Hypothesis Formation
Hypothesis formation is the discipline of turning a diagnosed funnel leak into a specific, falsifiable statement that names the change, the mechanism, the primary metric, and the expected lift. A well-formed hypothesis is the contract between the diagnosis and the experiment. It is what makes a result readable.
- A hypothesis is falsifiable. If it cannot be wrong, it cannot be tested.
- Every hypothesis names one primary metric. Secondary metrics are guardrails, not tie-breakers.
- Expected lift is anchored in the diagnosis, not in optimism.
- The hypothesis is written before the experiment is scoped, not after the data comes back.
Where does this stage earn its keep?
Teams skip straight from a diagnosis to a design change. When the result comes in, no one agrees on what the test was actually measuring, and the winner gets rolled back three months later because a downstream metric quietly regressed.
Terms this stage depends on.
A statement structured so that a defined outcome would disprove it. The opposite of a directional intent.
The single metric the experiment is powered to move, chosen because it maps to the sized opportunity.
A metric the experiment is required not to harm, monitored throughout the test.
What changes when this stage is done properly.
Illustrative comparison of hypothesis quality.
| Metric | Gut-driven attribution | Fully-instrumented data pipeline |
|---|---|---|
| Specificity | "Redesign the pricing page." | "Adding annual toggle raises trial-to-paid by 20% relative because it anchors on the lower monthly-equivalent price." |
| Metric | Not named. | Named. Powered. Pre-registered. |
| Guardrails | Discovered post-hoc. | Chosen up front. Monitored during the test. |
The operational shape of this stage.
- 01Translate every sized opportunity into a hypothesis in the form: change, mechanism, primary metric, expected lift, guardrails.
- 02Pre-register the hypothesis in the experiment log before scoping the design.
- 03Reject hypotheses that cannot be resolved within the traffic available in a reasonable cycle.
Hypothesis from a diagnosed SaaS trial funnel
A PLG SaaS diagnosis surfaced that free-trial users who invited a second seat converted to paid at materially higher rates. The hypothesis: prompting a seat invite inside the onboarding checklist will raise trial-to-paid by roughly 30% relative, powered on trial-to-paid as the primary metric, with activation-per-user and support-ticket volume as guardrails. The test ran and read out at a 33% relative lift at p=0.01.
Frequently asked about this stage.
What makes a growth hypothesis falsifiable?
+
What makes a growth hypothesis falsifiable?
+How many hypotheses should be in flight at once?
+
How many hypotheses should be in flight at once?
+See this stage run against your numbers.
A 30-minute Growth Audit. You leave with two or three specific findings, whether or not we ever work together.