“The app will succeed” sounds like a hypothesis, but it does not tell us what to observe, with whom, or when. A useful hypothesis allows us to imagine what result would support it and what result would challenge it before running the test. Strategyzer recommends precise, testable claims and separates a hypothesis from its experiment, measurement, and threshold [R01][R03].

A hypothesis is not a wish or a task

A wish says “We want referrals.” A task says “Create a landing page.” A hypothesis says “Restaurant managers struggling with cancellations will be willing to leave their contact details to request a demonstration.” The page is a possible instrument, not the hypothesis. Requesting a demonstration does not yet prove they will pay.

If it is unclear whether a claim describes an observed event, a statement, or an interpretation, revisit the differences between observations, assumptions, hypotheses, and evidence before selecting a metric. Measurement precision cannot rescue an unsupported inference.

Identify population, context, behavior, and period

First specify who: not “all customers,” but a segment with verifiable inclusion criteria. Then define when and where: small restaurants currently accepting telephone bookings during a two-week pilot. Finally state the behavior, condition, or result expected. If you use a quantity, define the unit and denominator.

A reusable editorial structure is: “We believe [defined segment], in [context], will [perform an observable behavior / meet a condition] within [period].” This is not an official mandatory template. It is a starting point; next we must select an appropriate experiment.

Make it refutable without claiming absolute certainty

“People value convenience” accommodates almost any result. “At least six of twenty eligible decision-makers will submit a pilot request under the presented offer” specifies a condition. A result of five is not automatic proof that the business will fail. It means the evidence did not meet this particular threshold, requiring interpretation of context, sample, and bias.

A hypothesis should not conceal several claims. “Restaurants want the product, can use it, and will pay S/100 per month” combines desirability, feasibility, and viability. Split the claims using hypothesis types.

Define measures and criteria separately

Strategyzer's Test Card separates four elements [R03]: hypothesis, test method, measurement, and threshold. For a pilot offer, the metric might be acceptance rate among eligible contacts; the denominator is everyone who actually received the same offer; the criterion is a threshold established before gathering results. Do not choose a threshold because a percentage looks attractive. It should be justified by minimum business economics, acquisition costs, or decision requirements.

Vague claim More specific hypothesis What still needs justification
“They need reminders” At least 8 of 20 eligible managers will report cancellations attributed to forgotten bookings in the past month Threshold evidence and reliability of recall
“The product is easy” At least 16 of 20 defined participants will complete a booking without assistance Participant selection and representativeness
“We can operate it” The team will process 100 simulated bookings in one shift with zero critical duplicates Appropriate volume and error tolerance
“We will make money” Contribution per account will be positive using costs observed in the pilot Real costs and time horizon

The numbers are illustrative, not universal recommendations. “How many participants do I need?” belongs in experiment design, not in copying numbers from a table.

From hypothesis to test record

Assign an ID, owner, version, risk, starting source, scope, expected behavior, measure, criterion, deadline, and possible decision. Leave space for contradicting evidence and limitations. Avoid rewriting the claim after seeing results; if it changes, preserve the prior version and create another.

During and after the experiment, use the method for documenting results while preserving the original version. This keeps the precommitted hypothesis separate from later observations and interpretations.

Worked example: H-D01 — “Among managers of restaurants with fewer than ten tables that currently accept telephone bookings, five of ten will agree to trial a digital booking log for fourteen days after a standard demonstration.” This is specific, but agreeing to a trial does not establish willingness to pay: a separate purchase hypothesis is required.

What you should have by the end: three distinct hypotheses — desirability, feasibility, and viability — with defensible measures and criteria, ready to prioritize. Main sources: [R01], [R03], [R18].

Bibliography and references