Discovery · knowing what to build
The trio discovers together and keeps every solution laddered to a real outcome. Interview for stories, test the riskiest assumption first, and — for AI — treat "the model is good enough" as a bet you test on a golden set before any UI.
Opportunity Solution Tree (Torres) — every solution ladders to a real outcome.
Above the fold
Discover as a weekly habit, as a trio.
PM + design + eng discover together — no sequential handoffs, so the people who build it hear the customer firsthand.
Interview for stories, never for features.
"Tell me about the last time you…" — interview to discover opportunities, never to validate a solution you already picked, and never ask "what features do you want?"
Test the riskiest assumption first.
Split a solution into desirability / viability / feasibility / usability — and for AI, Model-Capability is a bet: golden-set eval before any UI.
Continuous discovery · Torres
Torres
Trio (PM + design + eng) discovers together — no sequential handoffs; the people who build it hear the customer firsthand.
Opportunity Solution Tree: outcome → opportunities (needs/pains from interviews) → solutions → assumption tests. Keeps the team honest that every solution ladders to a real outcome.
Torres · continuous interviewing
"Tell me about the last time you…" — for churn: "…opened the app and then stopped."
Interview to discover opportunities, never to validate a solution you already picked.
never ask "what features do you want?"Torres · assumption testing
Split a solution into desirability / viability / feasibility / usability assumptions; test the riskiest first. + Model-Capability (AI only) — "the model is good enough for this job" is a BET; test it on a golden set before any UI. Distinct from feasibility (can we integrate?) and desirability (do users want AI here?).
Default category ordering when evidence is unclear:
Before the riskiest-assumption gate, run a written-first pre-mortem (Doshi/Klein) to dodge groupthink and feed surfaced failure modes back into the assumption inventory.
Cheapest-valid-test ladder · per category
| Category | Cheapest valid test |
|---|---|
| Desirability | story interviews / fake-door / concierge |
| Usability | 5-user / first-click |
| Feasibility | spike / POC |
| Viability | WTP / unit-economics |
| Model-Capability | offline golden-set eval |
Context & contrarian
Context-adapt: enterprise B2B swaps weekly interviews for an advisory board / design partners; hardware/regulated raises feasibility & compliance and makes tests slower.
Contrarian: for small reversible features "shipping is the experiment" beats discovery theater (Type-2); and riskiest-first can pass each assumption individually yet fail as a whole — run an end-to-end concierge / Wizard-of-Oz before full build.
2024-26 AI-era · Torres
AI speeds discovery but misses ~20-40% of opportunities/insights (directional) — keep a human in the loop.
Disciplines: decompose monolithic "do my research" prompts into narrow steps; run the error-analysis loop (read outputs → code failures → fix the prompt) like eval traces; verify against raw transcripts, never the AI's summary.
JTBD judgment · Christensen
The job-story format, the Four Forces churn diagnostic, and the three things the models miss about competition, consumption, and what a need actually is.
Christensen · canon
"When I [situation], I want to [motivation], so I can [expected outcome]"
Frame the struggle, not the feature. The three job types — functional / emotional / social — are canon.
Four Forces of Progress · the churn "why" diagnostic
Forces toward the switch
The struggle with today's situation that drives someone to look.
The appeal of the new solution.
Forces holding it back
Fear of the unknown solution.
The comfort of the current way.
Read it as a churn diagnostic
Anxiety + inertia are usually the dominant blockers for a new feature class, so reduce friction, don't just add pull.
What the models miss
Competition is the outside-category alternative — the Milkshake competes with bananas and boredom, not other shakes (Christensen). Frame the job around the circumstance, not the product category.
Consumption job ≠ purchase job — design for the moment of use, not just the buy decision; churn hides in the consumption job.
Need = direction + metric + goal ("cut time-to-first-invoice below a day"), not a feature request — a disciplined need statement is solution-agnostic and measurable.
Pointers
Direction → current state → target condition → obstacle → experiment. Converts vague strategic intent into the next measurable step.
Build only for a sharp problem, judged multi-signal — workflow compression (collapses many painful steps into few) and a visceral "I need this now" in interviews.
Pair with the Zone-of-Benefit bar — be ≥3× better on the dimension the customer actually feels (illustrative, not a law). Diffuse/weak signal → don't build; keep discovering.
DirectionalTry it
The Ulwick opportunity score, live: importance plus the satisfaction gap — added, never multiplied. High-importance + high-satisfaction lands where it belongs, served rather than a target.
The opportunity scorer needs JavaScript to run. Run it on the Tools page.