A holistic approach

to value-driven AI

The AI Canvas
Methodology

merantix-momentum.com
2022 Merantix Momentum
 AI Canvas
Methodology

Executive Summary
The AI Canvas Methodology - a holistic approach to deriving value from Artificial Intelligence

Data and AI-based business models are transforming every industry. Early adopters of AI have
gained significant financial returns [1], whereas those not implementing it risk falling behind
significantly. But applying artificial intelligence and deriving value-driven business models from it
are fundamentally different. Whereas many companies are showcasing singular and detached
applications, implementing AI at the operational core of a company and letting it sustainably
transform business models is far more complicated.

This challenge stems largely from the dynamic nature and inherent uncertainty of AI projects and
integrating them into running businesses. Changes in external and internal requirements, data and
technological infrastructure have major and often unpredictable effects on the success of a
project. At the same time, existing processes and structures impact the implementability of and
resulting returns from AI projects. Therefore companies need to plan and evaluate their AI
activities holistically, which includes assessing all aspects of AI cases before implementing them.

In order to enable companies to develop a successful and sustainable approach to AI adoption,

Merantix Momentum partnered up with the Institute of Computer Science at the University of
St. Gallen to develop the AI Canvas - a strategic toolkit for companies to guide them through
successful AI project evaluation, implementation and sustainable deployment.

The hierarchically structured framework addresses the described complexity drivers of AI projects
and is divided into four main blocks which each play an essential role in planning and
implementing AI initiatives:

Business Organization Technology AI Lifecycle

Understanding which Integrating AI into the Evaluating which data Designing AI models in
underlying business current organizational is needed to execute a sustainable fashion
case AI can and should structure and ongoing the respective cases and ensuring that
fulfill to derive processes to minimize and which ongoing value can be
impactful and friction and ensure technological derived by integrating
measurable value in continuous capabilities and a continuous life cycle
line with the company effectiveness of capacities are needed from the inception of
strategy. solutions. to support this. AI implementation.

The AI Canvas is the first step to holistically approaching value-driven AI. It supports companies of
various AI maturity levels, from those considering implementing their first use case to those
scaling their existing AI implementation. This whitepaper provides an ideal starting point through
an in-depth explanation of the four blocks of the AI Canvas and suggested courses of action. This
is illustrated using case studies of how Merantix Momentum has enabled companies to
successfully implement AI using the Canvas.

If you would like more information on our services, please do not hesitate to contact us at
momentum@merantix.com.

2
The AI Canvas
Methodology
Contents

Preface 4

1. The Importance and Difficulty of Successfully Implementing AI 5

AI is the defining competitive advantage of our century… 6
…And many companies are not ready 7

2. AI Canvas: A Framework for Planning and Scaling AI 8

The Business Block 10
Value Proposition 11
Performance Indicators 13
Sample Case 14
The Technology Block 15
Data & Technology 16
Operational Implications 17
Sample Case 19
The Organization Block 20
Sample Case 22
The AI Life Cycle Block 23

3. Using the AI Canvas 26

4. The AI Canvas offered by 30

Merantix Momentum

Conclusion 33

The AI Canvas
Methodology
 Preface

The 21st century is a truly transformational period as we are experiencing a

digital revolution which is redefining almost all aspects of our lives. Artificial
Intelligence (AI) is an immense value driver in these transformations; it is
estimated that AI will add a staggering $15.7T [2] to the global economy before
2030 - that is approximately the size of China’s GDP. We believe AI is an
enabler. It is not there to replace humans, it is not there to solve every business
problem. Instead, it will enhance human capabilities and provide them with safer
and more efficient working conditions.

The adoption of AI by companies has been accelerating in recent years, from

37% of companies implementing AI in some form in 2019 [3] to 56% in 2021 [4].
Because of the fast-paced developments in AI technologies and their adoption,
companies that come to the game later face significant competitive
disadvantages which they may never be able to recover from. On the other
hand, early adopters of AI have above average returns on investments in AI.
According to research by Deloitte, 82% of early adopters have gained significant
financial returns [1]. In light of this, it is essential for companies to act now to
remain competitive in an economy increasingly shaped by AI.

Yet, many companies struggle with effective AI implementation. They tend to

develop multiple proof of concept projects but find themselves unable to move
beyond this experimentation phase into commercial exploitation. Scaling AI in a
company is challenging as such endeavors often require organizational changes
to various departments and functions and a high degree of flexibility due to the
developing nature of the technology.

To help companies overcome these challenges, Merantix Momentum partnered

up with the Institute of Computer Science at the University of St. Gallen to
develop a methodology for approaching the topic of AI with a focus on
scalability - the AI Canvas. The AI Canvas is a strategic toolkit guiding
companies through AI project ideation, implementation and sustainable
deployment. It helps companies align on strategic priorities, identify any
road-blocks, understand the data landscape and requirements while taking into
account regulatory and ethical questions.

In the first section of this publication, we present the main challenges

companies face when implementing and scaling AI, carefully selected based on
our experience and conversations with companies. In the second section, we
explain the AI Canvas and dive into each of its building blocks. In the third
section, we present two case studies of how Merantix Momentum has enabled
companies to successfully implement AI using the Canvas.

4
The AI Canvas
Methodology
1.
The Importance
and Difficulty of
Successfully
Implementing AI

5
AI is the defining competitive advantage of our century…
Over the last decade, research on artificial the use of AI may further shift the degree of
intelligence (AI) has progressed markedly. automation that is possible in many fields.
While in the early 2000s the industrial use of
AI was constrained to specific technologies Clearly, due to AI’s enormous disruptive
and domains like expert systems in the potential, this transformation presents both a
medical field, the technology has matured to singular opportunity as well as a great threat
be applicable in practically any industry. to companies who do not effectively leverage
this potential. Those that understand and use
The drivers of this development have been the capabilities of this technology will hold a
the usage of highly parallelized computing competitive advantage that might prove
devices (GPUs), the availability of massive decisive for decades to come. The utilization
datasets, technology firms that invest heavily of AI not only allows for gains in productivity,
in the area (e.g. Google, Facebook, Amazon) but also accelerates development
and ever-advancing research all around the significantly and widens the gap between
world. competitors at increasing speed.

The impact of this change is often compared However, while companies’ adoption of AI has
to the impact of utilizing electricity during been accelerating in recent years, few are
industrialization. And indeed we can observe able to effectively realize the technology’s
many parallels between these two transformative potential. In our work with both
transformations. While during the national and international players we have
industrialization period many mechanical identified three common scenarios:
tasks could be automatized through the help
of machines,

The potential impact of AI may be recognized universally but the path

to implementation is still rocky for many organizations. The main
Organizations that do
issues they face are the lack of understanding of AI across all levels
not use AI at all of the organization as well as the lack of talent to develop AI from
inside of the organization.

The AI journey in many companies typically starts with conducting

Organizations that proofs of concepts for one or multiple use cases. While this
investment is important to educate teams and management on the
conduct isolated proof
potential of AI, the long-term business impact is oftentimes rather
of AI concepts but fail contained. This is due to the lack of an organizational setup for
to scale them scaling AI and alignment between the proposed value of the project
and the business goals of the organization.

Organizations that
scale use cases, but Leveraging AI in an efficient manner goes far beyond scaling
individual use cases. For a truly efficient set-up, organizations have
fail to transfer to think about how to create synergies, and share resources and
knowledge across the knowledge across the whole organization.
organization

6
6
The AI Canvas
Methodology
…And many companies are not ready
From the issues raised above we can see that While some approaches are sufficiently
becoming fully AI-capable requires strategic general to apply to nearly every strategic
planning and the right organizational setup. issue (for instance SWOT analyses), they do
not help in identifying all necessary
However, current tools that support strategic components of an AI strategy.
planning and ideation are not readily
applicable to AI setups. Notably we see four main peculiarities of AI
implementation and scaling that render
current solutions for strategic planning
impractical:

Fast-paced technological The need for redefining, monitoring,

development and updating AI products

Research and development in the field of Dataset shift, changes in the environment,
AI increases rapidly as the AI market is and the fast pace of AI development
expected to grow to a $126 billion industry require an iterative approach to monitoring,
by 2025 [5]. This requires stakeholders to defining, and updating AI products.
constantly reevaluate their strategy and
thus necessitates a very flexible tool.

Uncertainty of AI projects The need for a horizontal setup and

understanding of AI
The planning of projects that involve AI is
significantly complicated by the inherent In contrast to other technologies, AI is a
uncertainty of AI projects. Internal and topic that touches upon nearly every
external requirements, changes in the data aspect of a business. It is therefore
or infrastructure have large and often necessary to develop a broad
unforeseeable impacts on the outcome of understanding and horizontal setup for
a project. It is therefore necessary to implementing AI. Classical business cases
develop an understanding of all those are often conceptualized in isolation which
factors and integrate them into high-level is reflected in an equally narrow
strategic planning. perspective in tackling these cases.

7
7
The AI Canvas
Methodology
2.
The AI Canvas - A
Framework for
Planning and
Scaling AI

8
The AI Canvas - Deriving Business Value from AI
The AI Canvas is a strategic toolkit to guide Each of these are broken down into two further
companies through AI project implementation sub-blocks, reaching Level 2. After they are
and sustainable deployment. It serves as a covered, the Organization Block is explained and
planning tool for developing, implementing, and finally, the AI Life Cycle Block.
scaling AI use cases as well as for developing
long term AI strategies. Thereby, it aids Each of these larger building blocks and their
companies in addressing the difficulties outlined sub-blocks are divided even further into smaller
in the previous section. block elements on Level 3. These small blocks
provide the detailed aspects that should be
This chapter explains the building blocks of the considered when implementing AI projects and
AI Canvas. It is split into sections based on the scaling them, and practical guidance for how to
four main areas (Level 1) of considerations when go about this. In the following, each of these is
implementing AI. First, the Business block will be elaborated upon, including questions to answer
covered, and the Technology block will be for each small block, and a suggested format for
explicated afterwards. answering these questions.

Level 1

Level 2

Level 3

9
9
The AI Canvas
Methodology
 The Business
Block
The business block enables companies to assess
their needs, forming the foundation for an AI based
value proposition. It further aids with monitoring
and measuring success of the business case.

10
The AI Canvas
Methodology
 Value Proposition

1.1 Problem

This building block is about defining the Sample questions to answer within this block:
underlying challenge to be addressed - be it an AI
solution targeting a specific business area or a ● What is the problem?
more holistic approach to organizational AI ● What is the current workflow? What are the
transformation. pain points in this workflow?
● How much is it worth to solve this problem?
Especially with regard to AI, it is of great ● What is the size of the problem?
importance to understand exactly what challenge ● Have you already tried solving this and how?
you want to solve, derive an estimate of the ● Are there other problems that could use a
problem’s size, the value of its resolution and how similar solution?
well it can and must be solved with the help of
technology. Additionally, it is crucial to understand
the current workflow and existing pain points in
order to determine what type of AI solutions can
be applicable.

Problem & use case ideation workshop with the

Format
goal of delivering a prioritized use case list

1.2 User/Customer

Sample questions to answer within this The user building block defines the different
block: groups of people or departments that will use and
benefit from the AI solution.
● Who are the types of users? What is their
Knowing which use cases are of high priority,
workflow?
relevant internal and external users should be
● How coherent/different are the user
groups? identified. Knowing the users is crucial to
● How will they interact with the solution? understanding what the solution will look like, what
● If the product is to be sold externally, what kind of outcomes it needs, and how it can be
are the customer segments? What are their integrated into the current workflow. The identified
characteristics? users will be involved in the scoping and
development of the solution, according to their
respective characteristics and importance.

User journey development workshop with the

Format
goal of identifying main user pain points

11
11
The AI Canvas
Methodology
 Value Proposition

1.3 Products/Services

The product & service block defines a concrete AI

Sample questions to answer within this block:
solution by drawing upon the assessment of the
pain points and main user journeys.
● What are the defining functionalities?
● What scale is required to derive the
In addition to responding to user needs, successful
envisioned value?
AI products or services also require the
● How central is this product to your
specification of key functionalities in the form of
business?
parameters. This exercise is to ensure that the
● What will the interaction between human &
proposed solution correctly resolves the underlying
technology look like?
issues, particularly in terms of the required and
desired value creation and the potential strategic
importance.

Scoping session to identify the desired

Format functionalities and value the solution should
deliver

1.4 Partners

Sample questions to answer within this The partner block identifies internal and external
block: partners crucial to AI development.

In order to fully understand how AI will be

● Who are your internal partners? Who do you
introduced to the company, it is crucial to identify
need to get internal buy-in?
internal and external partners. Internal partners
● Who do you need to support us in
designing and building the product? must be identified to ensure their buy-in or to
● Who can help you with data collection? enable the design of the application. AI solutions
● Which key resources are you acquiring from often require collaboration with specific partners
internal/external partners? who can provide essential key resources or
● Which key activities do partners perform? activities that substantially enable and influence
their development. Examples of these include data
collection, data labeling, etc.

User story development workshop with the goal of identifying internal

Format
and external partners affected by and contributing to the process

12
12
The AI Canvas
Methodology
 Performance Indicators

2.1 Financial and non-financial returns

The financial and non-financial return block
focuses on setting up important KPIs and
Sample questions to answer within this block:
assessing related commercial risk.
● What are the overarching goals you are
It is important to formulate project-specific KPIs to
trying to achieve?
ensure that the AI solution is appropriately
● What could be indicators for progress
managed during its conception as well as
towards these goals?
sustainably deployed and maintained thereafter.
● Which of these indicators are measurable
These KPIs should accurately reflect what the
and practical as KPIs for project progress and
solution is intended to achieve and how this
success?
generated value relates to the core value
proposition of the business as a whole. With AI
solutions, such objectives often range from
financial goals to non-financial goals such as time
to market, customer satisfaction level, ESG, carbon
footprint etc.

Workshop to ideate, identify and

Format
rank the most relevant KPIs

2.2 Investment

Sample questions to answer within this The investment block dives deep into the
block: necessary up-front and long-term costs that a
company needs to invest when introducing AI.
● What are the cost implications for creating
When calculating the costs of AI solutions, various
value and scaling value?
● How much do you need to budget for a) dimensions must be taken into account. In this
data integration, b) technology, c) people context, one must consider the cost implications
development? for each stage of the project (PoC, scaling,
● How long does the integration take and organizational level). For each step, different costs
when can you expect financial returns? arise from both human and capital resources for
data integration, technology and human resources
development. These investments need to be
assessed and understood.

Project scoping with the execution team (internal or external). Focus is to

understand the major investment categories (data, compute, model development), Format
the cost drivers for each category and the respective uncertainty.

13
13
The AI Canvas
Methodology
Sample Case Business Block - Retail

Input - Request for engaging with the customer

Merantix Momentum partnered with a B2C retail company selling through its
e-commerce and offline store, but with no real personalisation for the customer. For
this they wanted to utilize AI, but had to first specify the business case.

The outcome of the first session was that they were essentially
Problem struggling to target their interactions and outreach to the customer.
This was a large problem for marketing, because it limited customer’s
engagement with the brand/company, and due to the huge customer
number, this issue could not feasibly be solved manually.

The main users having the problem were the marketing and sales User
department. They would need a personal profile of preferences for
each customer in an accessible format to target them. Another group
were the in-store staff, who could also utilise this information.

Product The product was defined to be a collaborative filtering-based

recommender engine which, based on existing and ongoing
customer data, would create an individual profile of preferences for
each customer to allow sales to target them directly.

The most important partners identified internally were the sales and Partners
marketing team as well as the data infrastructure team, providing and
managing access to the customer data needed in the solution

Financial and Despite not producing direct efficiency savings or revenue by itself, it
was determined that the solution would greatly improve sales and
non-financial marketing interactions, as well as increase customer interest and
returns engagement and drive revenue through product recommendations.

Based upon the product feature detailing, an investment plan Investment

containing the overall costs (data acquisition, model design, system
integration…) and duration was designed. Additionally, cost drivers
like update rate, access points and interfaces were detailed out.

Output - Detailed business case for an AI recommender system

After working through the business block of the AI Canvas Methodology, Merantix and
Takeaway
its partner had a clear understanding of the business purpose of the proposed
application. This meant which problem it was to solve, which features it would need for
that, who would have to be involved and how the investment and returns would look.
From here, the next step was to go into technological detailing of the solution.

The AI Canvas
Methodology
 The Technology
Block
Data and technology are the core enabling factors for
applying AI. Being able to analyse and understand them and
their limitations, as well as choosing the right framework to
operate them, guarantees value from AI in the long run.

15
The AI Canvas
Methodology
 Data & Technology

3.1 Data

The data building block assesses data needs and

Sample questions to answer within this block:
necessary infrastructure.
● How much data do you have?
Data can be considered to be the lifeblood of AI. It
● What is the data structure and quality?
is the necessary ingredient of any Machine
● What are the data sources?
Learning solution and is thus the most important
● Is there bias in the data?
asset of an AI-enabled company. Companies have
● Is data accessible through a common
to implement data governance principles that
interface or individual silos?
provide a common interface for accessing and ● Are processes in place to ensure compliance
using data throughout the organization and assess with data privacy and regulations?
the data availability, characteristics, specification
etc.

Data assessment workshop to understand

Format
existing data and further requirements

3.2 Technology

Sample questions to answer within this The technology building block looks into the
block: available technologies and sustainable software
development.
● What are the AI capabilities?
AI technology and research currently moves at a
● What degree of maintenance is needed?
very fast pace. To keep up with that speed and
● What frameworks can be used?
● What hardware platforms can be used? benefit from it means building technology in a
● Have common practices been established scalable, modular way. Implementing Machine
for the whole company? Learning solutions may incur a large hidden
● Are systems modularized so that they can technical debt. Committing to reducing this debt
be reused? and identifying it early on ensures the
technological stability of an organization.

Development of a technology
Format
scorecard specific to the client

16
16
The AI Canvas
Methodology
 Operational Implications

4.1 Governance & Compliance

The governance and compliance block addresses

Sample questions to answer within this block:
the internal framework for monitoring the AI
application and its implications.
● Can you pinpoint accountability?
Because of the novelty of AI software, it is crucial ● Who is liable for the AI software?
● How is your AI supply chain set-up?
for any newly developed applications to be
● Who owns the generated content?
embedded in a strong internal governance
● Have processes been established to enable
mechanism ensuring oversight and accountability
oversight and guarantee compliance?
for the model. This means that companies have to
clearly, reliably and transparently document how
and by whom the applications will be overlooked
and how the compliance with ethical and legal
maxims is ensured.

Governance and accountability outline showing the embedding

Format
of the application into current company compliance structures

4.2 Regulation

The regulation block addresses further regulatory

Sample questions to answer within this
challenges and ethical concerns.
block:

● What requirements for explainability are Lawmakers are starting to explicitly regulate the
needed? Are you able to fulfill these use of AI. General regulations (for instance GDPR)
requirements? and the need to fulfill certification requirements
● Are there any ethical concerns? may pose further external restrictions on the
● Is there a complete list of regulations that implementation of AI. This is inherently linked to
affect IT, data, AI across all of the company? the technology that powers a solution, since
traceability and reproducibility are essential
features that allow us to understand and explain
the decisions of AI systems.

AI documentation package - precompiled list of documents that can be used

Format
to record the behavior of AI systems to ensure traceability and reproducibility

17
17
The AI Canvas
Methodology
 Operational Implications

4.3 Safety

The safety block addresses questions related to

assuring that AI models are safe to deploy and do Sample questions to answer within this block:
not cause harm or damage.
● Has continuous monitoring and evaluation of
While we have a general and intuitive AI models been set up?
understanding of the safety and accuracy a human ● Do requirements for the documentation of
operator may provide, this is in general not true for the expected results of the AI model exist?
AI systems. To build a safe AI system, robustness ● What are the worst possible effects a
of the model has to be tested and it has to be deviation/failure of a model could have?
● How robust is the model?
documented thoroughly.

Risk analysis for all modes of failure and potential

Format
effects as well as mitigation and avoidance strategies

4.4 Security

The security block addresses platform security

Sample questions to answer within this
and assures that AI software is safe from external
block:
attacks or manipulations.
● Are processes in place for identifying
potential targets of attackers in the AI In addition to general IT security, AI systems allow
infrastructure? for additional attack vectors. Building a secure AI
● Did you ensure platform security? system comprises protection against data leakage
● Did you ensure access control? and adversarial attacks.
● Did you ensure model updates?

Exposure analysis for all points of entry or adversary as well as

Format
mitigation and avoidance strategies like access controls and hosting

18
18
The AI Canvas
Methodology
Sample Case Business Block - Dental Analysis

Input - Use case idea for detecting caries on dental scans

Merantix Momentum partnered with a dental scan company to develop an algorithm for
automatically detecting caries and classifying it by type. This would support dentists in
their work and increase efficiency and reliability of scan interpretations.

Data It was determined that a labeled data set containing at least 2.000
images would be necessary. The data would have to be labeled by
multiple experts to ensure reliability of labels as ground truth. Initial
assessments of bias and balance were also concluded.

The technology used was based upon the client's existing technology Technology
and utilised available company know-how as well as possible. For
example, a Rest API interface with the company’s systems and
remote development of the training infrastructure were selected.

Governance & A detailed accountability map for the solution was developed,
Compliance governing responsibilities for the monitoring and validation of model
outputs. Additionally, an escalation framework was designed for
cases of hazardous results or deteriorating model performance.

Regulatory frameworks around usage of healthcare data and AI Regulation

applications were analysed. As a result, a strong GDPR compliant
data management framework was conceptualized and a roadmap to
classification and regulatory approval of the model was designed.

Safety Model output implications and resulting worst case scenarios were
discussed and key risk drivers were determined. As part of the
governance framework, mitigation strategies for key risks were
ideated.

A risk map and infringement model for the application were outlined, Security
detailing possible areas of adversarial attacks and data leakage.
Similar to the safety block, mitigation approaches were ideated and
implemented into the technical design and setup of the system.

Output - Sound technical outline for a caries classification model

Takeaway
The output of the technical block offered a holistic overview of how the AI model
would look, which data it would use and how compliance, safety and security would be
ensured through governance. This output allows the technical team to design the
model structure and begin building features or elements.

The AI Canvas
Methodology
 The Organization
Block
The organization block assesses how AI use cases are to be integrated
into running processes and how an effective environment for AI
development and scaling can be built. It also highlights which cultural
and structural change should be initiated

20
The AI Canvas
Methodology
 Organization

5.1 Structural setup

Organizations can choose different internal setups
for their AI efforts such as centralized,
decentralized, or hybrid setups. Sample questions to answer within this block:

AI is the next disruptive general technology and ● Do you have/want to have a centralized,
thus requires a novel approach to leading and decentralized, or hybrid structure?
operating organizations. AI initiatives often face ● How do you set up an internal knowledge
organizational barriers that prevent them from sharing platform?
scaling, such as rigid culture, lack of knowledge, or ● Which AI-related competencies does your
talent. Because of these challenges, companies organization have?
tend to struggle to move past isolated proofs of ● What new roles do you need to introduce in
concept and scale projects to company-wide interdisciplinary teams?
programs. Therefore, to build an AI-first company,
leadership must consider which organizational
setup is most conducive to AI development, who
makes decisions about AI projects and
infrastructure, and which roles are necessary.

Workshop evaluating how AI activities are to be pursued, which departments/ roles

Format
they interact with and how they are to be embedded into the current organization

5.2 Development

The development block focuses on various ways

Sample questions to answer within this
of educating the company about AI.
block:

● Who do you need to educate about AI? One of the main barriers in AI adoption in
● What is the best format to educate companies is the lack of knowledge and weak
employees and management about AI? communication about the technology. AI
● Shall you do this internally or with external implementation requires management and
help? workforce to have a deeper understanding of
current AI technologies, their possibilities and
limitations. In order to develop a successful AI
solution, you need technical knowledge as well as
business understanding. It is therefore crucial to
educate employees about the benefits, risks, and
value of AI systems.

Conceptualization and development of an

Format
education program on AI for relevant stakeholders

21
21
The AI Canvas
Methodology
Sample Case Business Block - Semiconductors

Input - Desire for a strategic vision on AI competencies

Merantix Momentum supported a semiconductor manufacturing company with setting
up a full AI operation model for them to ideate, validate, prototype and develop AI use
cases internally and with external support.

In the initial phase, structures and processes were derived, defining

how AI innovations were to be ideated, validated and executed within
Structural the organisation. This included the roles and responsibilities to be
Setup covered within the company.

The main structural elements included an AI Innovation Funnel and

Validation stage, allowing effective and efficient ideation, evaluation,
validation and testing of the AI ideas.

In order to sustainably embed the AI activities into the company,

significant initial training was needed to develop understanding of
and acceptance for the technology and a development methodology. Development
This consisted of an AI crash course for all relevant stakeholders, as
well as a definition of current existing talent, capacities and needs for
respective supportive functions within the organisation or through
external support.

Output - AI coordination center with the company

The output of the organizational block was a clear structure for how AI activities were
going to be structured in the company, and which processes would support them in
ideation and development. It laid the groundwork for a strategic AI setup, enhancing
the company’s AI activities and organization.

The AI Canvas
Methodology
 The AI Life Cycle
Block
Keeping the long-term integration of the AI solution into the
business in mind whilst designing it is crucial for ensuring
sustained value. The AI Life Cycle block helps evaluate this
and conceptualize the ongoing utilization of the AI model.

23
The AI Canvas
Methodology
AI Life Cycle
Artificial Intelligence is an ever-changing, ever-evolving field of technology. Therefore, a
once conceptualized solution or approach can quickly become obsolete. Having the
continuous evolution and long-term goal of AI across all blocks of the AI Canvas in mind
when conceptualizing the initial approach is paramount to sustained value creation.

6.1 Business 6.3 Organization 6.2 Technology

Organizations have to
actually live the With new requirements,
Scoping, planning and
designated processes, shift in data distribution
measuring RoI of AI
and continuously monitor and other external effects,
products is a continuous
the system and its models might change or
process that has to be
function within the need to be updated,
done in accordance with
business and organisation. requiring constant
progress on the
Also the roles and monitoring and
technological side of the
responsibilities related to reevaluating of the data
life cycle.
AI are to be regularly and technology in usage.
re-evaluated.

This shows that the building blocks of the AI Canvas should not only be considered
when first implementing or scaling AI, but also need to be continuously looked at,
even when a company is a seasoned AI adopter.

24
24
The AI Canvas
Methodology
 From Ideation to AI Life Cycle
The AI Life Cycle is the final stage in the development of an AI solution. For an idea to progress to this
stage, the technical feasibility should first be tested through a Proof of Concept (PoC), after which the
value add of the solution is determined through the creation of a Minimum Viable Product (MVP). Each of
these stages involves iterative development and preparation for the AI Life Cycle stage, where the
technology, business case and organizational processes are continuously re-evaluated.

Idea

Goal of the PoC - Demonstrating the Defining the right scope of of a PoC is tricky.
PoC technology It should only be a testing rig for proving core
technical ideas, but all core aspects that will
The goal of a PoC is to demonstrate that become critical later on must be covered.
the technical assumptions you have
made throughout your ideation process Out of scope for a PoC are all features
hold true. At the end, you want to be beyond core concepts, like those pertaining
sure that the way you envision your to acceptance or customer/user satisfaction.
product or solution to work is feasible. They should be included in later phases.

Goal of the MVP - The scope of an MVP should include all core
MVP Proving the solution adds value features relevant to solution delivery and be
capable of being provided to users to test
The goal of the MVP phase is to implementability. The degree of integration
determine if your idea is adding business for the features is reduced to a functional
value for the issue at hand and is minimum to ensure cost-sensitivity.
functional for users. Here, you test
applicability and gather valuable
feedback for the final product Out of scope for an MVP are non-crucial
development. To maximize impact, the components relating to areas like aesthetics
MVP should be developed as quickly and or integration. These are postponed to
cheaply as possible. product delivery.

Making the leap to a product

Product
After having validated your MVP with users and stakeholders, it is ready to be translated
into a real product. At this point, the innovation cycle is completed and the idea is
transferred into the AI Life Cycle.

Entering the AI Life Cycle

AI Life Cycle
The AI Life Cycle encompasses the final product stage in which
operations and continuous development become an ongoing cycle.
The idea has now become a fully integrated and functional solution. At
this stage, development, operations, model maintenance (training,
deployment, evaluation, retraining) and management become crucial
elements to ensuring continuous delivery of the solution. They should
be regularly re-evaluated by looking at all Canvas blocks.

The AI Canvas
Methodology
3.
Using the AI
Canvas

26
 How to Use the AI Canvas - Evaluating Use Cases
The AI Canvas is first and foremost used for ideating and evaluating AI use cases. While
the Business and Technology blocks are most relevant to planning an individual use
case, it should be remembered that it will later be subject to scaling and ultimately,
become part of the organizational setup. Hence, it is advisable to go through each
sub-block of the Canvas and answer the questions detailed in Section 2 - examples
given below - with people from various functions and departments relating to the
respective block.

Starting with the idea of adopting an AI solution,

you go through each Canvas sub-block:

Value Proposition ● What is the current workflow and what

Determining a problem to solve, the are its pain points?
solution’s core functionalities, benefits ● What are defining functionalities to
and its users and partners. generate value and who would use it?

Performance Indicators: ● What are your overarching goals?

Determining relevant KPIs for measuring ● How can these be measured?
and evaluating projects progress and ● Which KPIs are to be tracked?
success (ROI, Investment Case) ● What are the cost implications?

Data & Technology

● How much (quality) data do you have?
Evaluating data quality and creating a
● What are the AI solution’s capabilities?
data acquisition strategy. Choosing the
● What existing frameworks can be used?
tools and frameworks for implementing
● Which capabilities will be needed?
and maintaining the AI solution.

Operational Implications
Evaluating internal AI governance ● Who is liable for the AI software?
processes. Understanding how to ● How robust is the model?
manage regulatory, ethical, safety and ● Did you ensure platform security?
security challenges.

Additionally, across all use cases you evaluate

company integration and life cycle implications…

Organisation
Designing an effective environment for ● What AI-related competences does your
developing and scaling AI by combining organization have?
existing structures and processes with ● Who do you need to educate about AI?
new purpose-built ones.

AI Life Cycle
AI projects require an iterative approach to implementation, monitoring and
improvement. This need for re-evaluation should be considered during planning
throughout all blocks.

…resulting in a fully assessed and outlined

application generating value through AI and a
road to implementation.

The AI Canvas
Methodology
 How to Use the AI Canvas -
Comparing and Prioritizing AI Applications
As part of the use case ideation and evaluation, you will likely devise several possible AI
applications. The AI Canvas also helps with identifying the use cases with the highest
potential and thus deciding which application idea(s) to prioritize. The matrix below
shows how this can be done using insights from the Business and Technology blocks.

Business Technology

The Business block enables you to The Technology block enables .

identify how much invest and how you to understand how feasible .
much resulting business value the the technical implementation of .
implementation of each use case the solution is, in terms of available data,
idea could bring. the difficulty of training, monitoring and
securing the model, as well as the impact
of regulatory and ethical issues.
implementation

HIGH Long-term projects

High potential Use cases with high

efficiency and potential
financial return, Assess
and clear KPIs
BUSINESS VALUE

Implement

Fast
Medium to low successes
measurable
added value Not
Assess
interesting

LOW

LOW TECHNICAL FEASIBILITY HIGH

Solvable ML problems
Complex ML
(high data quality and
problems (low data
large data quantity,
availability or
tested and reliable
difficult labelling)
models)

The AI Canvas
Methodology
How to Use the AI Canvas -
Roadmap to Implementation and Scaling
Hitherto, use cases have been
ideated and validated through AI Project Methodology
the AI Canvas, particularly the Business
Development and Use cases & their business
Business and Technology
execution standards for value, KPIs, users & partners
blocks. To realise the expected AI projects
business value and
technological feasibility, Organization & Life Cycle
Structural framework
organizational structures must AI governance, continuous
Use case assessment monitoring and improvement,
be purposefully developed for
through the AI Canvas skills and education,
AI adoption and continuous knowledge sharing
improvement. The AI Canvas,
particularly the Organization Impactful Use Case Funnel
Technology
and Life Cycle blocks, guides Business processes Data infrastructure, model
organizations on this road to incl. roles, decision operation and safety,
successful long-term AI gates and handovers handling security, legal
compliance and ethics
implementation and scaling.

Using the Canvas to set up these internal Moreover, the AI Canvas is also beneficial for
structures enables an organization to create an organizations who have already implemented
impactful funnel for validating use cases and AI use cases. Going through all Canvas blocks
implement them using AI-specific methodology. enables them to evaluate their current AI
This includes following the PoC-MVP-Life Cycle adoption holistically. Doing this while paying
stages of development. Moreover, it involves special attention to the Organization and Life
embedding the Life Cycle - the continuous Cycle blocks further helps companies create
improvement of methods, processes and an effective strategy for improving and
strategies - as well as cross-organisational scaling their existing AI adoption. Thereby,
knowledge sharing into company structures. the Canvas aids organizations with increasing
the value generated from AI implementation.
This is informed by insights from the other
Canvas blocks. For example, a healthcare Since understanding AI is quite complex, a
company might have identified various company does not need to identify and
operational implications concerning legal, ethical evaluate use cases or build an AI strategy
and data security issues when validating their alone. There are many options of partnerships
first use case. Due to the nature of this heavily and collaborations where an external service
regulated industry, similar issues are likely to provider with high expertise, like Merantix
apply to many future use cases. Hence, the Momentum, can introduce a company to the
organization can increase the long-term topic of AI.
effectiveness of AI implementation by
establishing frameworks and internal Such partnerships can work on both
governance processes for understanding and short-term and long-term basis and can be in
dealing with operational implications. the form of workshops, consulting services,
development services, deployment services,
Using the AI Canvas holistically thus enables maintenance services, and others. In the
companies first starting out with AI to build solid following section, we delve deeper into
foundations for successful long-term AI Merantix Momentum’s offering.
adoption from their initial use case onwards.

29
29
The AI Canvas
Methodology
4.
The AI Canvas
offered by Merantix
Momentum

30
How to Use the AI Canvas - The AI Canvas Workshop

● Increased Understanding of AI and data management

● Detailed and interactive introduction to use case development and
Key
evaluation with the AI Canvas
Takeaways ● 2-3 fully conceptualized AI use cases ready for execution
● Data assessment for customer-specific data sets on use cases
● Development roadmap from PoC to rollout for the use cases

Length: 4 days (2 of which are workshop)

Place: The workshop is carried out at the client site
Participants: 5-15 participants from all relevant departments
Format
Our Team: 2-3 AI Strategists & Machine Learning Engineers

Agenda per day

Onsite Workshop Onsite Workshop

Preparation with the Introduction to AI, Evaluation of existing Consolidation of

client and our team data management and data infrastructure workshop results
business case building
Selection of Evaluation of the Development of a
participants Collection of use case technical feasibility of detailed roadmap
ideas through team previously discussed approach for
Initial industry sparring use cases discussed use cases
research and sample
use case preparation Business case Development of an
development and understanding of
discussion of process operational implications
integration
Definition of final user
journey for discussed
solutions

Creation of a
development roadmap

The AI Canvas
Methodology
 Merantix Momentum - What We Offer

AI Strategy
Generating value through AI

With AI Strategy we enable companies to utilise AI effectively and efficiently,

thereby reaping actual value from it. This includes support on education,
ideation, business case building and structural set ups within customers’
organisations. We partner on all levels, from small workshops to transformative
strategies, to deliver success to our customers.

AI Development
Building AI solutions for partners

AI Developments means conceptualizing and delivering state-of-the-art

machine learning technologies. We take our clients challenges and transfer
them into ML approaches, ideating with leading AI engineers and researchers
on the best applicable technologies. Depending on the task at hand, we build
the AI solution, from PoC to fully scaled industry solution, for and with our
customers.

AI Operations
Ensuring continuous performance

AI Operations ensure that the solutions built by us continuously deliver value to

our customers. This means hosting, monitoring, securing and retraining
models, to ensure their performance continuously remains at peak levels. For
this, we work closely with our customers on their specific needs and domain
best practices.

If you’d like to know more, send us an email at

momentum@merantix.com - we will provide you with further
information and see how your company can derive value from AI!

The AI Canvas
Methodology
 CONCLUSION
AI is revolutionizing the way companies operate and is presenting
endless opportunities for generating value. However, the majority of
companies are struggling with sustainably deploying and scaling AI
across their organization. In this whitepaper, we have presented a
strategic framework for helping companies address these problems.

The AI Canvas guides companies through the topic of AI, from

conceptualising first use cases to scaling them and transforming
organizational structures. It offers a structured process of identifying
key topics and challenges in the building blocks of Business,
Technology, Organization, and AI Life Cycle.

We recommend using the AI Canvas as a first step into, and as a

way of securing a foothold in, the world of AI.
Merantix Momentum is here to support
you along this journey.
Are you ready?

33
About Us

Nicole Büttner
Founder & CEO Merantix Momentum is your company’s
external machine learning department.
nicole.buettner@merantix.com
We help you with ideation, identification,
implementation, and scaling of AI solutions
with high business value. We also provide
strategic advice on AI.
Alexander Fecke Our world class machine learning experts
NLP Lead have deep knowledge of data-driven
alexander.fecke@merantix.com business models and have designed solutions
for various industries including logistics,
manufacturing, retail, healthcare, legal
services, and others. We focus on three key
technologies: natural language processing,
computer vision, and predictive analytics.

Moritz Schröder Get in touch if you would like to learn more

ML Strategy Lead about how we can support you!
moritz.schroeder@merantix.com
+49 15752493220 momentum@merantix.com
merantix-momentum.com

The Artificial Intelligence and Machine

Learning [AI:ML] chair headed by Prof.
Damian Borth covers the area of deep neural
networks. Our research focuses on
representation learning through supervised
and unsupervised approaches with
applications to text-to-speech generation,
computer vision and remote sensing, and
financial time-series data.

https://ics.unisg.ch/

34
The AI Canvas
Methodology
Endnotes
[1] Deloitte Insights (2018) State of AI in the Enterprise, 2nd Edition.
Available at:
https://www2.deloitte.com/content/dam/insights/us/articles/4780_State-of-AI-in-
the-enterprise/DI_State-of-AI-in-the-enterprise-2nd-ed.pdf

[2] PwC (2019) Sizing the Price.

Available at:
https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-analysis-sizing-the-p
rize-report.pdf

[3] Gartner (2019) Gartner Survey Shows 37 Percent of Organizations Have Implemented
AI in Some Form.
Available at:
https://www.gartner.com/en/newsroom/press-releases/2019-01-21-gartner-surve
y-shows-37-percent-of-organizations-have#:~:text=%E2%80%9CFour%20years
%20ago%2C%20AI%20implementation,research%20vice%20president%20at%20G
artner

[4] McKinsey (2021) The State of AI in 2021.

Available at:
https://www.mckinsey.com/business-functions/quantumblack/our-insights/global
-survey-the-state-of-ai-in-2021

[5] Statista (2020) Artificial intelligence software market revenue worldwide 2018-2025.
Available at:
https://www.statista.com/statistics/607716/worldwide-artificial-intelligence-mark
et-revenues/

35
The AI Canvas
Methodology
merantix-momentum.com AI Campus Berlin
momentum@merantix.com Max-Urich-Str. 3
13355
Berlin, Germany