Digital-First AI Strategy

How to build a winning digital

strategy with generative AI?

Youngjin Yoo

Associate Dean of Research

Faculty Co-Director, xLab
Elizabeth M. and William C. Treuhaft Professor in
Entrepreneurship
Weatherhead School of Management
Case Western Reserve University

@YoungjinYoo
0.0
 Please download and install the
Slido app on all computers you use

How familiar are you with the

concept of artificial intelligence
(AI)?

ⓘ Start presenting to display the poll results on this slide.

 Please download and install the
Slido app on all computers you use

Which of the following best

describes your company's current
use of AI?

ⓘ Start presenting to display the poll results on this slide.

 Please download and install the
Slido app on all computers you use

What is the role of data in your

organization's AI efforts?

ⓘ Start presenting to display the poll results on this slide.

 Please download and install the
Slido app on all computers you use

Which of the following ways are you currently using

generative AI in your work? (Select all that apply)

ⓘ Start presenting to display the poll results on this slide.

 Please download and install the
Slido app on all computers you use

What are the questions that you have

about generative AI that you want to learn
today?

ⓘ Start presenting to display the poll results on this slide.

 Please download and install the
Slido app on all computers you use

Tell me what you want to know

about generative AI.

ⓘ Start presenting to display the poll results on this slide.

 Please download and install the
Slido app on all computers you use

Ask questions throughout the day

ⓘ Start presenting to display the audience questions on this slide.

1.0
(source: Heider & Simmel, 1944)
In a world of geometric beings, Triangle lived a solitary
life in the House of Shapes. One day, curiosity drove
Triangle to venture outside, where it encountered the
lively duo of Little Triangle and Circle. Fascinated,
Triangle invited them home, unknowingly setting in
motion a transformative journey. As the three
interacted, they danced in patterns of pursuit and
evasion, their movements a metaphor for the
complexities of friendship and understanding. Triangle,
once content in solitude, found itself alternately
exhilarated and overwhelmed by these new
connections. Little Triangle and Circle's partnership
grew stronger, eventually leading them to depart
together, leaving Triangle to reflect on its experiences.
This reflection sparked an epiphany: true growth meant
breaking free from old confines. As Triangle finally
stepped out of its comfort zone, the House of Shapes
began to crumble, symbolizing the beautiful destruction
of limiting beliefs and the embrace of a wider, more
connected world.

Claude 3.5 Sonnet

Industrial
Revolution
Digital
Revolution
Generative
Revolution
The shift from the era of the IT revolution to one of
the generative revolution involves the transformation
of computers from inorganic and syntactic
machines to organic and semantic machines
 Emergence of

Organic Semantic machines

Inorganic
machines

von Neumann
architecture
Inorganic matters are “characterized by repetition or near-
repetition, where separated elements are rendered capable of
returning to previous states or directly anticipating states to come.
These repetitions, past and future, are already contained within
the present: observing their current configurations in enough
detaile provides us with the capacity to understand their future
arrangements” (Grosz 2007)
Organic life is “a complex fold of chemical and
the physical that reveals something not given
within them, something new, emergence… Life,
that excess within matter that seeks to extend
matter beyond itself and its present forms, is not
the ‘origin’ of the virtual but rather one of its
modes of actualization, the potentiality of matter
itself, insofar as matter is the material of life as
well as non-life.” (Grosz 2007)
 Inorganic machines are machines
characterized by repetition or near-
repetition, whose future state can be
predicted with enough observations.
Organic machines are complex folds of
hardware, software, models, and data
connected through procrastinated and
temporary binding in run-time that reveal
something not given within them, something
new, emergence.
 Syntactic
Computing
Machines with bits
"The fundamental problem of
communication is that of reproducing at
one point either exactly or approximately
a message selected at another point.
Frequently the messages have meaning;
that is they refer to or are correlated
according to some system with certain
physical or conceptual entities. These
semantic aspects of communication are
irrelevant to the engineering problem.”

Claude Shannon
Dynamic context of word use
 What are LLMs - Embedding

How would you classify the token

that represents “apple”?

From Mike Rojas of AXS

 What are LLMs - Embedding

Because it’s fruit.

Because it’s round.

How would you classify the token
that represents “apple”?

From Mike Rojas of AXS

 What are LLMs - Embedding

If you say an apple is like a kind of

fruit, then we could choose “C” as
the place to embed the token
Apple.

So,
“Apple” is encoded as [5,5].

[5,5] is called a vector. It is two

dimensional in this example.

From Mike Rojas of AXS

 What are LLMs - Vector Space
Now imagine we want to encode
“apple” into a three dimensional
vector in our embedding. This way
we could encode round and fruit.

What would that look like?

It would be a cube.

We call the collection of all these

numbers vector spaces. Well we encode a lot more than round and
fruit. LLMs use 1000s of dimensions…

From Mike Rojas of AXS

Where it all began: Alex Net (2012)
Disentanglement (2013)
Variational Auto Encoder (2013)
Generative Adversarial Network (2014)
Word2Vec (2013)
Attention Mechanism (2015)
Transformer Architecture (2017)
ELMO (2018)
 Transformer-based Pre-trained Models

BERT (2018) GPT-1 (2018) GPT-2 (2019) GPT-3 (2020)

340 million 117 million 1.5 billion 175 billion
Discovery of Scaling Law (2020)
Hello, world!
 We now have a
generative machine that
is semantic and organic
What does this all mean?
A new form of separation:
Representations and Semantics
Context-agnostic semantic objects in continuous
high-dimensional semantic space
Meaningful
recombination as
an emergent
property from
dynamic
relationality among
semantic objects.
Generative machines, with their organic and
semantic capacity, are transforming the
relationships of work, tools, and humans with
potentially profound implications for economy and
society
 Sources of Organizing Financial
Energy Forms Instruments
Agricultural Kinetic Energy Tribal and State Barter System
Revolution Nation System
Industrial Thermodynamic Hierarchical Securitization
Revolution Energy Coordination & of Capital;
Control System Separation of
Ownership and
Management
IT Revolution Entropic Energy Network Derivatives
organizing
Generative Creative Energy ?? ??
Revolution
Some examples
2.0
Generative AI Hands-On
Exercise: Pet Industry
Product Strategy
 This exercise will guide
participants through the use of
generative AI for market analysis,
persona development, interviews,
Exercise unmet needs analysis, product
Overview ideation, go-to-market strategy,
financial projection, and more.
Step 1: Industry Overview & Persona
Development (15 mins)
Task: Analyze pet industry trends and consumer behavior to develop 3 key
personas using generative AI.

Suggested Data Sets:

• Generational Pet Market in the U.S.

• Pet Ownership Household Penetration (1988-2023)

Suggested Prompts:

• 'Create detailed personas for pet owners across various generations.'

• 'Simulate virtual interviews to understand pet owners’ behaviors and preferences.'
Step 2: Virtual Interviews & Transcript
Generation (15 mins)
Task: Use a second AI model to generate interview
questions, then conduct virtual interviews with the personas
created earlier. Generate full interview transcripts verbatim.

Suggested Prompt:
'What are 5 key questions to ask pet 'Interact with virtual personas and
owners about their unmet needs?' generate detailed interview transcripts.'
 Step 3: Task: Conduct cluster analysis on interview
transcripts using AI to identify unmet needs. Use

Unmet
these insights to generate 3-5 product ideas.

Needs Suggested Data:

Analysis & Interview transcripts generated in Step 2.

Product
Ideation Suggested Prompts:

(20 mins) 'Use the interview transcripts to

identify common unmet needs.'
'Generate new product ideas
based on identified needs.'
Step 4: Go-
to-Market
Strategy &
Financial
Projections
(15 mins)
 Step 5: Brand Identity & Landing Page
Development (15 mins)

Task: Use generative AI to design a landing page for Suggested Prompt:

the new product. Create a brand identity, including a
tagline, logo, and visuals.
'Create a landing page layout for a new pet product.'
'Generate a brand identity, including a logo and tagline, that resonates
with pet owners.'
Step 6: Final Presentation & Reflection (10
mins)
Suggested
Task: Teams present Prompts:
their go-to-market
• 'Summarize your
strategies and financial product idea, market
projections. Reflect on strategy, and financial
key insights gained from projections.'
using generative AI for • 'Reflect on how
generative AI helped
strategic decision- shape your decision-
making. making process.'
 Please download and install the
Slido app on all computers you use

Ask questions throughout the day

ⓘ Start presenting to display the audience questions on this slide.

Lunch Break
3.0
 Please download and install the
Slido app on all computers you use

What words come to your mind

when you think of AI?

ⓘ Start presenting to display the poll results on this slide.

Firms, Value
Creation,
and Value
Capture
Why do firms
exist?
Product – Market
Fit Search
A firm is…

a socio-technical
assemblage to perform the
value creation process
repeatedly at a scale
value creation
process

“the steps and

activities (within and
across firms or
industries) that lead to
the production of
marketable goods”
AI affects both.
Firms often begin with a
known market-product fit
at the category level.
Therefore, the search is
to make a known solution
better, faster, and
cheaper.
Disruptive
innovation
happens when
we challenge
this
assumption.
 Incremental innovation follows a “One to
Many” logic for value capture through
variation reduction with an assumption that
the future is the repeat of the past
Disruptive innovation
follows a “Zero to ONE”
logic for value creation by
embracing variations with
an assumption that the
past is not a reliable basis
for future creation.
Digital Innovation
is ALL ABOUT
value creation &
value capture
THREE
QUESTIONS
Why do I need a
EVERY new solution?

BUSINESS
MUST ASK What solutions
do I offer?

How do I create
and maintain
my solutions?
Digital-First DNA

Why: Meaning
Architecture

What: Value
Architecture

How: Operational
Architecture
Digital transformation is a process
of re-architecting the DNA of a firm
through these three strategic
questions in light of new capabilities
of digital assets.
The question of AI strategy
MUST be asked in this
context.
DIGITAL
FLIP
 sensing

stage
back
predictive &
embedded
time prescriptive
computing assets analytics

things people
computed
user experience

stage
front
space

ubiquitous
computing assets algorithmic
shaping
services

off-line online

Physical Assets Digital Assets

REVENUE CENTER REVENUE CENTER

Digital Assets Physical Assets

COST CENTER COST CENTER
Industrial value
creation process

products & service

Exchange of
raw materials
producer buyer
Value-in-
exchange with
complete
products for
BUYERS
The primary
source of value
in the industrial
economy is
scarcity.
The marginal
value of ALL
industrial assets
is always
negative.
 sensing

DIGITAL-

stage
back
predictive &
embedded
time prescriptive

FIRST VALUE
computing assets analytics

CREATION things people

PROCESS
computed
user experience

stage
front
space

ubiquitous
computing assets algorithmic
shaping
services

off-line online
Value-in-use
with
run-time
products for
USERS
The primary
source of value
in the digital
economy is
ubiquity.
The marginal
value of digital
assets can be
positive.
AI is a part of this
equation.
A trillion-dollar
formula
A powerful digital value loop
with an open third-party
service ecosystem that
creates on-going, dynamic,
and hyper-personal
engagement with users that
produces a positive
externality with a growing
profit margin and robust
recurring revenue streams.
 Context
Inference engine Intention

To construct a What is the most

value loop, a
coherent next action?
Prediction engine Social, institutional, and
material coherences

firm needs:
What is the most
Generative engine attractive next
engagement offer from
us?

Orchestration How can we make it

happen with minimum
engine friction?
Key Design Parameters of Digital Value Loop

COMPLETENESS VELOCITY OF VERACITY OF DEPTH OF THE GRANULARITY

OF THE LOOP THE LOOP THE LOOP LOOP OF THE LOOP
Completeness of a loop

• Who is my user (not buyer)?

• Do I have means to monitor their
actions?
• Can I influence their choice?
Velocity of a loop

• How fast do I iterate the value loop with

a user?
• Do I need a real-time loop?
• How much delay can I endure?
Veracity of a loop
• Not all loops have the same level of veracity
• The veracity of the loop depends on the consequences of the
actions taken by the user
• When users spend money through the loop, it has a higher level
of veracity. (e.g., Amazon)
• When users are actively engaging in the loop, it has a higher level
of veracity. (e.g., Google search)
• When users spend time in engaging with the loop, it has a higher
level of veracity (e.g., Spotify or Google Navigation)
• Social media services have lower level of veracity
Depth of a loop
• Depth of a loop is determined by the number
of layers in action paths that the loop is
engaged with.
• The layers include:
• Physical
• Material
• Emotional
• Cognitive
• Cultural
• The depth of a loop provides the basis of
intentional reference
Granularity of a loop

• Granularity of a loop is determined by

the resolution of the ability to monitor
and shape users’ actions
• Granularity consists of temporal and
spatial dimensions
• A higher granularity provides a greater
value, but demands a higher
computing power
 Operational efficiency

Premium pricing
4 AI value
models
Recurring revenue streams

Derivative revenue streams

 Per unit
$ $

Industrial Revenue Value

Economy with Cost

Price

Cost

Negative
Marginal Value # of
users
# of
users
 Per unit

Digital First $ Revenue $

Economy with Value

Positive Cost Price

Cost

Marginal Value # of users # of users

(log scale)
Positive
Externalities
by DESIGN
with AI
 Learning effects

A DIGITAL Network effects

VALUE LOOP • Direct network
SHOULD • Indirect network
• Relational network
CREATE • Co-creative network
POSITIVE • Data network

EXTERNALITIES Generative effects

THROUGH:
Transformational effects
Learning effects
Learning effects are the super-linear or non-
linear growth of value of the loop due to the
cumulative AI-based learning that produces
additional insights on users’ future needs
These “quantify me”
statistics are interesting.
…. but not for very long

• We quickly learn that we are

• incredibly boring and predictable.
Netflix: Delivering recurring value through
randomized positive reinforcement
via Long Tail matching
You don’t have to be right, all the times. Just often enough!
Disney+ is like a drug dealer for kids and parents with a
singular strong match.

• Kids can watch Frozen million times and they

will never get bored.

• In fact, if someone takes their favorite shows

away, they act like it is the end of the world.

• When kids act like that, parents feel like it is

their own end of the world.
Network effects
• Network effects are the non-linear growth
of value due to the growth of user base
due to the positive network
externalities.
• The more nodes on the network, the
greater the marginal value of a new node.
 Direct Network
Effects
Value Loop

users
 Indirect Network
Effects
Value
Loop
Users
(demand side)
service providers
(supply side)
 Value Loop
Data Network
1
Effects
Value Loop
2

Value Loop
3

users services
 Relational
Value
Loop
Network Effects

Value
Loop

users
 Co-creative
Network Effects
value
loop

co-created users and

content dynamic viral community
Transformation
Effect
• Transformation effects create non-linear value
growth of a digital value loop by transforming
the meaning of user experiences
• Transformation of user experiences can be
accomplished through the reconfiguration of
the relationship among actors, actions,
artifacts, time and space.
• Bundling (of artifacts), coupling (of actions),
and shifting (of time and space) are
foundational elements of transformation
effects
Generative
Effect
• Generative effect is a non-linear
growth of value of a digital value loop
by removing the upper limit of value
creation by leveraging the creative
potential of third-party actors. It
typically amplifies the other three
premium mechanisms (learning,
network, and transformation).
• Generative effect seeks a
procrastinated value creation over
time
4.0
AI Strategy Canvas: A
Framework for
Designing AI Solutions
Why an AI • Introduce the importance of having a structured

Strategy approach to AI projects. Highlight the complexity

of AI implementation, which involves aligning
technical, business, and ethical aspects.
Canvas?
AI Strategy
Canvas
Overview
 Describe the functions or
business units where AI will be
applied, such as customer
service, supply chain
optimization, or predictive
Identifying AI maintenance, and the specific
challenges it aims to solve.
Target Areas Consider categorizing target
areas by their business impact
and feasibility to prioritize AI
initiatives effectively.
 Elaborate on the specific benefits
the AI will bring. If you use it to
improve an existing business
model, choose improved efficiency,
better customer experience, or new
Defining AI insights. If you use it to create a
new digital-first strategy, how it will
Value improve the digital value loop.
Clearly align these benefits with
Propositions business objectives to demonstrate
ROI. Categorize value propositions
in terms of operational efficiencies,
enhanced customer experience, or
new product/service offerings.
AI Model Configuration

Explain why specific models (e.g., large or small

models, open or closed models) are chosen, and
how configurations like hyperparameters will be
managed and optimized for performance.
Consider different approaches such as
foundational AI models (augmented with context
data using retrieval-augmented generation), fine-
tuned models, or models built from scratch.
Highlight the relationship between model task
agnosticity and asset specificity.
 Expand on how data will be prepared,
which models will be trained, what
techniques will be used (e.g.,
supervised learning, reinforcement
learning), and how often retraining will
Designing the occur. Discuss the training
approaches in terms of open-source
Training vs. proprietary models and insourcing
vs. outsourcing.
Strategy
 Describe the roles needed, from data
engineers to ethics officers, including
skills required and how teams will be
organized for effective collaboration.
Include details about insourcing vs.
outsourcing considerations for
AI Team and personnel, as well as key roles such as
Expertise machine learning engineers, data
privacy specialists, and AI strategists.
Describe how team members can
work together in a cross-functional
setup to facilitate the deployment of
AI.
Integration of AI
Models
Describe the steps for automating data
collection, model versioning, CI/CD
processes, and monitoring performance
once in production. Highlight different
integration approaches such as horizontal
integration (generative AI used in
tools/services) versus vertical integration
(AI for creating deterministic
tools/services). Address the practicalities
of integrating AI models within existing
enterprise systems and how different
stakeholders might interact with these
models.
Crafting a Data
Strategy
Detail how you will source, clean, store,
and ensure the quality of data. Discuss
issues related to data governance, privacy,
and compliance. Include categories of
context data such as public, proprietary
third-party, enterprise-specific,
transaction-specific, or customer-specific
data, and their impact on the AI model.
Consider the implications of using different
types of data for context—whether to
provide rich context or rely solely on
system prompts for zero-shot learning.
 Provide details on cloud vs. on-
premises infrastructure, computing
requirements, networking, storage,
and scalability considerations that
support the AI solution. Include
AI aspects of hardware selection for
training and inference, whether to use
Infrastructure new or refurbished hardware, and
when to leverage cloud-based as-a-
Requirements service options. Discuss
considerations for timeliness of
inference (real-time vs. delayed
processing).
Understanding the
Cost Structure
Break down the fixed and variable costs
associated with AI projects. Discuss
the financial impact of using
proprietary vs. open-source models,
and how to optimize cost efficiency.
 Include detailed governance policies
that address ethical considerations,
decision-making responsibilities, data
privacy, safety, and adherence to
regulations. Discuss committees or
Governance roles that will oversee AI use, how
intellectual property and comparative
and advantage considerations will be
managed, and risk management
Compliance practices to mitigate biases or ethical
concerns. Address the flexibility of AI
customization during inference and its
implications for asset specificity.
Case Discussion:
ChemTech Co
5.0
 Please download and install the
Slido app on all computers you use

What words come to your mind?

ⓘ Start presenting to display the poll results on this slide.

Reflections
and
Questions
Remained