tmj-90 De-risk product bets before building
Discovery's job is to fail fast and cheaply — identify wrong assumptions before they're baked into shipped software.
Opportunity Solution Tree (Teresa Torres)
Desired outcome (metric to move)
└── Opportunity (unmet user need / pain / desire)
└── Solution idea (intervention)
└── Experiment (cheapest test)
Rules:
- Opportunities come from user evidence — interviews, support tickets, analytics — not internal opinions.
- One desired outcome per tree. Multiple outcomes = no prioritisation.
- Solutions are hypotheses; experiments are the cheapest way to test each hypothesis.
- The tree is a living document — update as evidence accumulates.
Assumption mapping
For each solution idea, map its assumptions across four risk dimensions:
| Dimension | Question | Example assumption |
|---|---|---|
| Desirability | Do users want this? | "Users will pay $10/month for this feature" |
| Viability | Does this create sustainable business value? | "This will reduce churn by 5%" |
| Feasibility | Can we build this? | "The API supports the required event granularity" |
| Usability | Can users use this without training? | "Users will understand the new onboarding flow without docs" |
Score each assumption: Risk (1–3) × Certainty (1–3 inverse — low certainty = high score). Highest scores = test first.
Validation methods (choose by cost)
| Method | Cost | Validates |
|---|---|---|
| Desk research | Hours | Market size, competitor landscape, existing solutions |
| Customer interview (problem) | Days | Pain existence, frequency, severity, willingness to solve |
| Fake-door test | Days | Demand signal (click-through to a "coming soon" page) |
| Prototype usability test | Days–week | Usability, core interaction |
| Wizard-of-Oz / concierge | Week | Desirability + willingness to pay without building the feature |
| A/B experiment | Week–months | Behavioural impact on a metric |
Fail fast principle: choose the cheapest method that can kill or confirm the assumption. Don't spend a month building a prototype when 5 interviews would do.
Discovery sprint structure (1–2 weeks)
- Week 1 — hypothesis formation + cheapest validation plan.
- Daily evidence reviews — what did we learn today? Does it change the tree?
- End-of-sprint decision gate — proceed (evidence supports the bet), pivot (evidence points to a different opportunity), or stop (evidence shows no opportunity).
Steps
- Define the desired outcome. One metric; baseline and target; time horizon.
- Build the OST. Map opportunities from existing user evidence. Diverge before converging — generate many opportunities; then cluster and score by frequency × severity.
- Map assumptions for top opportunities. Use the four dimensions. Identify the assumptions that would kill the bet if wrong.
- Plan experiments. For each killer assumption: cheapest method that provides signal in < 1 week.
- Run experiments. Document: hypothesis, method, result, what we concluded.
- Decision gate. Proceed / pivot / stop with explicit reasoning. Capture in lore.
Review checklist
- Desired outcome specific — one metric; measurable; time-bounded.
- Opportunities from user evidence — not internal brainstorms.
- Killer assumptions identified — the ones that, if wrong, kill the bet.
- Cheapest validation method chosen — not over-invested in expensive experiments for low-risk assumptions.
- Decision gate explicit — proceed/pivot/stop with documented reasoning.
Rules
- Discovery debt is real: every feature shipped without discovery is a bet placed blind.
- User interviews discover problems; prototypes test solutions. Don't use interviews to validate solutions.
- A "proceed" decision without documented evidence is not a discovery output — it's a guess with extra steps.