2. Our Mission
Partner with extraordinary entrepreneurs
to build enduring, industry-defining
companies
3. Some people are talking about
AI as the next “platform.”
AI is not a “platform,”
It’s an enabling technology.
4. Many “X-with-ML” startup business plans
(where X is some category of software)
…but not so simple.
5. How are startup SaaS companies
actually making ML part of their
competitive advantage?
6. 1. Tell Me Something New ++++
2. Replacing Rules-Based Systems +++
3. The Ironman Suit ++
4. Replacing Humans +
4 Models (Not Equally Common Today)
7. Model #1: Tell Me Something New
Improve customer experience
Data: Collect surveys, reviews/social, transactions, call logs, etc.
ML: NLP on customer interactions
Insight + Workflow: What (concretely) makes customers happy? Loyal?
Extract useful data from cheap, frequent satellite images
ML: Computer vision to recognize, count, measure, track objects
Find use cases: government, finance, oil & gas, etc.
Improve construction efficiency
Data: Collect timesheets, geo, cost codes, orders, notes, etc.
ML: Computer vision to tag images, NLP on notes and orders
Insight + Worflow: What impacts our productivity? Causes delays?
Problem—first:
Data—first:
8. Model #1: Tell Me Something New
Questions to Consider…
Do you have advantaged access to the data?
Do you need to collect the data?
What friction is involved in collection/integration?
Can you operationalize the insights?
Can you track the changes you’re trying to bring about?
Does the executive care? What’s the ROI?
9. Model #2: Replacing Rules
Replace rules-based credit models for marketplace
lending with ML-powered ones
Recognize and block malware based on behaviors,
(not signatures)
Offer a health insurance plan, drive down costs using
“population health management” – predict issues and
intervene early
Same business model, new tech:
New business model, new tech:
10. Model #2: Replacing Rules
Questions to Consider…
Trust and accuracy of your algorithm?
Regulatory hurdles to change?
Does your accuracy matter?
Is the ML approach less operationally costly?
11. Model #3: The Ironman Suit
Make your security operations team better/faster by
first surfacing insight at scale, then predicting
investigation/response actions
Guess your replies (1/3 of responses on mobile!)
Help business analysts and quants build machine
learning models quickly and easily (how meta!)
12. Model #3: The Ironman Suit
Questions to Consider…
Value of user time, talent, superpowers?
Are you solving a scarcity problem?
Does the superpower drive buying decision?
Friction to adopt a new system?
13. Model #4: Replacing Humans
Human-skill operational tasks accessible by API
e.g. creating training sets, content moderation
“Personal Assistant” for scheduling meetings by email
Improve medication adherence in clinical trials by
replicating “directly observed therapy” with
computer vision
AI—assisted humans:
Algorithm:
14. Model #4: Replacing Humans
Questions to Consider…
Can you provide an end-to-end experience?
How is the service consumed?
Is the accuracy sufficient?
Will it fail gracefully?
Even if your core service is efficient, is sales/success?
16. AI does not enable distribution, is not a
“platform,” but it may be part of your
differentiation.
17. STILL HAVE TO BUILD A GREAT SAAS COMPANY:
THE RIGHT TEAM
INTIMATE UNDERSTANDING OF CUSTOMER
UNIQUE + COMPELLING VALUE PROPOSITION
THE RIGHT TIMING
THE LAST MILE
CAPITAL-EFFICIENT GO-TO-MARKET
DEFENSIBILITY
18. Sarah Guo
s a r a h @ g r e y l o c k . c o m
@ s a r a n o r m o u s