One page tool · AI Startup Canvas
Show whether AI is the reason your venture wins
The AI Startup Canvas tests an AI venture idea on one sheet. Problem, value, mechanism, advantage, evidence.
AI Startup Canvas
BUSINESS DESIGN LABS
1 CUSTOMER PROBLEM
What problem are we solving, and for whom?
CUSTOMER
Who has the problem?
JOB OR PROBLEM
What are they trying to do? What gets in the way?
SIZE OF PROBLEM
How often does it happen? How serious is it?
2 VALUE CREATED
WHAT THE CUSTOMER GETS
What value does the customer get from this solution?
HOW AI CREATES VALUE · What does AI use and do to deliver the value?
WHAT AI USES
What information does the AI need?
WHAT AI DOES
What does AI do to solve the problem?
repeats
WHAT GETS BETTER
What improves through use or feedback? “Nothing yet” is a valid answer.
CUSTOMER OUTCOME
What changes because the problem is solved?
3 ADVANTAGE
WHY THIS IS HARD TO COPY
Why would competitors struggle to copy or replace this?
6 EVIDENCE AND ASSUMPTIONS
EVIDENCE
What supports this?
ASSUMPTIONS
What still needs to be true?
4 AI MODEL
What model will power the solution, and why?
MODEL CHOICE
Commercial API · open weight · specialist · build our own · mix. Which, and why?
5 DATA · Where will the AI get the data it needs?
OURS
CUSTOMER
PARTNER
PUBLIC
MISSING
Six areas. One sheet. The numbers follow the order you fill it.
Get the printable A1 versionThe signs you need it
Three moments founders recognise
01
The demo impresses everyone.
Nobody buys.
02
AI is in the pitch deck.
It is not in the mechanism.
03
An investor asks what is hard to copy.
The room goes quiet.
What it is
The AI Startup Canvas is a one page tool that tests an AI venture idea. It shows the problem, the value, the mechanism and the evidence together.
LEAN CANVAS ASKS
How does the business work?
THIS CANVAS ASKS
How does the AI work?
Use this one first when AI is the claim.
What it reveals
1
Whether a real problem sits under the idea, or only a capability.
2
Whether the value belongs to the customer, or only to the demo.
3
Whether anything compounds with use. “Nothing yet” is a valid answer.
4
Whether the model and data choices work today.
5
Which assumption carries the risk.
The six areas
One question per area
1
Customer Problem
What problem are we solving, and for whom?
A good answer names a person, a job, and how often the problem happens.
2
Value Created
What does the customer get, and how does the AI create it?
This area holds the mechanism. What the AI uses. What the AI does. What gets better with use.
3
Advantage
Why would competitors struggle to copy this?
A good answer points to data, workflow position, or access. Not to the model.
4
AI Model
What model will power the solution, and why?
Commercial API, open weight, specialist, build our own, or a mix.
5
Data
Where will the AI get the data it needs?
Ours, customer, partner, public, missing. Mark what is missing.
6
Evidence and Assumptions
What supports this, and what still needs to be true?
Evidence is what you have seen. An assumption is what you hope.
How to use it
Six passes. One question at a time.
Allow ninety minutes.
PASS 1 · AREA 1
Problem
Name the customer, the job and the size.
Decide: is this frequent and serious enough to pay for?
PASS 2 · AREA 2
Value
Name what the customer gets and what changes for them.
Decide: is the outcome in the customer’s words?
PASS 3 · AREA 2
Mechanism
Name what the AI uses and what it does. Then name what gets better.
Decide: does anything compound?
PASS 4 · AREA 3
Advantage
Name why this is hard to copy.
Decide: is the advantage real, or is it speed alone?
PASS 5 · AREAS 4 AND 5
Build
Choose the model approach. Map the data. Mark what is missing.
Decide: can this run with the data we can get?
PASS 6 · AREA 6
Evidence
Split what is supported from what is assumed.
Decide: which assumption do we test first?
EXIT
One test. One owner. One date.
Worked example
Abridge, before and after
Abridge turns a doctor and patient conversation into a clinical note. It is now one of the most funded AI ventures in healthcare. The useful part is what its canvas looked like before any of that.
AREA
CANVAS A · 2019
Seed stage
CANVAS B · TODAY
At scale
1 Customer Problem
2019Patients forget what the doctor said. Clinicians finish notes at home.
TodayClinicians in large health systems. Around two hours of admin for every hour of care.
2 Value Created
2019
Gets: a plain summary of the visit.
AI uses: the recorded conversation.
AI does: transcribes it and pulls out the medical parts.
Outcome: the patient remembers the plan.
Gets better: every recording adds real medical speech to train on.
Today
Gets: a finished note by the end of the visit.
AI uses: the conversation plus the record context.
AI does: drafts the note live and links each line to the transcript.
Outcome: the day ends at the clinic.
Gets better: every clinician edit sharpens the next draft.
3 Advantage
2019A practising cardiologist founded it. Research partnership with a university hospital. Access others did not have.
TodayBuilt into the record system. Top independent ranking gates procurement. Hundreds of health systems already live.
4 AI Model
2019Speech and language models built for medical conversation.
TodaySpecialist models for medical speech. The mix is not public.
5 Data
2019Ours: recordings from app users. Partner: one health system. Missing: notes good enough for the medical record.
TodayOurs: encounters and corrections. Customer: audio and records. Partner: record system and payers. Missing: outcome data by specialty.
6 Evidence and Assumptions
2019
Evidence: thousands of patients at one health system used a free app.
Assumptions: that doctors would pay. That the note could enter the record. That a patient app reaches an enterprise buyer.
Today
Evidence: a 90 day children’s hospital pilot cut documentation effort by 79 per cent. An independent study measured 13 minutes saved per clinician per day.
Assumption: those minutes are worth enterprise pricing.
The lesson
Canvas A holds access, usage and one partner. It holds no revenue and no proof. That is what a strong early canvas looks like.
A founder who fills Evidence with money raised has filled the wrong box.
Who uses it
Three users. One decision each.
Founders
Turn a product idea into a claim you can test.
Find the assumption that carries the risk.
Venture builders and cohorts
Compare every idea on the same six questions.
Decide which teams get the next block of time.
Investors
Ask the mechanism question in diligence.
See whether the advantage comes from data, position, or speed alone.
How it fails
Four common misuses
Capability first
The team starts at the model and works back to a problem. Start at area 1.
Outcome in product terms
“Users get an AI dashboard” is a feature. “The day ends at the clinic” is an outcome.
“It learns” with no path
If no one can name the feedback source, nothing compounds. Write “nothing yet”.
Evidence mixed with assumption
Funding is not evidence of value. Keep the two columns apart.
FREE DOWNLOAD
Get the AI startup canvas
WORK EMAIL
Plus a fortnightly email worth opening. Practical insights, no filler.