A one-page product experiment brief

Copy this brief before each test

A one-page brief prevents a test from turning into an activity with no decision attached. Keep it short enough to fill in before you begin. The thresholds below are examples; choose them from the decision you need to make, not from a universal startup rule.

Experiment name:
Owner and dates:
Decision this test will inform:

Target person / segment:
Recent behavior that makes them relevant:
Problem or outcome under investigation:

Assumption:
Why this assumption is risky:
Evidence that would raise confidence:
Evidence that would lower confidence:

Method and exact steps:
Recruitment source and inclusion criteria:
What participants will see or be asked:
What will be measured or observed:
How consent, privacy, and follow-up will work:

Time / money limit:
Stop condition:
Decision threshold (a learning rule, not a market statistic):
Possible decisions: continue / change / stop

Results:
What happened:
Counterexamples or surprises:
What the evidence does not show:
Decision and next test:

Make the test falsifiable and fair

Name one primary assumption. “Learn whether people like the idea” is too vague. “Among independent bookkeepers who reconciled expenses for a client in the last month, can we recruit five for a workflow interview within one week?” is observable. “Will 3 of 5 attempt a second use of a manually delivered weekly exception report?” tests a different claim. Do not combine these into one result.

Choose the least expensive method that exposes the assumption. Interviews are good for workflow discovery; they do not prove purchase. A prototype can reveal usability; it does not show whether a problem is frequent. A pre-sale tests a concrete commercial commitment, but only if you state what exists, what is promised, delivery timing, and how refunds work. A waitlist tests a message and willingness to share contact details, not usage or payment.

Protect participants and the quality of the evidence

Use a recruitment source you are allowed to use. Do not imply a feature exists when it does not. Say when a prototype is simulated or when a service is manual. Ask permission before recording; collect as little identifying data as possible; define who can see notes and when they will be deleted. Give participants an easy way to decline or stop.

Record the denominator and the selection process. “Four replies from 30 personalized invitations to people matching the criteria” is interpretable. “People are interested” is not. Keep an out-of-sample counterexample; do not quietly discard observations that conflict with the idea.

Close the loop

At the end, compare results with the threshold you chose before the test. If you change the threshold after seeing results, record why. State the next decision and the uncertainty that remains. A test is complete when it informs an action, even when that action is to stop. For help interpreting weak, mixed, or quantitative results, see decide what an experiment result means.

For the surrounding process, read idea to first experiment and customer interviews.