DSC DACH 25 - Nina Mrzelj & Sumeyye Suri.pdf

"Everyone is talking about using AI, and FOMO creeps in, you want to use it too. But… for what, exactly?

As the hype and interest around AI grow, so does the number of AI projects that seem to exist just because they involve AI. In this session, we will skip the buzzwords and give you practical tools for identifying AI use cases that makes sense for your organization. We will also introduce a “fail fast, recover smart framework” — a strategic approach to identify red-flags early and make sure that the AI projects you choose either succeed or fail productively. You will leave with a clearer understanding of:

Which AI use cases are worth investing your resources in and which ones are just expensive experiments dressed up as innovation,
The early signs of a failing AI initiative and how to avoid them,
Strategies for smart decision-making throughout a project, saving resources, repurposing assets and learnings.


By the end of this talk, we will have learned the impact of smart decision-making in choosing an AI use case, and walk away with a guide for stopping AI failures before they drain time, motivation, and credibility."

1. © 2025© 2025, Sclable Business Solutions GmbH
Hello!
15. October 2025

2. © 2025
2

3. © 2025© 2025, Sclable Business Solutions GmbH
How to Choose the Right AI
Use Cases and Survive the
Wrong Ones
15. October 2025

4. © 2025
4
Dr. Sümeyye Suri
Data Scientist
@ sclable

Nina Mrzelj
Data Science Lead
@ sclable

5. © 2025
5
The AI reality gap
Ambition ≠ Impact
98% 5%
of leaders report
increased urgency to
deliver on AI
of AI use cases make
it into production and
deliver value

6. © 2025
6
The majority of AI failures are not
technical — they’re strategic or
organisational.

7. © 2025
7
What AI really does (and doesn’t) do
finding patterns
and making
predictions
a silver bullet
for unclear
problems

8. © 2025
8
Does this problem need AI?
What is the business
problem?
Can it be solved with simpler
tools?
How will we measure
success?
Do we have the right data
and people?

9. © 2025
9
Evaluate value
Business
Strategy
Customer / User

10. © 2025
10
Evaluate value
Business
Strategy
Customer / User
How large is the estimated business value?
Additional revenue, reduced costs, reduced time, …
How large is the contribution to strategic goals?
Reduced risk, sales growth, new clients, sustainability, …
How big is the influence on customer/employee engagement?
How important is it in connection with user satisfaction?

11. © 2025
11
Evaluate effort
Data
People
Technology

12. © 2025
12
Evaluate effort
Data
People
Technology
Is the needed data defined and available?
Is it accessible in the right quality?
Can we get the stakeholders on board with management support?
Do we have the right people & skills to tackle the use case?
Are the required tech resources known and available?
Is the technology well-known? Or is there a substantial risk?

13. © 2025
13
Value vs Effort matrix
Value
Effort- +
-
+
Value vs Effort

14. © 2025
14
Value vs Effort matrix
Value
Effort- +
-
+
Value vs Effort
We want to
be here :)

15. © 2025
15
AI projects live in a messy,
unpredictable world.

16. © 2025
16
Fail often, fail fast
Recover smart

17. © 2025
17
What to keep in mind?
AI projects are iterative, not linear.
Data quality drives feasibility.
Stakeholder alignment has to happen often.
Choose experimentation over one expensive model.

18. © 2025
18
AI Project Kickoff
●Scope defined vaguely
●Success metrics unclear
●Assumptions:
○Data is ready
○More data can be found
○Proxy data is good
enough

AI Project End
The cost of failing slowly
●Team is stressed and burnt
out
●The data was NOT ready
●The data did not suit the task
●Models did not deliver
miracles
●Outcome is not actionable

19. © 2025
19
AI Project Kickoff
●Scope defined vaguely
●Success metrics unclear
●Assumptions:
○Data is ready
○More data can be found
○Proxy data is good
enough

AI Project End
The cost of failing slowly
●Team is stressed and burnt
out
●The data was NOT ready
●The data did not suit the task
●Models did not deliver
miracles
●Outcome is not actionable
What went wrong
●Constant data challenges surfaced
●Assumptions invalidated too late
●Model built in isolation w/o feedback
●Technical debt accumulated
●Decisions are reactive, not proactive

20. © 2025
20
AI Project Kickoff
●Scope defined vaguely
●Success metrics unclear
●Assumptions:
○Data is ready
○More data can be found
○Proxy data is good
enough

AI Project End
The cost of failing slowly
●Team is stressed and burnt
out
●The data was NOT ready
●The data did not suit the task
●Models did not deliver
miracles
●Outcome is not actionable
What went wrong
●Constant data challenges surfaced
●Assumptions invalidated too late
●Model built in isolation w/o feedback
●Technical debt accumulated
●Decisions are reactive, not proactive

21. © 2025
21
We need to set projects up in a way
that allows for multiple decision gates

22. © 2025
22
Decision gate
Team receives data sample
1-week data validation effort
Assumptions (in)validated early
Data sufficiency validated
Opportunity to kill/pivot and rescope

23. © 2025
23
Team receives data sample
1-week data validation effort
Business
understanding
Data
understanding
Model development
Deployment
Monitoring
Data preparation Evaluation

24. © 2025
24
Team receives data sample
1-week data validation effort
Business
understanding
Data
understanding
Data preparation
Model development
Evaluation
Deployment
Monitoring
Decision gate
Decision gate
Decision gate

25. © 2025
25
Team receives data sample
1-week data validation effort
Business
understanding
Data
understanding
Data preparation
Model development
Evaluation
Deployment
Monitoring

26. © 2025
26
How to fail fast and recover smart
Agile AI Development
●short sprints
●fast feedback
●learn, document,
move on

Versioning
●version data
●version code
●version experiments

Pivot/Kill
●early decisions
●multiple decision
gates

Fail often
●small experiments in
parallel
●analyze what’s/why’s

Lean AI
●test project feasibility
●low cost, low fidelity
models before scaling
up

27. © 2025
27
Takeaways
Don’t start with AI — start with the problem.
Evaluate opportunities with a structured approach.
Align AI projects with measurable business outcomes.

Validate your assumptions early on.
Choose experimentation over high fidelity prototypes.
Document, communicate, rinse and repeat!