CRITICAL POINT CRITICAL POINT CRITICAL POINT

(OWN) 2 (OWN) 3 (OWN) 4

Need for consideration Need to reconsider Need to apply a

to incorporate the single user as a dynamic DNE
additional multi-agent actor. perspective
characteristics
of the data which
feeds ML

THEY ARE LINKED TO:

CRITICAL POINT
(OWN) 1:
Need for value capture study using DNEs and ML

20
 2.6 Research
2. Critical assessment Question

RESEARCH QUESTION:

What are the factors that condition value capture from machine learning use and
DNE?

Research sub-questions:

• RQ1: Are there other data-related factors that affect or moderate value
creation and capture, apart from the quantity and quality?

• RQ2: To what extent does the consideration of a multi-agent user affect data
network effects?

• RQ3: How are value creation and capture through DNEs conceived from a
dynamic, temporal and evolutionary perspective?

21
 Content

1. Summary of the Data Network Effects theory

2. Critical assessment

3. Next steps: Research approach

22
 3.1 Research
3. Next steps
Approach

1. Approach:
• Machine Learning (ML) technologies enable some organizations to create and capture
value

• DNE theory* accounts for the factors that co-condition value creation from the use of
ML technologies.

• We assume a limitation presented by the DNE theory*:

– The lack of analysis in terms of value capture by DNE.

2. Research question and sub-questions:

What are the factors that condition value capture from machine learning use and DNE?

• Characterization of data in relation to value capture

Multi-agent user
Dynamic perspective

*Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2020). The role of artificial intelligence and data network
effects for creating user value. Academy of Management Review, (ja).
23
 3.1 Research
3. Next steps
Approach

3. Theoretical framework for answering RQ:

• Theory on value capture / value appropriation

• Business Model Theory (Amit and Zott, 2001)*

4. Methodology

• Literature review

• Case Studies

• Empirical study: QCA qualitative analysis methodology

– Study of the dynamic perspective of DNE

* Amit, R., & Zott, C. (2001). Value creation in e‐business. Strategic management journal, 22(6‐7), 493-520).
24
 3.1 Empirical
3. Next steps
study

Methodology: Qualitative analysis (QCA)

• What do we study? Pattern of interest:

– What conditions affect value capture using ML and DNE from a dynamic
perspective.

• Why with QCA?

– Suitable for small to intermediate N studies.
– Analyzes the necessary and/or sufficient conditions for an outcome of interest.
– It allows you to analyze different combinations to explain a specific result.
– Allows evaluation of equifinality with multiple configurations

• Objectives:
– identify causal issues that lead to outcomes.
– identify necessary and sufficient conditions, or conditional configurations, to
predict behaviors and develop a theory.
25
 3.1 Empirical
3. Next steps
study

Application to the pattern of interest.

– Definition of the sample (2 possibilities, or more?):

• Organizations that have innovated, transformed, their business models
over time, in relation to ML and DNEs.
• Case studies of companies, over a period of time, that are in a process of
change or innovation.

– Definition of the period to be studied.

– Definition of the conditions (and fuzzy sets) by means of the studied theory:
• Market position (high - medium - low).
• External motivating factors for change (all - mixed - internal only).
• Percentage of investment in innovation (up to X%: 1, up to X%: 3, up to
X%: 5).
• Identified data threshold (high - medium - low)
• Etc.

– Case studies, collection of information, data etc.

• In the different defined periods: T1, T2, T3...
26
 3.1 Empirical
3. Next steps
study

– Calibration of the information collected.

– Classification of cases according to their belonging to the given sets.

• According to model type (one side, two or more sides)
• According to sector (Advertising, fashion, financial...)
• Technology based, or non-technology based that uses ML in some
process.
• Etc.

– Conducting a comparative configurational analysis that examines level and

change as causal conditions.

– Relate the identified causally relevant combinations to specific cases at the

conclusion of the analysis.

27
 3.1 Empirical
3. Next steps
study

QCA
LIMITATIONS:
⮚ QCA involves imposing theoretical and relevant knowledge in examining
evidence that is not perfect.

⮚ QCA is useful when the sample has more than 12 cases (not less).

⮚ Results using QCA depend on the ratio of cases to causal conditions.

⮚ QCA is currently not designed to do truly dynamic longitudinal analyses

❖ The cases are never really similar.

❖ There are difficulties finding a similar case with a negative outcome.

❖ It is unlikely to find a crucial difference between cases.

❖ There is multicausality.
28