Webinar Series
Platform Approach to Scaling ML Across the
Enterprise
By Olalekan Elesin
Agenda
v Introduction
v Platform Thinking
v AI Platform Thinking
v AI Platform – Product Principles
v Now What?
Meet me …
vNigerian 🇳🇬
vRead 📖
vPlay 😎
vFootball (Soccer) ⚽ 🥅
Introduction
Usually Starts with
Alice, CPO Bob, Data Scientist
This is the
best job ever
We need to build a
recommendation
engine
That’s simple.
Let me get to work
on this
immediately
And then …
Alice, CPO Bob, Data Scientist
I don’t think I
can do this
aloneWe need a model
to optimize
conversion
Sounds like a task
I’m up for.
Wait, we just
wrapped up …
We hire more …
Alice, CPO Data Science Team
We need a model
to optimize
conversion
We need a model
…
We need to change
the world
…
And this happens…
Data Science Team
We need a model
to optimize
conversion
We need a model
…
We need to change
the world
…
…
…
We need a model
…
We are getting worked up. Spending
most of our time doing …
And this happens…
Data Science Team
We need a model
to optimize
conversion
We need a model
…
We need to change
the world
…
…
…
We need a model
…
More and more
with AI
I’m sure AI can
help us do this,
that and those
Platform Thinking
What is a Platform?
A platform is a group of technologies that are used as a
base upon which other applications, processes
or technologies are developed. – Techopedia
Platform Thinking
Platform
App 1 App 2 App n
App 1 App 2 App n
Platform Thinking
ØRemove friction from product/delivery teams by focusing on high quality,
self-service access to foundational technology.
ØCreating an ecosystem of technology and business capabilities.
ØInvest in a foundation for experimentation to ensure every delivery team
has access tools to make testing new ideas and validating learnings easy.
Source: ThoughtWorks - Art of Platform Thinking
AI Platform Thinking
AI Platform Thinking
Data Science Team
Engineering
80%
Data Science
20%
A day in the life of a
Data Scientist
AI Platform Thinking
Engineering
80%
Data Science
20%
A day in the life of a
Data Scientist
Data Science
80%
Engineering
20%
A day in the life of a
Data Scientist
What is an AI Platform?
An AI Platform is an internal product made up collection of
TOOLS, INFRASTRUCTURE, and SERVICES to enable the
development, deployment and management of
AI-driven products at scale
What is an AI Platform?
AI Platform
Initiative 1 Initiative 2 Initiative nInitiative 3 Initiative … …
By Olalekan Elesin
Scaling Use Cases with AI Platform
Pre
AI Platform
After
AI Platform
AI Platform Thinking – Long Tail
High
Priority/Business
Value Initiatives
Low Complexity and Quick win Initiatives
D
ata
Scientists
Machine Learning Enthusiasts
RiskandComplexity
AI Platform Thinking
• Increase ML use case
experimentation
• Reduce the Time-to-Market
• Increase the Number of AI Use Cases
• Increase the Number of ML
Practitioners
TimetoMarket(T2M)
No of AI Use Cases
T2M by AI Use Cases
AI Platform – Product Principles
Before We Begin
Machine Learning
Process
Data Science
Process
ML problem
framing
Data
integration
Data
preparation
and cleaning
Data collection
Data
visualization
and analysis
Monitoring and
debugging
Model
deployment
Feature
Engineering**
Model Training
and parameter
tuning
Model
Evaluation
Are
business
goals
met?
Dataaugmentation
Feature
augmentation
Prediction
Business
Problem
The Machine Learning Process
Legend
Data Platform
AI Platform
Human Judgement
AI Platform – Product Principles
Legend
People and Culture
Technology Model Training Model Serving
ML Model
ManagementFeature
Management
AutoML
Capabilities
AI Services
End-to-End Machine Learning Workflow
Trainings,SupportandWorkshops
ResearchandDevelopment
Model
Monitoring
Feature
Management
Data Labelling
Service
Now What?
Next Steps – AI Ready Enterprise
v Get Board and C-Suite Buy-in
v AI Evangelism, Community, and Education
v Create a Central Team with the mission to simplify AI Infrastructure
vMachine Learning Engineers, People Manager and Technical Product Manager
v Define AI Use Cases across Product Teams and Business Functions
v Enable Data Scientists to Focus on High Impact Initiatives
v Get Everyone Doing AI
Next Steps – AI Platform Product Management Tips
vFocus on internal customer outcomes
Example: Internal Customer Outcomes
AI Platform User Outcomes
Improve the technical knowledge required for product engineers/teams to get started with machine learning on the AI Platform
Minimize the amount of time it takes to test a trained model to production
Minimize the amount of effort required to make use of services that provide simple machine learning tasks
Minimize the effort required to provision model training and model serving infrastructure
Improve the likelihood of accessing the right dataset to create machine learning models
Improve the possibility of understanding the predicted results of trained machine learning models
Improve the likelihood to deploy machine learning workflows with continuous integration and continuous deployment pipelines
Improve the likelihood to monitor how models are performing over time
Next Steps – AI Platform Product Management Tips
vFocus on internal customer outcomes
vBuild by use case and generalize
vStrong collaboration with Data Science
• Email: elesin.olalekan@gmail.com
• Twitter: @elesinOlalekan
• LinkedIn: https://www.linkedin.com/in/elesinolalekan/
• Xing: https://www.xing.com/profile/Olalekan_Elesin
Thank you for your attending

Platform approach to scaling machine learning across the enterprise

  • 1.
    Webinar Series Platform Approachto Scaling ML Across the Enterprise By Olalekan Elesin
  • 2.
    Agenda v Introduction v PlatformThinking v AI Platform Thinking v AI Platform – Product Principles v Now What?
  • 3.
    Meet me … vNigerian🇳🇬 vRead 📖 vPlay 😎 vFootball (Soccer) ⚽ 🥅
  • 4.
  • 5.
    Usually Starts with Alice,CPO Bob, Data Scientist This is the best job ever We need to build a recommendation engine That’s simple. Let me get to work on this immediately
  • 6.
    And then … Alice,CPO Bob, Data Scientist I don’t think I can do this aloneWe need a model to optimize conversion Sounds like a task I’m up for. Wait, we just wrapped up …
  • 7.
    We hire more… Alice, CPO Data Science Team We need a model to optimize conversion We need a model … We need to change the world …
  • 8.
    And this happens… DataScience Team We need a model to optimize conversion We need a model … We need to change the world … … … We need a model … We are getting worked up. Spending most of our time doing …
  • 9.
    And this happens… DataScience Team We need a model to optimize conversion We need a model … We need to change the world … … … We need a model … More and more with AI I’m sure AI can help us do this, that and those
  • 10.
  • 11.
    What is aPlatform? A platform is a group of technologies that are used as a base upon which other applications, processes or technologies are developed. – Techopedia
  • 12.
    Platform Thinking Platform App 1App 2 App n App 1 App 2 App n
  • 13.
    Platform Thinking ØRemove frictionfrom product/delivery teams by focusing on high quality, self-service access to foundational technology. ØCreating an ecosystem of technology and business capabilities. ØInvest in a foundation for experimentation to ensure every delivery team has access tools to make testing new ideas and validating learnings easy. Source: ThoughtWorks - Art of Platform Thinking
  • 14.
  • 15.
    AI Platform Thinking DataScience Team Engineering 80% Data Science 20% A day in the life of a Data Scientist
  • 16.
    AI Platform Thinking Engineering 80% DataScience 20% A day in the life of a Data Scientist Data Science 80% Engineering 20% A day in the life of a Data Scientist
  • 17.
    What is anAI Platform? An AI Platform is an internal product made up collection of TOOLS, INFRASTRUCTURE, and SERVICES to enable the development, deployment and management of AI-driven products at scale
  • 18.
    What is anAI Platform? AI Platform Initiative 1 Initiative 2 Initiative nInitiative 3 Initiative … … By Olalekan Elesin
  • 19.
    Scaling Use Caseswith AI Platform Pre AI Platform After AI Platform
  • 20.
    AI Platform Thinking– Long Tail High Priority/Business Value Initiatives Low Complexity and Quick win Initiatives D ata Scientists Machine Learning Enthusiasts RiskandComplexity
  • 21.
    AI Platform Thinking •Increase ML use case experimentation • Reduce the Time-to-Market • Increase the Number of AI Use Cases • Increase the Number of ML Practitioners TimetoMarket(T2M) No of AI Use Cases T2M by AI Use Cases
  • 22.
    AI Platform –Product Principles
  • 23.
    Before We Begin MachineLearning Process Data Science Process
  • 24.
    ML problem framing Data integration Data preparation and cleaning Datacollection Data visualization and analysis Monitoring and debugging Model deployment Feature Engineering** Model Training and parameter tuning Model Evaluation Are business goals met? Dataaugmentation Feature augmentation Prediction Business Problem The Machine Learning Process Legend Data Platform AI Platform Human Judgement
  • 25.
    AI Platform –Product Principles Legend People and Culture Technology Model Training Model Serving ML Model ManagementFeature Management AutoML Capabilities AI Services End-to-End Machine Learning Workflow Trainings,SupportandWorkshops ResearchandDevelopment Model Monitoring Feature Management Data Labelling Service
  • 26.
  • 27.
    Next Steps –AI Ready Enterprise v Get Board and C-Suite Buy-in v AI Evangelism, Community, and Education v Create a Central Team with the mission to simplify AI Infrastructure vMachine Learning Engineers, People Manager and Technical Product Manager v Define AI Use Cases across Product Teams and Business Functions v Enable Data Scientists to Focus on High Impact Initiatives v Get Everyone Doing AI
  • 28.
    Next Steps –AI Platform Product Management Tips vFocus on internal customer outcomes
  • 29.
    Example: Internal CustomerOutcomes AI Platform User Outcomes Improve the technical knowledge required for product engineers/teams to get started with machine learning on the AI Platform Minimize the amount of time it takes to test a trained model to production Minimize the amount of effort required to make use of services that provide simple machine learning tasks Minimize the effort required to provision model training and model serving infrastructure Improve the likelihood of accessing the right dataset to create machine learning models Improve the possibility of understanding the predicted results of trained machine learning models Improve the likelihood to deploy machine learning workflows with continuous integration and continuous deployment pipelines Improve the likelihood to monitor how models are performing over time
  • 30.
    Next Steps –AI Platform Product Management Tips vFocus on internal customer outcomes vBuild by use case and generalize vStrong collaboration with Data Science
  • 31.
    • Email: elesin.olalekan@gmail.com •Twitter: @elesinOlalekan • LinkedIn: https://www.linkedin.com/in/elesinolalekan/ • Xing: https://www.xing.com/profile/Olalekan_Elesin Thank you for your attending