1. Leveraging AI for boosting your SaaS
product/business ROI’s
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
A Presentation by Vishal Sethi, Data Product Leader, Silicon Valley
AVP, Bristlecone, Founder Startupomega.com
2. Insights
Building AI Products
Integrating AI in SaaS Ecosystem
Role of ML in transforming SaaS Product
ML as a Service
Let us walk through
fundamental product design
and development thinking
behind building world class AI
products, that leverage AI to
enhance SaaS business drawing
examples from tech industry.
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
3. About Me?
20+ Years Cross Industry Contributions within Data, Analytics and AI
Product Leader, Investor, Technology evangelist, StartUp mentor
Brought a wide array of Enterprise B2B and B2C Products to life
What makes my experience unique is diversity of AI projects that I have worked on
Keynote speaker in C level summits
MSC – Data Science, MBA – strategy, Six sigma – champion, and Master in Finance management
Heartfulness meditation practitioner
Vishal.sethi@bristlecone.com
vishal@startupomega.com
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
4. Goal of this talk today?
Inspire and inform product leaders
Ideas and methodologies and technologies great products leverage
Applied examples to illustrate above
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
5. Thinking of an AI Startup?
Open AI - 5-year search growth: 99X+, Funding: $1B (Corporate Round)
◦ creating artificial general intelligence (AI) to benefit humanity.
Frame AI - 5-year search growth: 462%, Funding: $17.9M (Series B)
◦ The Voice of the Customer engine
Moveworks - 5-year search growth: 1000%, Funding: $305M (Series C)
◦ platform is able to support employees’ issues end-to-end
Cloudminds - 5-year search growth: -100%, Funding: $468.6M (Series B)
◦ open end-to-end software for robots
H2O.AI - 5-year search growth: -100%, Funding: $251.1M (Series E)
◦ open-source AI platform that allows developers to import algorithms for different use cases
Argo - Search growth status: Exploding, Funding: $3.6B (Corporate Round)
◦ first fully integrated self-driving system
Eightfold - Search growth status: Exploding, Funding: $396.8M (Series E)
◦ AI to power a suite of HR-related products aimed at retaining, training, and finding the best talent
Source: Crunchbase
Unicorns
Moorethread – 528.5M
Mobvoi – 252.8M
Scale.AI – 602M
Insider – 167 M
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
6. Rethinking products and business models
Walter Thompson and Microsoft, 2016
148 MP, 168263 scans, 300 paintings
Features –Caucasian male, 30-40 yrs, Facial
hair, hat, white collar, facing right
Rembrandt Harmenszoon van Rijn, 1606
lead white pigment and oils like linseed oil
Titian, Hendrick ter brugghen, High
viscosity and slow drying of oil paint
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
7. AI and Product-Market Economics?
Business model Digital Abundance New Possibilities
AI Is changing the
world
AI helps you capture
digital abundance
AI helps you leverage
economies of scope and
learning beyond scale.
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
8. Algorithms at the core od producst?
More data
Better
algorithms
Better
service
More
usage
Algorithm delivers customer
experience and operational
processes, and thus learn and
becomes better over a period of
time.
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
9. Why AI for SaaS and Digital Economy?
Source
Competing in the Age of AI, Marco Lansiti
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
10. Idea of an AI factory
AI Factory as Operational Foundation
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
11. Idea of an AI factory
Source - Sciencedirect.com
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
13. Quick review and thought reference?
How to augment human touchpoints with AI for
networking and learning?
How to build and leverage network effect for value?
How to create learning effects to build competitive
advantage?
Mapping business networks for value for targeted user
groups?
What networks are key to providing that service, and
what are their characteristics?
How to overcome challenges with network clustering?
Multihoming? Disintermediation?
Where are we experiencing or likely to experience
strong learning or network effects?
AI
Learning
Effects
Network
Effects
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
14. Convergence of technologies
Virtualization Cloud
Networking IoT/Device
AI
Compute cost
Storage cost
Data sourcing and
generation
Data abstraction
AI for AI
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
15. Product Planning
Business
Understanding
Data
Understanding
Data
Preparation
Modeling
Evalution
Deployment
Source - CRISP-DM model
O'Reilly
Not just magnitude but sentiments
Clear vision on focus for feature
engineering
Experimentation and appetite for failure
High touch custom productionable
architecture
Monitoring to observability
E.g. fraud monitoring for credit card burst
Learning system for physically challenged
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
16. Roadmap Execution
Data Quality
and
Standadization
Interface
Design
Prototypes and
MVP’s
Right Scope
Augmenting
product
with
Technical
Leadership
Testing AI/ML Products
UX driven design, user do not care about AI
Apple sense of design scope making
things work
Minimize black box
Create appetite for experimentation and failure
Create a ecosystem of product design
Form right technical partnership
A/B, Multivariate testing, Model
evaluation, fit and recalibrations, Data
biases and more
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
18. Netflix – Can a algorithm save a billion
dollar?
1 M $ price to build a baseline algorithm
Algorithm that saves 1B in customer retention
Recommendation system influence 80% content
geolocation, time, weather-data, device, voice recognition
etc to recommend the best and most relevant content
Viewing history, time and duration
Similar members with taste and preferences
Featuers such as Genre, categories, actors, releases
Personalization of thumb nails
Trending now
Continue watching
Because you watched a movie
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
19. Netflix and AI
Personalized
recommendations
Auto generation
of thumbnails
Location scouting Streaming quality
Movie editing
Personalized
learning to rank
Context awareness
Presentation
effects
Social
recommendations
Full page
optimization
Cold start
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
20. Netflix – A sneak peek
Strategy Metric Tactic/Project
Personalization RMSE Modal algorithm test, Voice ID,
Movie personality quick,
Language detection
Original content % of members who watch at
least 10 hours month of a
original content
Cold start merchandising test,
weekly release test, episodic
micro docs
Watching experience % of customers who watch at
least 40 hours/month
Ultra HD, customer playback
speed, shared viewing, lip synch,
algorithms (40 languages)
Interactive storytelling % of members who watch at
least one hour interactive
content per month
Support for real timing
branching prototyping, Kimmy,
Schmidt, launch
Strategy Q2 Q3 Q4 Q1
Personalization Mood algorithm
test
Voice recognition Language
detection
Movie personality
quiz
Original content Cold start system Weekly release
test
Support for
episodic micro
docs
Expert panel
forecasting
Watching
experience
Shared viewing Customer
playback speed
Automated lip
syncing in 40
languages
Ulta HD custom
mobile devices
Interactive
storytelling
Kimmy scmidt
launch
Real time branch
prototyping
Voice activated
decisions
Banderstrach#2
Source - Gibson Biddle, Productled
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
21. Netflix Data Stack a quick glance
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
Source Netflix Blog
22. A Guiding Framework
• Pain and gain
• User benefits
• MVP – HIP, HBS
• Product KPI’s
Value
Proposition
• Automation,
Assistance, and
Personalization
• ML technique
• Model metrics and UI
Problem
Framing • Engineers and Data
scientists
• Data and variable
engineering
• Inhouse or Incloud
Skills Data,
Platform
• Architecture – training
time, infrastructure as
a service
• Model best practices
• Integrations
Microservices
• Enhance UX
• Enhance model
• Enhance data
Experiment
and Iterate
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
23. ML Platform Conceptual View
Source
Towards data science
1. Data ingestion and
engineering
2. Feature store
3. Model management
and obersvability
GCP, AWS, Azure offer a
verity of services that can
be architected to enable
this in a short time
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
24. After deployment product management
Inputs to pipeline, confidence of model,
output it produces
Inputs are complete, comply
distributions, trigger alarms, model
retaining and shutdowns
GPU/TPU performance and caching. SLI’
SLO and SLA’s
Time based model retaining, Continuous
retaining
Create a ecosystem of product design
Form right technical partnership
Michelengalo, Zipine, H2O.ai, Mlflow,
Kubeflow, Seldon.io, Dask
Debugging
I/O validation
Task Speed and SlO’s
Durability and Monitoring
Frameworks
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
25. Embedding AI/ML in SaaS Ecosystem
Create MVP
don’t disrupt
Feature
Evaluation
Project
Estimation
Cloud platform
and Open stack
Teams and
Skills
Secure your
product
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
26. Driving Value with AI in SaaS ecosystem
AI’s capacity to
learn from a user’s
prior experiences
can be used to
customize
interface design in
SaaS.
Personalization
Human-machine,
machine-machine
processes may it
be repetitive or
intelligence can be
automated with AI
and mesh
technologies.
Automation
Machine learning
can help to predict
user preferences
or behavior,
product
performances,
then perhaps
trigger alerts or
actions when it
appears the user is
disengaging
Prediction
An easy and
intuitive search
reduces the
friction leaving
customer satisfied
in getting relevant
results of their
searches.
Search
AI can augment
SaaS developers
own coding
abilities by
providing the
necessary checks
that the coding is
good. This avoid
early release
crashes and bugs
while significantly
reducing release
cycle times.
Release
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
27. Some cool examples
Personalized
styling and
clothing
recommendation
Personalization
Customer support
bots are able to
login to systems
and reset
passwords based
on user request.
Automation
Uber predicts
surges in demand
to determine
pricing for peak
period and
optimizes its
margins.
Prediction
Power BI offers
voice services to
query dashboards
and reports.
Search
Alipay Tencent
analyzes the data
through machine
learning
algorithms to
inform and
automate an
expanding variety
of services.
Release
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
28. Take aways for Product Leaders
Problem Framing
Ethics
Planning and managing
project
Metrics
it’s even possible for an AI product “intervention” to move an
upstream business metric, e.g. is recommendation even good?
The scale and impact of a product over the
difficulty of product development.
AI performance tends to degrade over time
Is it a problem that should be solved? How can the solution
be abused?
Markkula Institute at the University of Santa Clara
Fault tolerance vs fault intolerant
Guardrail metrics, they ensure that the product analytics
aren’t giving the wrong signals. E.g reduce pick up time
per user vs maximize trips per user
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM
29. Trends for Product Leaders to watch
Human in the loop
MLOps and FinOps
Observability and Automation
No Code
Data Fabric
COPYRIGHT@ VISHAL SETHI, VISHAL@STARTUPOMEGA.COM