Building AI-first products
Shift AI
Virtual Event, April 2020
Title text here
Subtitle here
AI applications in
our industry
Highly confidential – Do not copy
Generating Alpha and
Returns
Digitize/Automate
Workflow
for Operational Efficiency
Improving Content
Distribution
Managing Risk
Core opportunities in AI within investment management
Examples:
Big data and alternative data offers
up a world of possibilities for
generating additional alpha, and
not only for equities (fixed income
and real estate opportunities are
emerging)
Understand investor preferences in
real time; more effectively manage
and tailor content; and deliver it
with greater agility and speed to
clients.
Use AI and advanced automation
to continuously improve efficiency
of operations and transform
traditional cost centers into AI-
enabled “as a service” offerings.
AI can bolster compliance and risk
management functions by automating
data analysis, reducing admin
activities and refocus employees’
time to higher value-add activities.
AI-enabled intelligent
dashboards: adapt to every
interaction that advisors have with their
customers to make critical information
accessible on-demand
*There are no assurances that these outcomes will be achieved and actual events may differ materially
*Based on (with our additions): Artificial Intelligence, The Next Frontier for Investment Management Firms, Deloitte, 2019
BUILDING AI-FIRST PRODUCTS
Compliance: Using AI to monitor and
detect suspicious trading activity
Legal: Using AI to review legal
contracts for impacts related to
regulatory and market changes
Unstructured data: Widespread use
of NLP for extracting signals from
news, regulatory filings, and social
media
Alternative: Using geo-location data
to make smarter investment decisions in
real estate
AI Product Development
Highly confidential – Do not copy
Deep Learning Timelines
BUILDING AI-FIRST PRODUCTS
1974
Backpropagation
1943
Neural nets
1960
Adaline
1980
Self-organizing
map
1940
Dark era
1958
Perceptron
1969
XOR problem
1980
Neocogitron
1982
Hopefield
network
1985
Boltzmann
machine
1986
Multilayer
perception
1986
RNNs
1986
Restricted
Boltzmann
machine
1990
LeNet
1997
LSTMs
1997
Bidirectional
RNN
2006
Deep Belief
Networks -
Pretraining
2006
Deep
Boltzmann
machines
2012
Dropout
2014
GANs
2017
Capsule networks
Highly confidential – Do not copy
BUILDING AI-FIRST PRODUCTS
Embedding expertise
Data science becomes BAU
2012 – 2015
Entrenching data & analytics
Establish Data Science consulting practice.
2017 onwards
AI-first product development
Build out design, product and AI teams
2019
Founding of data
Science team
Data support and services
Scaling
AI ability
Fully integrated
AI products
Develop predictive
modelling capability
Pivot from data science to
AI-products
AI
20162012-2015
Competitiveadvantage
Time
AI product focusData science focus
Evolving focus from data science to AI-first products
Title text here
Subtitle here
What does it mean for a product
to be AI-first?
Title text here
Subtitle here
“If I had asked people what they wanted,
they would have said faster horses.”
-Henry Ford
Highly confidential – Do not copy
AI product life cycle
DESIGN
& BUILD
VALIDATE
IDEATE
RESEARCH Proof of Concept Ç
Release 1:
First Iteration of
Product (MVP)
Release 2: Incremental
improvements based on
user feedback and model
improvement
BUILDING AI-FIRST PRODUCTS
Highly confidential – Do not copyHighly confidential – Do not copy
AvailableProductMetrics
Product Lifecycle
BUILDING AI-FIRST PRODUCTS
Dissecting the
user process
Qualitative feedback
and user interviews
Product analytics
for user behavior
Highly confidential – Do not copy
Moving forward, ensure the right mindset
and expectations are in place
Stage 1:
Ideation & Proof of Concept (POC)
Stage 2:
Minimal viable product (MVP)
Stage 3:
Generate value
Stage 4:
Business as usual
Explore
Create
Scale
BUILDING AI-FIRST PRODUCTS
Title text here
Subtitle here
AI product types and
the teams behind them
Highly confidential – Do not copyHighly confidential – Do not copy
AI consulting & team augmentation support (not
products)
Full-stack AI Products
Model/API/Data Products
AI products vary in size
and form
Each tier requires a different
product team composition
BUILDING AI-FIRST PRODUCTS
Highly confidential – Do not copy
BUILDING AI-FIRST PRODUCTS
AI Product Team
AI scientists
(PhD in CS, NLP, CV,
Stats, etc.)
AI engineer
(advanced degree
in CS.)
Strategy, BA,
Market
Research
AI Product
Manager
Data Engineers
(preparing /
processing SQL and
NoSQL data for AI)
UX/UI Designers
Back-end dev
Front-end dev
The full cast behind AI-first products
Stakeholders,
Users, Customers
Strategy,
Business Analyst
Highly confidential – Do not copy
BUILDING AI-FIRST PRODUCTS
AI Product Team
AI scientists
(PhD in CS, NLP, CV,
Stats, etc.)
AI consulting & team
augmentation
Full-stack AI Products
Models & APIs
AI consulting & team augmentation
Highly confidential – Do not copy
BUILDING AI-FIRST PRODUCTS
AI Product Team
AI scientists
(PhD in CS, NLP, CV,
Stats, etc.)
AI engineer
(advanced degree
in CS.)
AI Product
Manager
Data Engineers
(preparing /
processing SQL and
NoSQL data for AI)
AI consulting & team
augmentation
Full-stack AI Products
Models & APIs
Model/API/Data Products
Strategy,
Business Analyst
Back-end dev
Highly confidential – Do not copy
BUILDING AI-FIRST PRODUCTS
AI Product Team
AI scientists
(PhD in CS, NLP, CV,
Stats, etc.)
AI engineer
(advanced degree
in CS.)
Strategy,
Business Analyst
AI Product
Manager
Data Engineers
(preparing /
processing SQL and
NoSQL data for AI)
UX/UI Designers
Back-end dev
Front-end dev
AI consulting & team
augmentation
Full-stack AI Products
Models & APIs
Full-stack AI Products
Identifying the right
AI product manager
Highly confidential – Do not copy
What makes a great
AI Product Manager?
Highly confidential – Do not copy
Subject matter
expertise
AI technical
expertise
General
product
management
skills
Great AI Product Manager
BUILDING AI-FIRST PRODUCTS
Highly confidential – Do not copyHighly confidential – Do not copy
AI Product Team
Stakeholders
AItechnicalexpertise
Subject matter expertise
BUILDING AI-FIRST PRODUCTS
Highly confidential – Do not copyHighly confidential – Do not copy
AI Product Team
Stakeholders
AItechnicalexpertise
Subject matter expertise
BUILDING AI-FIRST PRODUCTS
Highly confidential – Do not copy
Identifying the right AI-first product manager
Highly confidential – Do not copy
Start from a position of strength and are
ready to ship AI-first products.
Start leading products from a PM perspective
while cultivating business domain and AI
experience on-the-job.
Potential growth into a great
AI-first PMs.
Start with deep technical knowledge and can
transform into a AI-first PM quickly. The risk due
to deep solution expertise is a focus on solution
space (& "cool ideas”) rather than problem
space (& real user pains)
Start as strong team leaders and can grow
into AI-first PMs by shadowing AI-first
and/or general PMs over multiple
products
General PM
Skills
Subject
matter
AI
experience
Subject
matter
AI
experience
General
PM Skills
General PM
Skills
Subject
matter
AI
experience
General PM
Skills
Subject
matter
AI
experience
AI-first PM General PM Technical PM-to-be Business/Generalist PM-to-be
BUILDING AI-FIRST PRODUCTS
Thanks!
Ehsan Yousefzadeh
ehsan.yousefzadeh@aig.com

Shift AI 2020: Building AI-first Products - Ehsan Yousefzadeh (AIG Investments)

  • 1.
    Building AI-first products ShiftAI Virtual Event, April 2020
  • 2.
    Title text here Subtitlehere AI applications in our industry
  • 3.
    Highly confidential –Do not copy Generating Alpha and Returns Digitize/Automate Workflow for Operational Efficiency Improving Content Distribution Managing Risk Core opportunities in AI within investment management Examples: Big data and alternative data offers up a world of possibilities for generating additional alpha, and not only for equities (fixed income and real estate opportunities are emerging) Understand investor preferences in real time; more effectively manage and tailor content; and deliver it with greater agility and speed to clients. Use AI and advanced automation to continuously improve efficiency of operations and transform traditional cost centers into AI- enabled “as a service” offerings. AI can bolster compliance and risk management functions by automating data analysis, reducing admin activities and refocus employees’ time to higher value-add activities. AI-enabled intelligent dashboards: adapt to every interaction that advisors have with their customers to make critical information accessible on-demand *There are no assurances that these outcomes will be achieved and actual events may differ materially *Based on (with our additions): Artificial Intelligence, The Next Frontier for Investment Management Firms, Deloitte, 2019 BUILDING AI-FIRST PRODUCTS Compliance: Using AI to monitor and detect suspicious trading activity Legal: Using AI to review legal contracts for impacts related to regulatory and market changes Unstructured data: Widespread use of NLP for extracting signals from news, regulatory filings, and social media Alternative: Using geo-location data to make smarter investment decisions in real estate
  • 4.
  • 5.
    Highly confidential –Do not copy Deep Learning Timelines BUILDING AI-FIRST PRODUCTS 1974 Backpropagation 1943 Neural nets 1960 Adaline 1980 Self-organizing map 1940 Dark era 1958 Perceptron 1969 XOR problem 1980 Neocogitron 1982 Hopefield network 1985 Boltzmann machine 1986 Multilayer perception 1986 RNNs 1986 Restricted Boltzmann machine 1990 LeNet 1997 LSTMs 1997 Bidirectional RNN 2006 Deep Belief Networks - Pretraining 2006 Deep Boltzmann machines 2012 Dropout 2014 GANs 2017 Capsule networks
  • 6.
    Highly confidential –Do not copy BUILDING AI-FIRST PRODUCTS Embedding expertise Data science becomes BAU 2012 – 2015 Entrenching data & analytics Establish Data Science consulting practice. 2017 onwards AI-first product development Build out design, product and AI teams 2019 Founding of data Science team Data support and services Scaling AI ability Fully integrated AI products Develop predictive modelling capability Pivot from data science to AI-products AI 20162012-2015 Competitiveadvantage Time AI product focusData science focus Evolving focus from data science to AI-first products
  • 7.
    Title text here Subtitlehere What does it mean for a product to be AI-first?
  • 8.
    Title text here Subtitlehere “If I had asked people what they wanted, they would have said faster horses.” -Henry Ford
  • 9.
    Highly confidential –Do not copy AI product life cycle DESIGN & BUILD VALIDATE IDEATE RESEARCH Proof of Concept Ç Release 1: First Iteration of Product (MVP) Release 2: Incremental improvements based on user feedback and model improvement BUILDING AI-FIRST PRODUCTS
  • 10.
    Highly confidential –Do not copyHighly confidential – Do not copy AvailableProductMetrics Product Lifecycle BUILDING AI-FIRST PRODUCTS Dissecting the user process Qualitative feedback and user interviews Product analytics for user behavior
  • 11.
    Highly confidential –Do not copy Moving forward, ensure the right mindset and expectations are in place Stage 1: Ideation & Proof of Concept (POC) Stage 2: Minimal viable product (MVP) Stage 3: Generate value Stage 4: Business as usual Explore Create Scale BUILDING AI-FIRST PRODUCTS
  • 12.
    Title text here Subtitlehere AI product types and the teams behind them
  • 13.
    Highly confidential –Do not copyHighly confidential – Do not copy AI consulting & team augmentation support (not products) Full-stack AI Products Model/API/Data Products AI products vary in size and form Each tier requires a different product team composition BUILDING AI-FIRST PRODUCTS
  • 14.
    Highly confidential –Do not copy BUILDING AI-FIRST PRODUCTS AI Product Team AI scientists (PhD in CS, NLP, CV, Stats, etc.) AI engineer (advanced degree in CS.) Strategy, BA, Market Research AI Product Manager Data Engineers (preparing / processing SQL and NoSQL data for AI) UX/UI Designers Back-end dev Front-end dev The full cast behind AI-first products Stakeholders, Users, Customers Strategy, Business Analyst
  • 15.
    Highly confidential –Do not copy BUILDING AI-FIRST PRODUCTS AI Product Team AI scientists (PhD in CS, NLP, CV, Stats, etc.) AI consulting & team augmentation Full-stack AI Products Models & APIs AI consulting & team augmentation
  • 16.
    Highly confidential –Do not copy BUILDING AI-FIRST PRODUCTS AI Product Team AI scientists (PhD in CS, NLP, CV, Stats, etc.) AI engineer (advanced degree in CS.) AI Product Manager Data Engineers (preparing / processing SQL and NoSQL data for AI) AI consulting & team augmentation Full-stack AI Products Models & APIs Model/API/Data Products Strategy, Business Analyst Back-end dev
  • 17.
    Highly confidential –Do not copy BUILDING AI-FIRST PRODUCTS AI Product Team AI scientists (PhD in CS, NLP, CV, Stats, etc.) AI engineer (advanced degree in CS.) Strategy, Business Analyst AI Product Manager Data Engineers (preparing / processing SQL and NoSQL data for AI) UX/UI Designers Back-end dev Front-end dev AI consulting & team augmentation Full-stack AI Products Models & APIs Full-stack AI Products
  • 18.
    Identifying the right AIproduct manager
  • 19.
    Highly confidential –Do not copy What makes a great AI Product Manager? Highly confidential – Do not copy Subject matter expertise AI technical expertise General product management skills Great AI Product Manager BUILDING AI-FIRST PRODUCTS
  • 20.
    Highly confidential –Do not copyHighly confidential – Do not copy AI Product Team Stakeholders AItechnicalexpertise Subject matter expertise BUILDING AI-FIRST PRODUCTS
  • 21.
    Highly confidential –Do not copyHighly confidential – Do not copy AI Product Team Stakeholders AItechnicalexpertise Subject matter expertise BUILDING AI-FIRST PRODUCTS
  • 22.
    Highly confidential –Do not copy Identifying the right AI-first product manager Highly confidential – Do not copy Start from a position of strength and are ready to ship AI-first products. Start leading products from a PM perspective while cultivating business domain and AI experience on-the-job. Potential growth into a great AI-first PMs. Start with deep technical knowledge and can transform into a AI-first PM quickly. The risk due to deep solution expertise is a focus on solution space (& "cool ideas”) rather than problem space (& real user pains) Start as strong team leaders and can grow into AI-first PMs by shadowing AI-first and/or general PMs over multiple products General PM Skills Subject matter AI experience Subject matter AI experience General PM Skills General PM Skills Subject matter AI experience General PM Skills Subject matter AI experience AI-first PM General PM Technical PM-to-be Business/Generalist PM-to-be BUILDING AI-FIRST PRODUCTS
  • 23.