Value Hypothesis
Definition
Value Hypothesis
A value hypothesis is a single testable statement that a named group of customers will get enough value from a product to pay for it. It is written so evidence can prove it wrong, which is what separates it from a forecast.
The term belongs to the lean startup method popularised by Eric Ries, and to the investor Andy Rachleff, who paired it with a growth hypothesis.
The value hypothesis asks whether the product works for anyone at all. The growth hypothesis asks how new users keep arriving. Teams that skip the first and fund the second end up scaling something nobody wanted.
Steve Blank’s 2010 first-principles essay defines a startup as an organisation searching for a repeatable and scalable business model, assembled from hypotheses the founder has not yet tested.
Key takeaways
- A value hypothesis names one customer, one problem and one promise, in a form evidence can disprove.
- It precedes the growth hypothesis, because scaling an unwanted product only burns money faster.
- Every hypothesis needs a pre-agreed threshold that counts as confirmation and one that counts as failure.
- Invalidation is a result, not a setback; it should trigger a documented pivot or a clean stop.
How it works
Writing the statement is the discipline. A usable value hypothesis names the customer segment, the problem, the promised change and the observable behaviour that would settle the question, all before any build work starts.
The shape is easiest to see laid out in parts.
| Part of the statement | Worked example |
|---|---|
| Customer | freelance bookkeepers carrying 5–20 small clients |
| Problem | month-end reconciliation eats three evenings every cycle |
| Promise | automatic bank matching brings that down to under an hour |
| Confirming evidence | 40% of trial users finish a reconciliation inside an hour in week one |
| Disconfirming evidence | fewer than 10% complete a single reconciliation at all |
Thresholds go in writing before the test runs. Agreeing the number afterwards — once the data is already in — is how a team talks itself into a result it wanted.
The experiment should be the cheapest thing that could still produce a wrong answer. A landing page, a concierge service run by hand, or a single-feature build will each do the job if the behaviour they measure is real behaviour.
Interviews on their own rarely settle anything. People predict their own behaviour badly, so a statement is only tested once someone does something costly: paying, switching tools, or abandoning a workaround.
Leading indicators carry more signal than totals. Repeat use within a week, willingness to pay before the feature ships, and unprompted referrals all say more than a sign-up count that never converts into use.
Write one hypothesis at a time. Bundling three claims into a single test produces a result nobody can attribute, and the team spends the next fortnight arguing about which part worked.
Invalidation needs a plan attached. The honest responses are a pivot to a different customer or problem, another experiment with a tighter design, or a decision to stop — and picking among them is easier when the threshold was written down.
Evidence for the stakes is unflattering. A March 2026 CB Insights teardown of 431 venture-backed shutdowns found poor product-market fit behind 43% of the failures, second only to running out of capital at 70%.
A validated hypothesis is also perishable, so most teams fold the surviving statement into a north star metric and keep measuring it.
Examples
Three widely documented launches show what testing a value hypothesis looks like in practice. Each one spent almost nothing to find out whether the core promise held before the engineering bill arrived.
Dropbox is the standard case. In 2008 the team published a short screencast demonstrating file sync that did not yet exist at scale, and the surge in beta sign-ups tested demand before the hard engineering was funded.
Buffer took the same route in 2010 with a two-page site. The first page described the scheduling product, the second showed pricing, and clicks on a plan that could not yet be bought measured willingness to pay rather than curiosity.
Airbnb tested a narrower claim. The founders rented air mattresses in their own apartment during a sold-out San Francisco conference in 2007, which answered whether strangers would pay to sleep in a private home at all.
Not every test ends well, and that is the point. A promising sign-up rate that collapses in week two has disproved the hypothesis just as cleanly as no sign-ups at all.
Service businesses run the same play. A back-office provider pitching a new reconciliation service can run it manually for two clients first — a minimum viable product in everything but code.
Related terms
A value hypothesis rarely travels alone. The terms below cover the method it belongs to, the artefacts used to test it and the habits that keep testing going after the first answer arrives.
- Lean Startup Methodology: the build-measure-learn approach that gave the value hypothesis its name.
- Minimum Viable Product: the smallest build that can generate validated learning about a hypothesis.
- Customer Discovery Process: the structured interviewing that surfaces the problem a hypothesis should name.
- Product Discovery Framework: the working system teams use to decide what to test next.
- Experimentation Program: the operating setup that runs tests at a steady cadence rather than ad hoc.
- Continuous Discovery Habits: the weekly customer contact that stops a validated hypothesis going stale.
FAQ
What is the difference between a value hypothesis and a growth hypothesis?
A value hypothesis asks whether customers get real value from the product; a growth hypothesis asks how new customers will find it. Testing growth before value scales a product nobody wants.
What makes a hypothesis falsifiable?
It states, in advance, an observable result that would prove it wrong. If no realistic outcome could disprove the statement, it is an opinion rather than a hypothesis.
How large does a test need to be?
Large enough that the result is not noise, and no larger. Many early hypotheses are settled by a few dozen genuine users behaving consistently.
What should happen when a value hypothesis fails?
The team pivots to a different customer or problem, redesigns the experiment, or stops, and the pre-agreed threshold decides which of those is honest.
Who should write the value hypothesis?
The team that will act on the answer. A statement written outside the build loop rarely carries a threshold anyone is willing to honour.
Does a validated hypothesis stay validated?
No, because markets, substitutes and expectations move.
Browse the Outsource Accelerator directory when a validated hypothesis needs delivery capacity you do not want to build in-house.
Independent
