A Data-Driven Company: 21 Lessons for Large Organizations to Create Value from AI, by Richard Benjamins, Chief AI and Data Strategist at Telefónica.
*Machine Learning School in The Netherlands 2022.
1. From talking the walk to walking the talk
Practical lessons to become a data-driven
company and create value from AI
Richard Benjamins
Chief AI & Data Strategist – Telefónica
Co-founder of the Spanish Observatory for Ethical and Social Impacts of AI (OdiseIA)
Board member CDP (Climate Change ONG)
@vrbenjamins
8. Becoming a data-driven company requires a
complex journey Artificial Intelligence
Cognitive Power
- AI
2017 2018 2019
AURA
4th Platform
Internal Use Cases
Social Good Ethics
Privacy
Trust
8
9. There are many lessons to learn from others and accelerate the
journey
9
Organization
Where to place
CDO?
Data vs IT
AI vs Data
Data maturity
External
monetization
Business
Select use cases
Measuring
economic impact
Funding of data
journey
Open Data and
business
SMEs and AI
Technology
Cloud or on-
premise
Global vs local
storage and data
model
Where to run
analytics
Data collection
Outsourcing?
People
Winning over
sceptics
Data
democratization
Creating
momentum with
data
Responsibility
Ethical and
societal
challenges
Responsible AI
Data for Good
• Important decisions to take
explicitly (options, +/-),
depending on data maturity
• 70% similar across sectors
11. 11
How to measure economic impact?
Challenges to measure economic impact
1. Data is almost never the only contributing factor
2. Hesitations to “publish” results for fear of
consequences
• Start with new use cases, heavily based on data
• Measure the uplift (control groups)
• Fear for less budget (savings) or higher targets (revenues)
• Over time, this usually disappears
Three types of economic benefits
today in 5-10 years
Reducing IT
costs
Optimization
of business
External
monetization
New business
12. 12
How to fund the data journey? Corporation versus business units
Data maturity
Business funding
Corporate funding
Pilot Deployment Production
Business funding
Corporate funding
Pilot Deployment Production
Business funding
Corporate funding
Data maturity
Business funding
Corporate funding
Implementation of corporate strategy
Pilot Deployment Production
Business funding
Corporate funding
Asset development
13. 13
Where to place the Chief Data Officer?
Area Pros Cons
CMO Marketing and sales provide “use cases”
with direct impact
Usually focused on B2C, forgetting the B2B
area, not capturing value in other areas
CFO Financial ledger requires high-quality data Less business focused, and financial
management doesn’t need big data
CIO Technology used according to company
standards
Governed by technological criteria, not
business
CTO Take advantage of the latest technological
innovations
Driven by new technology, rather than by
business
CSO (security) Good for security and privacy of customer
data
Less focus on business
CRO (resources) Cost savings go directly to the bottom line Driven by efficiency, rather than by growth
CEO-5 CEO-4 CEO-3
CEO-2
CEO-1
CDO, CTO, …
14. 14
How to relate the IT department with the CDO?
IT
Data …
Technology-driven
…
Data IT
Good!
Data IT
Frequent
Data IT
Data IT
Reasonable
Data
Data IT
IT
Better
16. 16
How to organize external monetization? Existing Big Data department New business unit
Concept Pros Cons Pros Cons
Platform No additional
costs
Adapt to
external use
cases
Built for
external use
Additional
costs
Skills Team in place No
businesspeopl
e
E2E profiles
Budget Leverage
existing
investments
Mixing P&L
with cost
centre
Clean P&L Some
duplication of
investments
People Recognize
existing data
professionals
Ignoring
existing
professionals
Innovation Mixing
operation with
innovation
Innovate at the
edge
Some
duplication
SLA Internal SLA
not sufficient
for external
clients
Client-driven
SLA
Data
governance
In place Define again
Privacy In place Only for
internal use
Specific for
external use of
data
Data
sourcing
In place Focus on
internal use
cases
Dedicated
data sourcing
for external
use
Some
duplication
Anonymous data
Insights
Business
solutions
Cost
of
provision
Business
value
Go to Market
• Existing sales force vs.
creating
• Generic sales force vs.
specialized
17. 17
Global versus local storage and data model?
Typical
organization
Local data model Unified data model
Local data storage Starting organization Data mature
organization
Global data
storage
Organization focused
on synergies (savings)
Digital native
organization
18. 18
Where to perform the analytics? Global versus local
MNC
Operating business level
OB1
OBn
OB3
OB2
HQ
OB2
BU1
BUn
BU3
BU2
Analytics
centre
19. 19
You need a data collection strategy
What should a data collection strategy cover?
• What data to collect and when.
• Where and how to store the data.
• Estimation of costs and budget assignment.
• Effort of breaking data silos.
THE BATTLE FOR
VOICE DATA
THE BATTLE FOR DATA:
NETWORK DATA
OWNERSHIP OF DATA
Data collection by design ORGANIZATIONAL SILOS
20. 20
When to (not) outsource?
Why do organizations with external parties?
• Lack of knowledge
• Create more bandwidth
• External assessment
• Innovation
• Data democratization
Modes of collaboration
• One-off – get it done
• Long term – partnerships
• Learn and internalize – BOT
• Acquisition – escalate fast internally
Risks
• Lock-in
• Dependent on 3rd party
• Lack of knowledge and experience
21. 21
Winning over sceptics
Best is to start with champions …
Sceptics
• “I know exactly what I need to do”
• “I do this for 20 years now, so don’t tell me anything about my business”
• “Come back when you have proven in this company that it works”
• Data is power and power is hard to share
• Departmental silos
I know exactly
what I need to do
22. 22
Data democratization – don’t confine value in a small department
Bring the value of data and AI to full organization.
• A layered approach (like an onion from inside out)
• Training
• Tools
Additional benefits
• Avoid bottlenecks
• Specialist work on hardest problem
• Motivation
• Retention
• Specialist focus on new things
AML
23. 23
Why creating momentum with data and AI?
• Organizations have limited patience for seeing results
• After 18 months, a presentation for the board
• Don’t wait until you are asked for a presentation to the board
• Keep a record of all results
• Publish results early on the internet
• Done versus perfect (data scientists)
• Organizations listen to external sources
• Work with external purposeful organizations
• Publish work externally
25. 25
How to implement the responsible use of AI in your organization?
What AI principles to choose?
Components of Responsible AI by Design
• Principles
• Awareness, training
• Questionnaire with recommendations
• Tools
• Governance model
• Actionable ethical principles
• Unintended consequences
• AI-specific versus generic
• Sector-specific considerations