Assumption Mapping: How to Find and Test Your Riskiest Bets
Every roadmap item rests on beliefs nobody has checked. Assumption mapping sorts them by importance and evidence, so you test the one that could sink the bet before you build.
Share
Assumption mapping is a team exercise that lists everything that must be true for a product idea to work, then plots each belief on a 2x2 grid by how important it is and how much evidence supports it. Beliefs that are important but unproven are your real risks. Test those first, as cheaply as possible, before anyone writes code. David J. Bland popularised the method in Testing Business Ideas.
Key takeaways
The two axes are importance (would the idea fail if this is wrong?) and evidence (do we have observable proof?). Test the important, no-evidence quadrant first (Strategyzer, 2020).
Map across desirability, feasibility and viability. Most teams over-test feasibility and under-test whether anyone wants the thing.
Pick 3-5 assumptions per session. Testing everything spreads the team thin.
Rank evidence by cost to the customer: what people say < what people do < money or a signature.
The stakes are real: 80% of features in the average software product are rarely or never used (Pendo, 2019).
What is assumption mapping?
An assumption is anything your plan treats as true without proof: customers have this problem, they'll pay this much, the API can handle this load. Mapping makes those beliefs explicit and ranks them, so discovery time goes to the ones that could sink the bet.
The method is most closely associated with David J. Bland. He describes it in Testing Business Ideas, co-written with Alexander Osterwalder and published by Wiley in November 2019 (Wiley, 2019). Strategyzer published an excerpt in August 2020 that sets out the full exercise (Strategyzer, 2020).
Teresa Torres uses the same 2x2 in her continuous discovery work. Her core advice: stop testing whole ideas and test the assumptions underneath them instead (Product Talk, 2023). An idea can fail for a dozen reasons. An assumption fails for one, which makes the result readable.
Two axes. Evidence runs left to right, from "have evidence" to "no evidence". Importance runs bottom to top, from "unimportant" to "important". Bland's test for importance: which hypothesis, if proven wrong, would cause the business idea to fail?
Have evidence
No evidence
Important
Build on it. Keep checking it hasn't changed.
Test now. Your riskiest assumptions live here.
Unimportant
Ignore. True and low-stakes.
Park. Revisit if the plan changes.
Strategyzer's version tells teams to focus near-term experiments on the top-right quadrant: important hypotheses with the least evidence (Strategyzer, 2020).
You can draw this on a whiteboard, in Miro, or in the free assumption map template, which plots notes by impact and evidence and tells you what to test first. No signup needed.
Which types of assumptions should you map?
Different authors slice risk differently. The labels matter less than covering all of them, as Torres notes.
Bland and Osterwalder
Marty Cagan's four risks
Teresa Torres
Question to ask
Desirability
Value
Desirability
Does the customer want this enough to change behaviour?
Feasibility
Feasibility
Feasibility
Can we build and run it with our time, skills and technology?
Viability
Business viability
Viability
Does it work for the business: pricing, sales, legal, brand?
—
Usability
Usability
Can people figure out how to use it?
Adaptability
—
—
Can it survive changes in the market or environment?
Cagan also assigns owners: the product manager owns value and viability, the designer owns usability, and the lead engineer owns feasibility (SVPG, 2017). That is a useful check on who should be in the room.
Bland suggests mapping desirability first, then feasibility, viability and adaptability. Engineering-led teams tend to do the reverse, because "can we build it?" feels answerable. Desirability is where products more often die: in CB Insights' 2026 study of 431 VC-backed shutdowns, 43% cited poor product-market fit (CB Insights, 2026).
How do you run an assumption mapping session?
Plan for 60-90 minutes with 4-8 people who build, sell and support the product.
Name one bet. "Auto-create Jira tickets from customer calls", not "Q3 roadmap". One map per bet.
Write silently for 10 minutes. Each person writes one belief per sticky note, using "We believe that…". Cover every category in the table above.
Rewrite vague notes as testable ones. Add who, what and how much (see the next section).
Place each note on the evidence axis first. Ask: what observable evidence do we have, and how strong is it? An opinion from the CEO counts as no evidence.
Then place it on the importance axis. Ask Bland's question: if this is wrong, does the idea fail?
Argue about placement. Disagreement is the useful output. It shows where the team is guessing.
Circle 3-5 notes in the top right. These are the leap-of-faith assumptions for this bet.
Design one test per circled note before anyone leaves, with an owner, a deadline and a pass threshold written down in advance.
Book the re-map. Two to four weeks later, move notes based on results and decide what happens to the bet.
How do you write a testable assumption?
An assumption that can't fail can't teach you anything. Make it specific enough that a result could prove it wrong.
Vague
Testable
Users will like it.
At least 6 of 10 PMs we interview describe spending 2+ hours a week turning call notes into tickets.
People will pay.
At least 3 of 20 trial accounts accept a €49/month add-on after 14 days.
It's technically possible.
Our extraction finds 80%+ of action items in 30 real transcripts.
It's easy to use.
4 of 5 first-time users create a ticket without help in under 3 minutes.
If a belief feels too big to test, split it into smaller beliefs that must all be true. If any one fails, the big one fails too, and it's far cheaper to find that out on a small piece.
How do you test the riskiest assumption?
Use the cheapest test that could prove you wrong. Torres suggests starting with the smallest assumption test you can complete in the next hour (Product Talk, 2023).
Rank the evidence you collect
Strength
Evidence type
Example
Weakest
What people say
"I'd definitely use that."
Early signal
Directional data
Survey answers, search volume, interview stories about past behaviour
Strong
What people do
Clicks on a fake door, completed tasks in a prototype, repeat use
Strongest
Money or a signature
Pre-orders, paid pilots, letters of intent
Match the test to the risk
Desirability: customer interviews about recent behaviour, a fake door test, a concierge MVP.
Feasibility: a time-boxed technical spike, a Wizard of Oz test, conversations with partners or suppliers.
Viability: a pricing survey, a mock sale, a letter of intent.
We believe that [assumption].
To check, we will [test] with [who, how many] by [date].
We'll measure [metric].
We are right if [threshold]. If not, we will [decision].
Writing the threshold first stops the team from moving the goalposts when the result comes in.
Worked example: an AI call-notes feature (illustrative)
Say a six-person B2B SaaS team is considering a feature that turns recorded customer calls into Jira tickets automatically. They map 12 assumptions and score each from 1 to 5 on importance and evidence. To rank them, they use a simple rule: risk = importance × (6 − evidence), so an important belief with no evidence scores 25 and a trivial, proven one scores 1.
#
We believe that…
Type
Importance
Evidence
Risk
A1
PMs spend 2+ hours a week turning call notes into tickets
Desirability
5
1
25
A2
PMs will accept AI-created tickets without rewriting most of them
Desirability
5
1
25
A3
Customers' security teams allow call transcripts to be processed by our AI vendor
Viability
5
1
25
A4
Customers will pay a €49/month add-on
Viability
5
2
20
A5
Transcription works for non-native English speakers
Feasibility
4
1
20
A6
We can extract 80%+ of action items correctly
Feasibility
5
3
15
A7
PMs can route tickets to the right Jira project without training
Usability
3
1
15
A8
The feature reduces churn in accounts with 5+ PMs
Viability
3
1
15
A9
Most customers already record their calls
Desirability
4
3
12
A10
Engineers prefer AI-drafted tickets to PM-written ones
Desirability
2
1
10
A11
Users notice the feature in the call view
Usability
2
2
8
A12
Jira's API supports the ticket volume we need
Feasibility
3
5
3
The top three (A1, A2, A3) are all in the top-right quadrant. Notice that the one the engineers were worried about, A12, is the least risky item on the map.
Tests, run in one week:
A1, eight interviews about last week's calls. Pass if 5 of 8 describe 2+ hours. Result: 6 of 8. Supported.
A2, a Wizard of Oz test. A PM on the team hand-drafts tickets from 5 customers' real calls. Pass if 60% go into Jira without major rewrites. Result: 41%. Contradicted.
A3, questions to 6 customers' security contacts. Pass if 4 of 6 would approve. Result: 2 of 6. Contradicted.
Decision: the problem is real, but "auto-create" is the wrong solution and the data-processing rules block most accounts. The team reframes the bet as suggested tickets that a PM reviews, offered only to customers whose policies allow it, then re-maps in two weeks. One week of tests replaced a build the team had estimated at six weeks.
Assumption mapping template
Copy this into a doc or spreadsheet, or use the free interactive assumption map tool.
Bet: [one sentence]
Date mapped: [date] Next re-map: [date]
| # | We believe that... | Type (D/F/V/U/E) | Importance 1-5 | Evidence 1-5 | Risk = I x (6 - E) | Test | Pass if... | Owner | Due | Result |
|---|--------------------|------------------|----------------|--------------|--------------------|------|------------|-------|-----|--------|
| 1 | | | | | | | | | | |
| 2 | | | | | | | | | | |
| 3 | | | | | | | | | | |
Top 3-5 to test this cycle: #_, #_, #_
Decision after results: continue / change / stop
What mistakes make an assumption map useless?
Mapping only feasibility. The engineering questions feel concrete, so they crowd out desirability and viability.
Untestable notes. "Users want a better experience" can't fail. Rewrite it or drop it.
Counting opinions as evidence. A confident stakeholder is not data. Neither is a demo compliment.
Testing everything. Three to five per cycle. The rest wait.
Setting the pass threshold after the test. You'll always find a way to call it a success.
Running it once. A map drawn at kickoff is wrong within weeks. Results land in a doc and the sticky notes never move.
Mapping the solution only. Map the problem and the customer too. "The customer has this problem" is usually the most important assumption of all.
When a full map is overkill
Small, reversible changes don't need a workshop. If a change takes a day to ship and a minute to roll back, ship it and watch the data. Save the workshop for bets that cost weeks of build time, touch pricing, or are hard to undo.
How does an assumption map connect to product-market fit?
Mapping assumptions is how you de-risk a bet before you build. Product-market fit measures whether the bets you shipped added up to something customers need. The desirability assumptions you test early, such as who has the problem and how badly, become the segment you later survey with the Sean Ellis question. See our guides to AI product-market fit and the product-market fit survey.
Most assumption maps go stale because the results live somewhere else. In Specky, each assumption on the board links to the experiments that tested it and records the strength of the evidence, from what people say to money or a signature. Any bet that rests on a contradicted belief shows it, so the roadmap moves with what you've learned.
Frequently asked questions
What is assumption mapping and risk validation?
Assumption mapping is the step where you list and rank the beliefs behind a plan by importance and evidence. Risk validation is what follows: running cheap tests on the important, unproven beliefs until each is supported or contradicted. Together they make sure build time goes to bets that have survived an honest attempt to break them.
What is the difference between an assumption and a hypothesis?
An assumption is something you're treating as true for now while accepting the risk that it isn't. A hypothesis is an assumption you've decided to test, written precisely enough that a result could prove it wrong, with a metric and a pass threshold. Every hypothesis starts as an assumption; most assumptions never need to become hypotheses.
How many assumptions should you test at once?
Three to five per cycle. A map can easily surface a dozen or more assumptions; Torres's own example lists ten for three ideas and notes dozens more are possible. Only a handful will sit in the important, no-evidence quadrant. Testing more than five at once spreads a small team thin and delays the answers that matter most.
Who should take part in the workshop?
The people who carry each risk: a product manager for value and viability, a designer for usability and an engineer for feasibility, as Marty Cagan's model assigns them. Add someone from sales or support who hears customers every week. Four to eight people is enough; larger groups produce more notes and less argument.
How often should you update an assumption map?
Every time a test result lands, and at least every two to four weeks while the bet is active. Move notes as evidence comes in, and re-score importance when the plan changes. A map that hasn't changed in a month is either finished or forgotten.
GDPR-aligned. EU workspace data is stored in Frankfurt, Germany. GDPR and EU data Built in Vienna, Austria. A product of influence solutions fm GmbH (FN 675374y). Imprint 14-day trial. Card required, and nothing is charged until day 14. 30-day refund after subscribing. See pricing