AI is a Team Sport

[CLASSIFICATION: PUBLIC/PERSONAL]
AI is a team Sport
Harvinder Atwal

// Harvinder Atwal
MoneySuperMarket
// Web
dunnhu
mby
{"previous" : "Insight Director, Tesco Clubcard"}
LLOYDS BANKING GROUP
{"previous" : "Senior Manager, Customer Strategy and Insight"}
{"Current" : “Chief Data Scientist"}
@harvindersatwal
BRITISH AIRWAYS
{"previous" : "Senior Operational Research Analyst"}
{"about" : "me"}
@gmail.com

£2bn
SAVINGS
2020 estimate total of household savings
1993 80% 13.1 million 1000+
We started life
as mortgages
2000
of UK Online
Adults visit one
of our websites
each year
MoneySuperMarket
Active users
2019
Product
Providers

“The science of getting computers to act without
being explicitly programmed” – Andrew Ng
6
AI
Algorithm
Output
Code
Rules
f(x)
Input
Data
Regular programming AI
Input
Data
Output
Rules
f(x)

Automated Claims Processing
Automated Product Descriptions
Automated
Trend
Detection
at Unilever
Automated
Image
Selection at
AirBnb &
TripAdvisor

Netflix - Automated Artwork
HireVue – Automated Interviews
Factory
Automation
at Kawasaki
Heavy
Industries

Applications aren’t in short supply
Demand Forecasting
Capacity Forecasting
Marketing automation
Supply chain management, automatic ordering
Automatic scaling of infrastructure
Document Classification
Image Annotation
Customer Service
Machine Translation
Anomaly Detection
Product Recommendation
Fraud Detection
Image Selection
Text Generation
Predictive Maintenance
Automated Pricing
Automated routing
Medical diagnosis

85% of AI projects
eventually fail to
achieve their
intended business
results*
*Source: TechRepublic

Only 53% of AI
POCs make it
into production
Source: Gartner

12
Your Machine Learning
Algorithm
What’s needed for
successful
operationalisation

model.fit(X_train, y_train)
is actually the easiest part
Data Governance
Data Quality
Data Security
Test Data Management Version Control
Access
Control
Team
Organisation
Stakeholder
Buy-in
Outcome
Measurement

Correctly coded logic will work correctly
Build Test Deploy Manage

Model development lifecycles are more
complex than software development lifecycles
Data
acquisition
& wrangling
Data Prep &
Feature
Engineering
Model
selection &
Training
Model
Evaluation
Online Inference
Offline Training Artifacts
Model Inference code,
Data pre-processing
and feature
engineering code &
configuration
management
Operationalisation
Deployment
in
application
Model
Inference
Model
Monitoring
Model & data
transformation code
Production Data
Data Repositories

AI Operationalisation doesn’t happen in a
vacuum

People and culture challenges are a bigger
barriers to AI adoption than technology

.
”It’s also striking that some of the biggest gaps between AI high
performers and others aren’t only in technical areas, such as using
complex AI-modeling techniques, but also in the human aspects of AI,
such as the alignment of senior executives around AI strategy and
adoption of standard execution processes to scale AI across an
organization.”
There really is a “playbook” for success
The state of AI in 2020
The online survey was in the field from June 9 to June 19, 2020, and garnered responses from 2,395 participants representing the full range of regions, industries,
company sizes, functional specialties, and tenures

Data-driven culture
and AI success starts
at the (very) top

Have a sound data
strategy
McKinsey - Companies with the most successful analytics programs are
2.5x more likely to report having a clear data strategy than their peers

Data Strategy balances data defence and offence
requirements
Data Security
Data Privacy
Data Cataloguing
Data Quality
Data Integration
Self-service
Faster Data
Corporate Objectives
& Key Results
Data defence Data offence
Insight
Model-Driven
Recommendations
Experimentation/
Optimisation
Reporting
Support strategic
and financial
commitments
through data
analytics
Ensure data
security, privacy,
integrity, quality,
regulatory
compliance, and
governance

A successful data strategy links organisational goals with
management of the data lifecycle
Store
Share Use
End-of-
Life
Capture
Raw Data
Process
Data Producers
Align here
Data Sharing: ETL, Exports, Data Matching, APIs
Descriptive Analytics: Reporting
Diagnostic analytics: Data exploration, Insight
Predictive Analytics: Predictive modelling
Prescriptive Analytics: Recommendations
Corporate
Strategy
KPIs,
Targets or
Key Results

Create a strategic
vision beyond proofs
of concept – Tightly
aligned with the
organisation’s
objectives

AI really is a team sport
Sorry unicorns!

HIRE DATA
SCIENTISTS
Operationalising AI is much more than
hiring data scientists
1 2 3
MONEY
FLOWS
HOARD
DATA
4

Research shows 92 percent of AI success
stories involve multi-disciplinary teams
92% of Strategic Scalers leverage and embed
multidisciplinary teams across the
organisation.
And, 72% say their employees fully
understand how AI applies to their roles
Source: “AI: Built to Scale” and produced by Accenture Strategy and Accenture Applied Intelligence, is based on a global survey of 1,500 C-level executives across 16 industries designed to
understand how companies are implementing AI across their organizations.

Multi-disciplinary teams of specialists need a
Chief AI/Data/Data Scientist/Analytics Officer
as a strategic champion

All Products (including operationalised
AI) have a lifecycle to be managed
Concept Inception Development Transition Production Retirement
Identify projects
Prioritize projects
Develop vision
Assess feasibility
Engage stakeholders
Obtain funding
Build team
Setup environment
Iterative development Final testing
Documentation
Training
Deployment
Operate
Monitor and support
Fix and enhance
Migrate
Remove

The Manager - AI Products need a Product
Manager
Understand objectives and requirements of the AI
product
Prioritise work.
Engage with the business, leadership, end users,
and data experts to scope the work, estimate
value and effort, create mock deliverables, agree
on timeline, validation and approval gates.
Set KPIs

The ML Lifecycle can be simplified to
identify remaining personas
Data
acquisition
& wrangling
Data Prep &
Feature
Engineering
Model
selection &
Training
Model
Evaluation
Online Inference
Offline Training Artifacts
Model Inference code,
Data pre-processing
and feature
engineering code &
configuration
management
Operationalisation
Deployment
in
application
Model
Inference
Model
Monitoring
Model & data
transformation code
Production Data
Data Repositories

Personas are required to support every
stage of MLOps
ML Dev pipeline
Online Inference
Offline Training
Operationalisation
Deployment
Operations
Data Pipeline
Data Engineering

Central Player - The Data Scientist
Identifies datasets and
algorithms, processes
training data, experiments,
builds and tests AI models

Core player – The Data Engineer
A software engineer
specialising in data pipelines
and platforms.
Manages data integration,
quality,
transformations/feature
stores, and platform
infrastructure

Key persona – The ML/AI Engineer
Operationalisation specialists
for ML/AI.
Responsible for the ML/AI Dev
Pipeline that enables data
scientists to train, version and
deploy production-ready
scalable models and also
Operations monitoring
feedback.

The integrator – The Developer
Integration specialists for
ML/AI.
Integrate ML artifacts, data
pipelines, APIs and predictions
into the main application

Crucial member –The Business Domain expert
Coordination and insight
specialist.
Help technical experts
understand possibilities and
turn business problems into AI
solutions. Help business teams
understand the value of
models.

The Glue - The AI architect
Strategic specialists for ML/AI.
Work with Data scientists, AI
Engineers and Developers to
plan and integrate solutions,
identify risks, choose the right
technologies and evaluate
architecture evolution

The DevOps
Engineer
Continuous delivery and
quality specialist.
Manage and monitor the
application lifecycle by
bringing together
development, testing and
operation of software
development.

The Data analyst
Insight specialist.
Coordinate with product
managers and business units
to uncover data insights to
drive the product roadmap.

#1 Establish the most valuable AI use
cases aligned to organisational goals

#2 Create the right culture - Educate and build
trust in AI applications from the top down

Lack of talent and technology are not a
barrier to successful AI Operationalisation
Gartner 2020 AI in Organisations survey: 70% of respondents state that a lack of AI talent is not a major barrier to
successful AI deployments

#3 Establish diverse, multidisciplinary
teams
AI Operationalisation is a much broader discipline than data science

Core
Personas
Data Engineer
AI Architect
Data Scientist Manager
Developer
ML/AI Engineer
Supporting
Personas
BI Developer
DevOps
Engineer
Security Expert
Domain Expert
Designer Data Analyst

#4 Have a strategy and operating model
for scaling AI
Nearly three-quarters of those that
were successful with AI said they had
a clearly-defined strategy and
operating model for scaling AI
projects in place, while only half of
the companies in proof of concept
were able to claim the same thing.

#5 Build the competency to integrate AI solutions with
your organisation's infrastructure and policies or you will
not scale
Tackle the system integration, security and privacy issues early by engaging with
Enterprise IT, compliance, and data security teams early.

#6 Keep solutions
& Robust

Not clever &

// Harvinder Atwal // Web
var current: {
companyName : "MoneySuperMarket",
position : “Chief Data Scientist"
};
var previous1: {
companyName : "Dunnhumby",
position : "Insight Director,"
+ "Tesco Clubcard"
};
var previous2: {
companyName : "Lloyds Banking Group",
position : "Senior Manager"
};
var previous3: {
companyName : "British Airways",
position : "Senior Operational Research Analyst"
};
{"about" : "me"}
var username = "harvindersatwal";
var linkedIn = "/in/" + username;
var twitter = "@" + username;
var email = username + "@gmail.com";

AI is a Team Sport

More Related Content

AI is a Team Sport

Editor's Notes