Hypotheses
Product hypothesis template: how to write one you can actually test
Updated
Most product ideas arrive as a feature request: “we should add favourites”. A hypothesis turns that into a claim you can be wrong about. It names who the change is for, what it should do for them, and how you will know. This article gives you a template, shows a worked example, and lists the checks that separate a testable hypothesis from a wish.
The template
We believe we will [measurable outcome] if [who it is for] get [the benefit they feel] with [the feature or change]. We will know when [metric] moves from [baseline] to [target] within [timeframe].
Every blank does a job. The outcome is what the business or the user gains, not the feature itself. Who keeps you from building for everyone and helping no one. The benefit is the change in the user’s day, in words they would use. The feature comes last on purpose: it is one way to get the outcome, not the point.
A worked example
Say interviews with ICU nurses keep turning up the same complaint: they rebuild the same infusion order from scratch every shift. Analytics agrees: infusion prescriptions take about 15 minutes, three times longer than any other order.
We believe we will cut infusion prescribing time to under 5 minutes if ICU nurses and prescribers get their unit’s most-used infusions suggested and fully prefilled with suggested infusions on the prescribing screen. We will know when median prescribing time moves from 15 to under 5 minutes within 6 weeks of release to two pilot units.
Notice what the example does not say. It does not say “nurses want favourites”, which is a solution dressed up as a need. And it does not say “make prescribing faster”, which nobody could ever prove wrong.
Five checks before you test it
- Could it fail? If no result would make you drop the idea, it is not a hypothesis yet. Write down what “wrong” looks like.
- Is the baseline real? “From 15 minutes” needs a number you have measured, not a guess. If you do not have one, measuring it is your first experiment.
- Is the who narrow enough? “Clinicians” is too wide. “ICU nurses on night shift” is something you can recruit for and measure.
- Does it rest on evidence? Link it to the interviews, tickets or analytics that made you believe it. A hypothesis with no evidence behind it is an opinion.
- What is the riskiest assumption? Name the one thing that, if false, sinks the whole idea. Test that first, cheaply.
Write down how confident you are
Put a rough confidence on each hypothesis, say 40%, before you test it, and update it as evidence comes in. It forces an honest conversation about which ideas are bets and which are near-certain, and it shows which ones deserve a proper experiment and which only need a quick check.
Common mistakes
- Bundling two bets in one. “Prefilled infusions and a new layout will speed things up” cannot tell you which change worked. Split it.
- Choosing the metric after the results. Decide the metric and target before you start. See success metrics and guardrails.
- Letting the hypothesis drift. If you change it mid-test, you are running a new experiment. Record it as one.
- Writing it alone. A hypothesis improves when someone challenges it. Ask a colleague to argue the opposite.
Where this fits
A hypothesis sits in the middle of a chain: customer insights lead to a hypothesis, an experiment tests it, the result becomes a finding, and findings justify requirements. Keeping those links is what lets you answer “why are we building this?” months later. The requirements traceability article covers the end of that chain.
Keep your evidence and decisions linked
Groundwork takes your team from customer insights to evidence-based requirements. Free to start, with your whole team.
Get started