8 Lenses for Working with Hypotheses: How to Evaluate an Idea from Different Angles and Avoid Mistakes
Beyond ordinary validation, there are 8 powerful lenses that help product managers comprehensively evaluate a hypothesis: from probability to causality control and Premortem.
Hypothesis testing is not just about confirming your ideas. It's about creating conditions where the weakest ones can "fail" — cheaply and early. To approach this process systematically and avoid self-deception, it's useful to look at each hypothesis through several "lenses."
Here are 8 powerful mental models that will help you comprehensively evaluate an idea and make an informed decision.
1. Probability Instead of Belief
Stop "believing" in hypotheses. Instead, treat them as statements with a certain level of confidence that you constantly update based on new data.
- How it works: "I am 40% confident that this change will increase conversion." After a weak signal from five interviews, your confidence might drop to 20%. After a strong signal from an A/B test, it might rise to 80%.
- Why this is needed: This approach protects against emotional attachment to ideas and forces you to think in terms of bets and signals.
2. Pre-commitment
Define rules for interpreting results and thresholds for success/failure before starting the test.
- How it works: Write down in advance: "If conversion increases by 5% or more, we scale. If less than 1%, we kill the hypothesis. If between 1% and 5%, we conduct additional analysis."
- Why this is needed: This is the main safeguard against self-deception and "fitting" interpretations to the desired outcome after data is received.
3. Hypothesis as a Decision
Formulate hypotheses not as questions, but as triggers for specific decisions. What exactly will you do based on the test results?
- How it works: Not "We think users need this feature," but "We are testing whether users are willing to pay for feature X to decide: invest in full development or close the direction."
- Why this is needed: Connects hypothesis testing with real actions and resources, cutting off "research for research's sake."
4. Decomposition into Assumptions
Any large hypothesis consists of several smaller assumptions. Break it down.
- How it works: The hypothesis "People will buy our new module" breaks down into:
- Do they need it? (Is the problem real?)
- Will they be able to use it? (Is the UX clear?)
- Will they buy it? (Does the value exceed the price?)
- Can we deliver and support it? (Technical and operational feasibility)
- Why this is needed: Allows choosing the cheapest test to verify the riskiest assumption first.
5. Triangulation
Never trust a single data source. Use at least two independent signals to confirm a conclusion.
- How it works: Combine quantitative and qualitative data. For example, if an A/B test shows an increase in conversion (behavior), confirm this with 3-5 interviews to understand "why" (words). Or combine willingness to pay (money) with repeated use (repetition).
- Why this is needed: Reduces the risk of drawing conclusions based on noise, data errors, or an unrepresentative sample.
6. Premortem and Inversion
Instead of asking "how can we succeed?", ask: "Imagine we have already failed. What went wrong?"
- How it works: The team generates a list of potential causes of failure. "Users didn't understand the value," "integration turned out to be too complex," "a competitor released an analog faster."
- Why this is needed: Transforms future risks into concrete assumptions that can and should be tested now, before they become reality.
7. Red Team
Assign one or more people the role of "devil's advocate." Their task is to deliberately look for weaknesses in the hypothesis, data, and test methodology.
- How it works: The Red Team asks uncomfortable questions: "What if the sample was biased?", "Maybe this effect is explained by seasonality?", "Are we sure this isn't just a novelty effect?".
- Why this is needed: This is a formalized way to combat groupthink (groupthink) and confirmation bias.
8. Control and Causality
Always strive to separate correlation from causality.
- How it works: The best tool is a controlled experiment (A/B test). If that's not possible, use holdout groups (those not affected by the change), A/A tests (to check for randomness), or "placebo" (changes that should not affect the metric).
- Why this is needed: To be sure that a change in a metric is caused precisely by your intervention, and not by random noise.
Using these eight lenses transforms hypothesis testing from a gamble into a disciplined process of risk management and knowledge production.