AAC2025_Danninger_Fail fast succeed smarter.pdf

Every company is at a different stage in the introduction of data science or AI. Not every use case fits every company, and finding the “right one” is often a challenge. Limited resources and a lack of expertise are common obstacles. This presentation will explore this challenge using an agile process to identify, develop and successfully implement impactful data science and AI projects.

1. FAIL FAST, SUCCEED SMARTER:
AN AGILE PROCESS FOR
EFFECTIVE AI IMPLEMENTATION

2. DO SOMETHING NEW. DO SOMETHING BETTER. WHAT WE DO
We are a design-led data,
TRA1SMO o fWR7E 07THNOTT
that specialises in AI &
advanced analytics.
2
Azure Open AI Preferred Partner 1 of 70 AI partners worldwide.
1 of 10 specialist OpenAI partners in the UK.

3. Mriehel (Malta)
Plovdiv Sofia
Basel
Munich
Graz
Vienna
Linz
Porto
Lisbon
Bristol
London
2002
Founded
550+
Employees
1.500+
Digital projects
Hamburg
"One of the most
efficient teams
I have worked with
in my entire career".
VP Engineering, LinkedIn Learning

4. 1.
Challenges & roadblocks of a successful AI implementation
2.
Agile process to implement AI projects
3.
Benefits of an agile process
4.
Lessons learned
Agenda
DO SOMETHING NEW. DO SOMETHING BETTER.

5. Challenges & roadblocks of a
successful AI implementation

6. 6

7. 1.
No approval by management
2.
Lack of suitable data science / AI use cases
3.
Is the existing data suitable?
4.
Lack of technical knowledge to implement an identified use case
5.
The customer’s data science / AI team encounters roadblocks
5 identified challenges when implementing data science or AI projects DO SOMETHING NEW. DO SOMETHING BETTER.

8. Agile process to implement
AI projects

9. We offer an approach to bring data science projects into production in the long term
DO SOMETHING NEW. DO SOMETHING BETTER.
Data & AI Strategy
Data Science / AI
Data Engineering &
Platforms
Data Engineering &
Platforms
… but the devil is in the details

10. No use cases
found
Suitable use
cases found
Predictability
Check (PC)
PC
unsuccessful
PC successful
Are the existing
data suitable?
Lack of
technical
knowledge
Establish an
own DS team
Data quality, more
data required, etc.
3.
4.
DS Team
encounters
roadblocks
5.
Data Science
Expert Review
Let’s dive deeper into the process
DO SOMETHING NEW. DO SOMETHING BETTER.
Successfully
implemented
project
Challenge
Ascent
Service
Other stakeholders or
department required
Envision
Workshop
No approval by
management 1.
Lack of suitable
use cases 2.

11. Envision Workshop to identify Data Science/AI use cases
DO SOMETHING NEW. DO SOMETHING BETTER.
INTRODUCTION AI USE CASES CRAZY 6 NEXT STEPS
ENVISION WORKSHOP
PRIORATIZING

12. DO SOMETHING NEW. DO SOMETHING BETTER.
FEASIBILITY
BUSINESS VALUE
Offer Automation
hoch
low
einfach difficult
Competition monitoring
Demand
Prediction
Customer Service
Envision Workshop – Prioritizing
Which value cases are technically feasible and offer the highest impact/ROI?

13. Data Science
Expert Review
Successfully
implemented
project
PC successful
Lack of
technical
knowledge
Establish an
own DS team
Data quality, more
data required, etc.
4.
DS Team
encounters
roadblocks
5.
Let’s dive deeper into the process
DO SOMETHING NEW. DO SOMETHING BETTER.
Task
Envision
Workshop
No approval by
management 1.
Lack of suitable
use case 2.
Predictability
Check (PC)
No use cases
found
Are the existing
data suitable? 3.
Suitable use
cases found
Challenge
Task /
Outcome
PC
unsuccessful
Ascent
Service
Other stakeholders or
department required

14. Predictability Check to evaluate the customer’s current data
DO SOMETHING NEW. DO SOMETHING BETTER.
PREDICTABILITY
CHECK
DATA PROVISION
DATA
EXPLORATION
PROTOTYPINGEVALUATION

15. Predictability Check – Evaluation
What are the key factors that determine the outcome of the predictability check
DO SOMETHING NEW. DO SOMETHING BETTER.
Unforeseen external
influences can reduce
predictability by
introducing noise and
variability.
Data with consistent
historical patterns
generally offer better
predictability.
Clean, well-structured
and relevant data is
crucial for any
predictive analysis.
Larger data sets can
provide more
comprehensive
insights and patterns
and improve
predictability.
Scope of the data Data quality
Historical
consistency
External factors

16. Successfully
implemented
project
Establish an
own DS team
Let’s dive deeper into the process
DO SOMETHING NEW. DO SOMETHING BETTER.
Lack of suitable
use case
No use cases
found
Suitable use
cases found
Predictability
Check (PC)
PC
unsuccessful
PC successful
Are the existing
data suitable?
Envision
Workshop
No approval by
management
Data quality, more
data required, etc.
1. 2.
3.
Lack of
technical
knowledge
4.
Data Science
Expert Review
DS Team
encounters
roadblocks
5.
Challenge
Task /
Outcome
Ascent
Service
Other stakeholders or
department required

17. Data Science Expert Review to support the customer’s team
DO SOMETHING NEW. DO SOMETHING BETTER.
DATA SCIENCE
EXPERT REVIEW
DATA
PREPARATION
MODELLING &
EVALUATION
DEPLOYMENT
DEVELOPMENT
SETUP
x
o
x
o

18. Let’s dive deeper into the process
DO SOMETHING NEW. DO SOMETHING BETTER.
Lack of suitable
use case
No use cases
found
Suitable use
cases found
Predictability
Check (PC)
PC
unsuccessful
PC successful
Are the existing
data suitable?
Envision
Workshop
No approval by
management
Data quality, more
data required, etc.
1. 2.
3.
Lack of
technical
knowledge
4.
Challenge
Task /
Outcome
Data Science
Expert Review
DS Team
encounters
roadblocks
5.
Establish an
own DS team
Successfully
implemented
project
Ascent
Service
Other stakeholders or
department required

19. We offer an approach to bring data science projects into production in the long term
DO SOMETHING NEW. DO SOMETHING BETTER.
Data & AI Strategy
Data Science / AI
Data Engineering &
Platforms
Data Engineering &
Platforms
… and in practice?

20. Demand prediction for Ölz der Meisterbäcker
DO SOMETHING NEW. DO SOMETHING BETTER.
Is the existing
data suitable?
Lack of
technical
knowledge
Lack of
technical
knowledge
Predictability
Check (PC)
The customer already
had the idea for the
demand prediction
use case
•High data quality
•Great prediction accuracy
on a global scale
•Successful Predictability
Check
•Gained insights of PC
used
•Data preprocessing
pipeline
•Feature engineering
•Deployment using Azure
ML
•Retraining pipeline
•Regular accuracy
monitoring
•Validation together with
the customer

21. Benefits of an agile process

22. 1.
Easily adaptable process to the customer’s current needs or challenges
2.
Fast gain of insights in every stage of the process
3.
Low cost before ramping up a large project
4.
Strong guidance throughout the process
5.
Always the possibility to quit a project fast
FAIL FAST, SUCCEED SMARTER DO SOMETHING NEW. DO SOMETHING BETTER.

23. Lessons learned

24. 1.
Customers have different levels of maturity for implementing data science or
AI projects 2.
Gaining insights fast is the key to support custome rs
3.
Having a battle tested blueprint inspires trust
4.
Structured services / frameworks streamline the process —
Data Thinking Workshop
—
Predictability Check
—
Data Science Expert Review
One agile process for different challenges, industri es, use cases … DO SOMETHING NEW. DO SOMETHING BETTER.

25. DO SOMETHING NEW. DO SOMETHING BETTER.
25
Philipp Danninger Data Scientist [email protected] Contact me on Linkedin