How to test a risky idea before you build too much

Building becomes dangerous when it answers a question that is not yet decisive. A team spends weeks proving it can implement the product while the largest uncertainty is whether target users experience the problem, will change an existing workflow, can approve adoption, or can be reached economically.

Name the assumption whose failure would invalidate the current plan. Define an observable, costly behavior that would support or weaken it. Then build the smallest responsible experiment that can create the conditions for that behavior. The artifact is only useful insofar as it buys a decision.

The Lean Startup supplies the Build–Measure–Learn loop and minimum viable product. The Mom Test keeps the evidence honest by distinguishing past behavior and commitment from opinions, compliments, and imagined futures.

Risk belongs to assumptions, not feature size

An idea contains several kinds of uncertainty. The problem may be infrequent or adequately solved. The proposed value may not be large enough to change behavior. The buyer may differ from the user. Distribution may cost more than the customer relationship can support. A technical or regulatory dependency may make delivery impractical.

Write each as a falsifiable statement about a particular group and behavior. “People need this” cannot be tested cleanly. A useful assumption identifies who, in which situation, currently does what, and what commitment the proposed change is expected to produce.

The riskiest assumption is not always the one with the least evidence. It is the one whose failure causes the largest change in strategy and cannot be cheaply avoided. A difficult engineering problem may be secondary if there is no evidence the outcome matters. Strong demand may make technical feasibility the next decisive risk.

Rank assumptions by consequence of being wrong, current evidence, and cost of testing. Test the one that can invalidate the most downstream work soonest.

Design the evidence before choosing the artifact

The Build–Measure–Learn loop should be planned backward. First state what decision will follow. Then define the observation that distinguishes plausible outcomes. Only then choose what to build.

If the risk is problem frequency, reconstruct recent events and inspect existing workarounds before presenting a solution. If the risk is comprehension, a realistic representation may be enough to see whether target users recover the intended value. If the risk is willingness to change workflow, the test must include the relevant setup, data, stakeholder, or switching cost. If the risk is willingness to pay, a free signup does not reach the uncertainty.

Choose evidence proportional to the claim. A click can show attention to a specific offer under specific conditions. It does not establish retention or payment. A scheduled pilot can show organizational interest. It does not establish repeated value. Each test should state what it cannot conclude.

The measurement must be defined before results arrive: target group, exposure, behavior, time window, threshold, and guardrail. Otherwise the team can treat any positive signal as validation and every negative signal as an artifact problem.

An MVP is minimum relative to the learning target

Ries’s minimum viable product is the least work needed to complete a learning loop. It may be a concierge service, a manual workflow, a prototype, a demonstration, or working software. It does not need to scale if scale is not the current question.

Zappos’s early test, described by Ries, used photos of shoes from local stores and purchased an item at retail after an order arrived. The process could not support a mature retailer, but it exposed the demand question before inventory and distribution infrastructure were built.

The minimum must preserve the part of the experience relevant to the risk. A manual service can test whether the outcome is valuable but may not test whether users can operate a self-serve product. A polished prototype can test comprehension but not whether the backend is feasible. State which uncertainty remains after the shortcut.

Minimum does not excuse deception, unsafe handling of data, avoidable harm, or damage to trust. Safety, privacy, and reputation may be constraints the experiment must respect from the start. The correct minimum is the smallest responsible test, not the cheapest thing that can produce a number.

Commitment is evidence that the test reached a real constraint

The Mom Test warns that praise and predictions are cheap. A target user can sincerely like the idea without changing anything. Stronger evidence asks for time, access, reputation, money, or workflow change appropriate to the stage.

Commitment should match the remaining uncertainty. An introduction can test access to the buyer. A sanitized artifact can expose the actual process. A scheduled trial with owner and success criteria can test adoption. Payment can test value capture when the offer is concrete enough to price.

A refusal is also evidence. It may reveal privacy, timing, authority, switching cost, or low priority. Record the reason and observation rather than collapsing every no into lack of demand or every yes into validation.

Do not manufacture commitment before the participant can evaluate the offer. Asking for money against a vague promise may test trust in the founder rather than the product hypothesis. The cost must connect to the claim being tested.

Results need a precommitted update rule

Before running the experiment, define what will cause the team to persevere, pivot, run another test, or stop. Include magnitude, target segment, time window, and critical guardrails where possible. A threshold is not certainty; it prevents the interpretation from moving after the result.

Separate failure of the hypothesis from failure of the test. If target users never encountered the offer, the distribution method may be the unresolved risk. If they understood it but would not take the relevant next step, value or priority may be weaker. If they committed but could not achieve the outcome, delivery is the issue.

Use cohort or case-level evidence so aggregate attention does not hide who acted and under which version. Preserve observations and inferences separately. “Three qualified teams completed setup” is different from “the market wants the product.”

The decision may be to narrow the segment, change the mechanism, test a downstream risk, or abandon the thesis. Learning becomes valuable only when it changes resource allocation.

The next move is a one-week assumption test

Write the current plan’s five most important assumptions across problem, value, access, feasibility, and growth. Mark which one would make the largest amount of planned work unnecessary if false. State the costly behavior that would support it and what result would change the plan.

Choose the smallest artifact capable of eliciting that behavior from real target users. Remove features that cannot affect the observation. Add the trust, safety, and context required for the test to remain interpretable. Run it for a defined cohort and record behavior, refusals, and constraints rather than compliments.

At the review, make the promised decision. If the result is ambiguous, name which missing evidence created the ambiguity and design the next test around it. The success condition is not a positive result. It is that the team spent a small amount to make a consequential assumption more accurate before spending a large amount to automate it.