B.COM Unit – 4 ( CORPORATE SOCIAL RESPONSIBILITY ( CSR ).pptx
Data is not the new snake oil
1. Presented to:
Data Analytics Club, ASB
4 Oct 2021
Akshay Regulagedda
Senior Manager, Data Science
Data Is Not the New Snake Oil
Formulating an AI Strategy and Implementing It
Photo by Isaac Smith on Unsplash
2. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Executive Summary
2
Here are the key “takeaways” from this presentation.
• “Data Is Not The New Snake Oil”:
• Analyzed data has been seen as a competitive advantage by many first-mover organizations.
• But many organizations have been unsuccessful in executing AI.
• Objective Function:
• A well-crafted data/ AI strategy can generate value and create competitive advantages.
3. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Data has been seen as a competitive advantage for companies such asTesco.
4
Data (techniques to extract value from data, e.g. AI) has been compared with basic commodities.
Coined by Clive Humby
• Humby & his wife, Edwina Dunn
started a data consultancy,
Dunnhumby in 1989
• Engaged by Tesco in 1995 to build a
loyalty card system
Chairman of Tesco’s board:
“What scares me is that you know more
about my customers after 3 months than I
know after 30 years”
Dunnhumby’s “Blueprint”
Statistic profiles of customers to generate
reward/ incentive vouchers and coupons
Results
• World’s first supermarket loyalty card,
Clubcard
• Tesco doubled its market share in 1995 a
result of the loyalty card
Clive Humby, in 2006:
“Data is the new oil.”
4. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Data has become plentiful, which increased the hype around data.
5
An explosion of memory and processing resources led to more data… and more interest in using data.
Explosion of Data in Global “DataSphere”
Using Google searches as a zeitgeist:
• No searches till 2009
• Took off after 2017, and again in 2018
• Many articles: particularly in The Economist (2017), Guardian, Forbes,
Wired etc.
• Asia focus: Pronouncements in late 2017/ early 2018 by prominent
regional CEO’s
• Global DataSphere: Amount of data created, captured,
replicated each year
• 99.91% of available data created in the last 18 years.
• Change in Type of Data:
Historic, structured data => Real-time flow of
unstructured data
5
1,277
2,837
5,900
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
2002 2010 2012 2020
Exabytes
Explosion in people saying “Data is the new oil”
5. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
But many companies have not seen benefits.
6
Most engagements have not seen value, even though the promise still holds for most.
You’re in a boardroom, listening to the promise of how data would transform the entire company.
“The missing piece is data. We’ve never been able to truly understand our customer, how they have or how
to optimize our processes and products. That’s why we need to invest heavily into this data space, today. If
we don’t, we’re going to fall behind.”
If you’re reading this, you might note to yourself “that isn’t just 2017…” You’re right.
Eric Weber, Data doesn’t automatically deliver, Sept 2021
Survey of 2500 decision-makers by
BCG and MIT
“Little Impact So far”
70%
Decision-makers who have
“significant investments”
“Can’t report any benefits”
40%
Survey of decision-makers (PWC)
Plans to deploy AI
20%
2019
4%
2020
Survey of decision-makers (PWC)
“Already implemented AI”
27%
2019
18%
2020
Economist, “Business are finding AI Hard to Adopt”, 11 Jun 2020
However:
Survey of 2500 decision-makers by BCG and MIT
“Business Opportunity”
90% Survey of 2500 decision-makers by BCG and MIT
Risk of Competitors Implementing AI First
45%
Feedback:
“The state of AI hype has far exceeded the state of AI science, especially when it pertains to validation and readiness for implementation in patient care.” – Eric Topol, 2019
BCG – MIT Survey
“Financial Benefits”
1 in 10
6. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Change in large companies is hard, even as data / AI has its limitations.
7
Here are some of the reasons why executing AI successfully in large companies can be hard.
People &
Processes
AI Talent
Tech Costs
Data Issues
Finding Use-
Cases
• Initial gains in companies that were nimble and fast. Much harder in established, large co’s.
• Change is hard. Electricity took 30 years to be adopted in rich countries.
• Need cross-functional teams: depth and breadth
• Expensive: “Only the tech giants and the hedge funds can afford to employ these people”
• Initial investment is extremely low, but operational costs can scale up very rapidly.
• Need trained talent to monitor costs
• Data is plentiful, but useable (“labelled”) data extremely hard to find (e.g., electricity data)
• Data cascades: Projects take years before they realized the initial training data was faulty
• Moravec’s Paradox: Machines find complex arithmetic and formal logic easy, but struggle with
tasks like coordinated movement and locomotion.
• Defining business “boundaries” is a rare skill
• Easier to build models that can play Go than to do a customer chatbot
“Production-
ization”
• Integrating even well-constructed models into workflows is hard
• Explaining models and their output can be a challenge
Economist, “Business are finding AI Hard to Adopt”, 11 Jun 2020
7. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
AI high performers have similar elements in their AI playbook.
8
“The biggest gaps between AI high performers and others aren’t just in technical areas, but in human aspects of AI. They
develop/ customize AI capabilities in-house”.
Strategy
AI Talent
• Vision & roadmap prioritizing AI initiatives linked to business value & broader strategy
• Senior management aligned and committed.
• Active program to develop & manage AI ecosystem partnerships with companies & academia.
• AI talent needs understood
• Tailored curriculums by role to develop AI skills for in-house staff
• Central coordination of AI initiatives balanced with close connectivity to end users
Ways of Working
• Clear framework for AI governance covers all steps of Data Science Lifecycle and manages risks
• Standard protocols to build and deliver AI tools
• Advanced processes (e.g. data operations, microservices) to deploy AI
• Uses design thinking, involves end-user in development of AI tools
Models, Tools,
Tech
• Standardized end-to-end platform and toolsets for data-science, data-engineering and app-dev
• Understand how frequently AI models need to be updated
• Track AI-model performance and explanations to ensure outcomes improve over time
Data
• Data strategy that supports and enables AI with data dictionary.
• Processes for ensuring data quality and labelling AI training data
• Generate synthetic data to train AI models when there are insufficient natural data-sets
Adoption
• Enact effective change management (e.g. leaders model behaviours)
• Process to move AI solutions from pilot to production
• Systematically track comprehensive set of KPI’s to measure impact
McKinsey, State of AI in 2020, Nov 2020
8. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Data science is the science of learning from data.
9
If data science is about improving measurable outcomes, then the data-science tools themselves can be improved upon as well.
Open-ended “detective” work to expose unexpected
features, identifying anomalies, …. “cleaning” data.
Implementing appropriate transformation
to restructure data to reveal more insights.
Computational optimization, cloud computing,
developing workflows, repeatable artifacts, and so on.
1) Data Exploration:
2) Data Transformation:
3) Computing with Data:
Through a single statistical model (Generative
Modelling) or methods constructed that predict
well over a dataset (predictive modelling)
4) Data Modelling:
Explaining data-science methods &
results though dashboards &
storytelling.
5) Data Products:
Data-science principles applied to business
functions — measurement, analysis,
improvement —to be applied to data-science
tools themselves.
6) Science about data-science:
9. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Data product strategy has 4 broad quadrants.
10
A well-executed product strategy sits at the center of the four domains.
Business Execution
• Product Fit: Business case for product (Strategy fit
for use-cases, measuring potential & actual uplift)
• Org Fit: Marshalling resources (computational,
technical and domain)
• Product execution: Manage stakeholders, timelines
• Workflow Fit: Ensure integration with workflow
and measure results
• Gathering data: Getting what data, how frequently
• Sharing resources: Maintaining data dictionary
• Managing data: Structured, semi-structured,
unstructured data served through data-lake
• Data Pipelines: Reliability of EDM pipelines
• Toolset: Maintaining common sets of internal tools for
data-scientists and designers to use
• Identify wants versus needs
• Users will not be able to articulate needs in an AI
context.
• Design-thinking to generate use-cases
• Design UX as replicable toolkits (platforms, classes
of screens, in addition to ‘golden samples’)
AI’s magic sauce:
• Exploratory data models, transformations
• Measuring model lift, trying out various approaches
• Execute data-science lifecycle, share model output
Strategy sits here: equal elements from all four quadrants
Data Engineering
Design Data Science
10. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Strategy Fit: Helps to look at a firm as a whole in through progressive lenses.
11
This is an open-ended enquiry. Ideally, it should be done separate from a product strategy.
1) Problem Statement
Summary of current situation
• Situation:
Who, when, where
• Current Strategy:
What is the firm’s focus?
• Identifying the Key
Issue(s):
• Timeline?
• Consequences?
• Decision-making
criteria?
• Success factors?
2) External Analysis
What’s happening outside the firm
• General Environment:
Trends in external environment
that could influence industry &
firm
• Industry Environment:
Five Forces, profit potential etc.
What is the value-chain of
industry & is it changing?
• Competitor Environment:
Strategic groups (firms that
follow similar strategies)? Key
competitors? Capabilities and
competitive advantages?
3) Internal Analysis
What’s happening inside the firm
• Resources:
• Tangible: financial, org, physical, tech
• Intangible: Human, innovation,
reputation
• Capabilities:
What has the firm created in combining
resources (e.g. HR, Marketing, R&D etc)?
• Core Competencies:
Are these capabilities considered core to
the firm?
• Competitive Advantage:
Are the capabilities non-substitutable?
Advantageous? Sustainable?
• Evaluation of Current Position:
Where does a firm sit relative to its
competitors?
• Corporate & Business Strategies:
Have firm’s strategies changed over
time? What is the firm’s value chain? Is
it creating value?
4) Alternatives
Evaluating MECE alternatives to solve issue
• Current Strategy:
Will current strategy provide the
firm a competitive advantage?
• Different Paths:
New strategy or modifications?
What strategic paths exist (e.g.
outsource, acquire, divest,
restructure)?
• Decision Criteria:
Constraints or opportunities that
will let you pick choices
• Quantitative: Margin, ROI, Market
share, capacity, delivery time, risk,
cash flow, inventory turn,
productivity, staff turnover, quality,
growth, quantity
• Qualitative: Competitive
advantage, customer satisfaction,
morale, image, ease, synergy,
ethics, safety, appeal, culture etc.
• Evaluation:
Compare and pick.
11. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Strategy Fit: Implementation plans more valuable than analysis.
12
You may or may not need to present this to management. But it’s best to be able to articulate this when someone asks.
1) Recommendation
Why did you pick this alternative?
• Defend the picked alternative.
• Outline Expected Results
• Financial implications
because of suggested plan
• Success Criteria: how do you
know when you are
successful
• How does it fair against
decision criteria
2) Implementation Plan
How will you execute the pick?
• Identify who, what, where and how
• Resource requirements and justify their
availability
• Timeline for suggested plan
3) Risks/ Assumptions
What can go wrong?
• Assumptions:
• Indicate reasonable assumptions
made in the selection of the
alternative
• Usually quantifiable, but not
necessarily so
• Risks:
• What can go wrong in your
assumptions? How would you
mitigate them?
12. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Product Fit has three parts to it.
13
Plan
• What are the firm's products?
• Why are they doing it?
• Can you quantify the objectives?
• If not, why not
• If so, what are the metrics
you're tracking
• How do they relate back to
qualitative objectives
Execute
• Objective Function: What
is the one (main) thing you
are trying to solve
• Framework in next slide
Once (an informal) corporate strategy has been articulated, start thinking in terms of product.
Measure Results
• Establish a baseline before project
execution
• Ground up from frontline staff
• Broad agreement from
management
• Goal of project is model-uplift
• Iterate through stages & phases
• Start small, dream big
• Qualitatively: What is the effect of
the firm doing it
Iterate with same use-case
Tackle complimentary
use-cases
• Why aren’t we achieving
model-uplift: data, people,
approach
• Tackle each factor separately
• “Free” dashboard screen
reusing existing parts
• Doodle, brainstorm, deepen
13. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Find an Objective Function for the product through a search tree.
14
Objective Function: Mathematical function that maximizes a specific variable (e.g. revenue, costs, new customers etc)
Build in-house
+ Value of
product
- Tech/ cloud
costs
Profits
Per deployment
Revenue Costs
Buy from vendor
+ Value of
target
- Purchase price
+ Additional
value/ costs
- Integration
costs
+ / - Vendor
management
Cultural / legal
difficulties
-
Price Volume
* Variable Costs Volume
* Fixed Costs
+
• Industry baseline
• Adj. per segment
• Segment market to
understand customer base
• Better segmentation via AI?
Short Term
• Value in tracking / adjusting
price in real time?
• Amt per customer (loyalty
prog etc)
Long Term
• Increase options per
product (AI?)
• More differentiation per
customer segmentation
• Value Chain to see all costs
Improve Costs
• Reduce frequency
• Predictability
• Objective repeatability on gut-
feeling-based cost reduction
14. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
AI requires a Data Science Lifecycle, different from an SDLC.
15
Data product lifecycles are iterative and non-deterministic. You are likely to iterate through these stages over and over.
15. Data is not the new snake-oil: Designing a data strategy for enterprises
Akshay
Conclusion
16
Here are the key “takeaways” from this presentation.
• “Data Is Not The New Snake Oil”:
• Analyzed data has been seen as a competitive advantage by many first-mover organizations.
• But many organizations have been unsuccessful in executing AI.
• Objective Function:
• A well-crafted data/ AI strategy can generate value and create competitive advantages.