5. Goal of this section
Goal: Provide a framework for
thinking about how data science
can deliver value
Topics:
o Ways to get value from your
data
o Creating smart products and
services
o The Value Chain of Data
Science
o Apply the value chain to a case
Let me explain how
data science can deliver
value to your business!
6. How to get value
from your data?
There are several options
1. Directly sell data (to customers)
2. Use data to generate insights
• Allow better decision making
• Sell them to customers
3. Use data to develop AI solutions
• Improve the jobs and tasks of
employees
• Create smarter products for customers
7. AI solutions leverage data science to get as close as
possible to influencing the action
My keywords for an AI solution are: “learning” and “end-to-end”!
Human
Decision
Action
Analytics
Discipline
Insights
Level
Descriptive
Whathappened?
Human Input
Reporting
Analysis
Prescriptive
Whatshould be done?
Data
Science
Data
Value
Diagnostic
Why did it happen?
Decision Support
Automated Data Products
Predictive
Whatwill happen?
8. Content
Recommender
Dissecting an AI solution like the Netflix personalized
series recommender
Suggested
series list
Series to
suggest
Expected
duration
Already
Watched
Business requirements:
• Increase VCR
• Apply PG Ratings
• Push Netflix Originals
User requirements:
• Good suggestions
• Ease of use
• Great experience
Duration
Prediction
Data sources
Rank &
Filter
Measure KPIs
(VCR, CTR, ratings, etc)
9. The generic architecture of a full fledged AI solution
Advised
actions
Predictive insights
Predictive
insights
Business
requirements
User
requirements
Model
Decision
engine
Measurement
Data
Other
Data/insights
Model
UI
10. We can summarize this conceptually with the “Value
Chain of Data Science”
Actions
Insights
Data
Measure
Optimize
Predict
Value
What an AI solution does…
…and its resulting (intermediate) outputs!
11. It all starts with data, preferably lots of
relevant and high-quality data!
Actions
Insights
Measure
Optimize
Predict
Value
Data
12. Algorithms need data for training,
validating and testing
Actions
Insights
Measure
Optimize
Predict
Value
Data
Train
Validate
Data set split
Train model
Determine performance
Test Determine generalizability
Study
Sample
exam
Exam
13. Apply algorithms to make predictions
Actions
Insights
Measure
Optimize
Predict
Value
Data
14. Use the simplest possible algorithm to
predict, but not too simple
Actions
Insights
Measure
Optimize
Predict
Value
Data
Your model is too
simple to predict reliably
Model predicts future
observations reliably
Your model is great in
memorizing, but does not
generalize
Underfitting Just right Overfitting
15. Nowadays getting all kinds of predictive
insights is “easy”
Actions
Insights
Measure
Optimize
Predict
Value
Data
Regression
(numeric target
variable)
Classification
(categorical target
variable)
Clustering
(groups of
observations)
E.g. number of products,
minutes of delay, number of
visitors, entity salience score,
…
E.g. churn yes/no, risk rating
high/medium/low, car brand,
…
Groups with similar behaviour
(e.g. website visits, customers,
…), Groups with suspicious
behaviour (e.g. fraudulent
transactions)
16. Apply optimization to transform those
predictions into prescriptions
Descriptive
Analytics
What
happened? Diagnostic
Analytics
Why did it
happen?
Predictive
Analytics
What
will happen?
Prescriptive
Analytics
What should I
do?
Prediction
Optimization
Information
Hindsight
Insight
Foresight
Actions
Insights
Measure
Optimize
Predict
Value
Data
17. So the insights become actionable
Actions
Insights
Measure
Optimize
Predict
Value
Data
18. Make sure you store the taken actions
and relevant metrics
Actions
Insights
Measure
Optimize
Predict
Value
Data
Model
metrics
Busines
s
Metrics
Measures related to business value and KPI’s, e.g. revenue,
churn rate, CPA, service level.
Usage
Metrics
Measure related to the model and optimization performance,
e.g. RMSE, MAPE, accuracy, precision
Measures related end-user usage and satisfaction, e.g. % of
followed action, # of calls made, feedback on UI/UX
19. Make it all the way to the end, to finally
create some value
Actions
Insights
Measure
Optimize
Predict
Value
Data
20. Example case: Reduce the waste of fresh
sandwiches sold during flights
Actions
Insights
Data
Measure
Optimize
Predict
Value
21. Example case: beer color prediction
Actions
Insights
Data
Measure
Optimize
Predict
Value
22. Forward navigate the value chain of data
science to discover relevant use cases
Actions
Insights
Data
Measure
Optimize
Predict
Value
23. To discover viable and feasible use cases, work
backwards from value to data
Optimizing what
actions?
Using what insights?
Learned from
what data?
What is the
opportunity?
Measure
Optimize
Predict
Ask more business-oriented questions to uncover good ideas:
• What’s your strategic focus for the coming year?
• Where do you see opportunities for Advanced Analytics?
• How do these opportunities relate to (upcoming) projects?
• What are the business gain/pains that are being addressed?
• Who stand to benefit from these projects, and how?
24. Many organizations start at the data and only go
halfway
Business challenges:
• Optimize end user journey
• Facilitate taking of actions
• Measure outcome and value
• Achieve business objectives
Technical challenges:
• Collect and clean the data
• Build a predictive model
• Generate and collect insights
• Optimize and make actionable
Actions
Insights
Data
Measure
Optimize
Predict
Value
25. Many fail because they focus on details, in isolation
Value
Action
Model
Data
Insights
Not this… ...but this!
Value
Action
Insights
Data
26. Key learnings:
o Many ways to generate value with data
o Use the Value Chain of Data Science to
build analytics solutions
o Start from the value, arrive at the data
o Iterate over the whole value chain
o Start small, create proof, then scale out
27. Quiz time!
1. Take out your phone
2. Go to kahoot.it
3. Enter the pin that will be shown on the screen
4. Type in your nickname
5. The questions appear on the screen. Answer
these with your phone (multiple choice)
6. The fastest correct answer gets most points
29. What does it take to become a Data & AI-driven
organization?
_ WHY Why do we need to become a Data & AI-driven organization?
_ WHAT What do we need to mature our Data & AI capability?
_ HOW
How do we deploy our Data & AI capability to developAI
products?
30. A restaurant analogy: kitchen & service area as one
team
Service Area
aka your Data
Science team
aka the “business”
Kitchen
31. Running a restaurant vs. writing a recipes book
One of the main reason why businesses fail, is hiring Analytics researchers vs. people that can
apply analytics to solve real-life problems
As a restaurant, you need more than chefs.
And you need the service area and kitchen to be aligned.
VS
32. Data & AI maturity can be measured along two axes
Analytical
Capability
Business Adoption
• People & Skills
• Tools &
Technology
• Data
• Executive Support
• Funding
• Implementation
33. A typical Data & AI maturity journey consists of 4
phases
Continuous
Experimentation
Enterprise
Empowerment
Initialization
Business Adoption
Analytical
Capability
Initialization
• Find and initiate firstuse cases: identify
opportunities, boot up data, people & tools
Continuous Experimentation
• Expand team and infrastructureas you are
gradually implementing moredata & AI products.
Enterprise Empowerment
• Grow data & AI practice across allbusiness units and
put business in driver’s seat; buy-in required!
Data& AI Democratization
• Data & AI literacy in genes of company; anyone
has skills required to make data
(& AI) driven decisions
Data & AI
Democratization
34. Each maturity phase has a different focus
Initialization
• Find and initiate firstuse
cases: identify
opportunities, boot up data,
people & tools
• Proving value
• “Can wedo it?”, “How hard
is it?”, “What are the
opportunities?”
• Expand team and
infrastructureas you are
gradually implementing
moreAI products
• Capability building
• “How to organize?”, “What
skills and tech is needed?”,
“How do I repeat and
scale?”
• Grow Data & AI practice
across all business units and
put business in driver’s seat
• Business Adoption
• “How to involvethe
business?”, “Who is paying
for new and matured use
cases?”
• Data & AI literacy in genes of
company; anyonehas skills
required to make Data (&AI)
driven decisions
• Enabling everyone
• “How to supportData & AI
driven decision making across
the organization?”, “How to
measureeffects?”
Focus
Questions
State
Enterprise
Empowerment
Continuous
Experimentation
Data & AI
Democratization
35. Measure your Data & AI maturity to spot improvement
opportunities
Metrics
Axes
Score
Curious about your AI maturity score?
Check our AI maturity self-assessment:
https://xebia.ai/ai_maturity_scan
Team
L&D
Knowledge sharing
Categories
36. Become a Data & AI-driven organization
WHAT do we need to mature our Analytical Capability?
• People & skills
• Tools & technology
• Data
38. A successful team needs to get the following
things ‘right’:
• Project sponsorship
• Analytics translation
• Data science
• Data engineering
Tasks related to these topics can be done by
different roles/people.
39. These roles form minimal viable project team
Analytics Sponsor
• Create a data & AI
strategy and vision
• Make budget and
resources available
• Oversee use case
portfolio
BIZ IT
D&A
Analytics Translator
• Collect, prioritize, and
validateideas
• Drive the solution
developmentprocess
BIZ IT
D&A
Data Scientist
• Explore the dataand
train models
• Explainfinal solution
performance
BIZ IT
D&A
Data Engineer
• Extract data from data
sources
• Embed modelsin
runnablesolution
BIZ IT
D&A
40. Different use cases require additional skills and roles
Software engineering
Agile coaching
Legal & Compliance
UI/UX design
Domain expertise
Plant Operator
43. Upskilling your kitchen crew involves training,
technology and talent
• Develop a talent strategy
• Make choices on the AI
enabling technology
• Based on the talent and
technologies you have,
close the gaps with training
• Tutor-led training
• Self-paced training
• On-the-job learning
• Coaching
Technology
Proper tooling to
develop AI solutions
Training
Foundation of
data experts
Talent
Mix in seniority, theoretical
expertise and business experience
45. Not only on company level, but also at bigger scale
Xebia Data empowers medical students with AI
46. Training is just the start though
Tutor-led and self-
paced learning
On the job learning &
Coaching
Tracking KPIs and
celebrating examples
3
1 Academy 2 Use cases Reinforcement 4 Community
Making it real for
business
48. Hackathons & innovation days
Energizing and fun way to try
out new tools & technologies.
Knowledge sharing helps you to spot new tools &
tech, and prevents you from making the same
mistake twice
Regular project demos
Adopt best practices between
teams and prevent makes
from being made twice.
Meetups, conferences
The more you share
knowledge, the more it
grows.
49. Maturing your People &
Skills Analytical
Capability
Immature Mature
• Mostly reliant on
external consultants
• Senior talent in-house
• No vision & strategy for
growing in-house talent
• Clear career paths and
training curriculum in place
• Knowledge sharing is
minimal or ad hoc
• Constant knowledge
sharing; learning is part of
culture
Examples of maturity indicators
50. Key learnings:
o Your minimum viable team of sponsoring,
translation, data science and data
engineering
o Additional expertise will need to aid the
team
o Provide your technical talents growth
perspective and training
o Foster knowledge sharing to start a
flywheel