Hypothesis

Origin & evolution

The idea travels into product work from the scientific method, where a hypothesis is a falsifiable conjecture tested against observation👁️ObservationUser ResearchA specific behaviour or statement observedView reference →. Eric Ries imported it into startups in The Lean Startup (2011), reframing a new venture as a bundle of untested beliefs. He named the two most dangerous ones the value hypothesis (will customers find this valuable?) and the growth hypothesis (how will they find it and spread it?), and called them collectively the leap-of-faith assumptions: the parts on which the whole plan depends.

Jeff Gothelf and Josh Seiden gave teams a sentence to write them in. Lean UX (2013) popularised the template "We believe [this] will result in [that]; we will know we are right when we see [this signal]." The format does quiet but real work. The "we believe" forces the belief into the open, the "will result in" names the expected effect, and the closing clause commits the team to a measurement before the build starts, so success cannot be redefined after the fact.

The field then sharpened the distinction between a belief and a tested belief. David Bland and Alexander Osterwalder's Testing Business Ideas (2019) treats hypotheses as the output of Assumptions🤔AssumptionStrategyA belief taken as true that underpins a strategyView reference → Mapping: surface every assumption, plot it by importance against evidence, and convert the riskiest, least-evidenced ones into hypotheses worth testing first. The hypothesis became the unit you prioritise, not a label you attach to a plan after writing it.

How it works in practice

A team running a project-tracking tool notices that 70% of new sign-ups never invite a teammate, and trial-to-paid conversion sits at 4%. They suspect the empty first project is the problem. The raw belief, "onboarding is weak", is untestable, so they write it as a hypothesis: "We believe that pre-filling a sample project for solo sign-ups will raise week-one teammate invites; we will know we are right when invite rate among new accounts rises from 18% to 30% within two weeks."

Every clause is now load-bearing. The change is specific (a pre-filled sample project), the population is named (solo sign-ups), the metric🔢MetricStrategyA unified metric that measures progress, health, or behaviour across the productView reference → is chosen in advance (week-one invite rate), and the threshold is stated (18% to 30%). If the rate climbs to 22%, the hypothesis is not vaguely "trending positive"; it failed its own bar, and the team learns the empty project was not the real blocker. The prediction earned its keep by being wrong in a way they could see.

Hypothesis vs. its neighbours

• Assumption is an unexamined belief the team is currently acting on, often without noticing. A hypothesis is an assumption that has been made explicit, framed as a prediction, and queued for a test. The standard move is to surface assumptions, then promote the riskiest ones into hypotheses.
• Experiment🧫ExperimentValidationA test designed to validate a hypothesisView reference → is the test you run to gather evidence. The hypothesis is the claim under test; the experiment is the procedure that confronts it with data. One hypothesis can be probed by several experiments, and one experiment can bear on several hypotheses.
• Theory carries far more weight in scientific usage: a well-substantiated explanation supported by a body of evidence. A product hypothesis is a single, local, disposable prediction, designed to be cheap to kill. Confusing the two leads teams to defend a guess as though it were established fact.

In the graph

In the Unified Product Graph, hypothesis🧪HypothesisValidationA testable belief about a solution is a hub in the Discovery, Research & Validation region, whose mental model is a loop: question, hypothesis, test, evidence, decision⚖️DecisionStrategyA recorded decision with context, rationale, and consequencesView reference →, repeat. Its lineage is explicit across a boundary edge: a strategy-side assumption🤔AssumptionStrategyA belief taken as true that underpins a strategyView reference → crosses into validation via assumption_becomes_hypothesisAssumptionbecomesHypothesiscausal, capturing the promotion from unexamined belief to testable claim. A solution🔧SolutionDiscoveryA proposed approach to address an opportunityView reference → proposes it (solution_proposes_hypothesisSolutionproposesHypothesiscausal), it aims at a strategic outcome🏆OutcomeStrategyA desired business or user outcomeView reference → (hypothesis_targets_outcomeHypothesistargetsOutcomecausal), and it accumulates proof through hypothesis_has_evidenceHypothesishas evidenceEvidencehierarchy. The loop closes when a learning🧠LearningValidationAn insight gained from an experimentView reference → writes back via learning_updates_hypothesisLearningupdatesHypothesiscausal, so a belief that survives a test is visibly distinct from one that has never met data.