Double diamond framework for product interviews

The double diamond is a design process model from the British Design Council (2005), not a PM framework. That origin matters: design-heavy companies (Figma, Airbnb, Apple) treat it as core vocabulary; pure-execution shops may not. In interviews, its value is structural: it makes the most common failure mode visible, candidates who skip the first diamond entirely.

Two diverge-then-converge passes. Diamond 1 works the problem space: Discover (diverge) then Define (converge to a problem statement). Diamond 2 works the solution space: Develop (diverge) then Deliver (converge, pick one, test it). You do not open Diamond 2 until Diamond 1 is closed.

Phase shorthand: Discover means low-engagement users probing for the job being done. Define means a problem statement you could write a test for, plus a viability gate. Develop means meaningfully different solution vectors where the real question is “does AI make this more lovable or more obnoxious?” Deliver means one solution with a metric that would catch a failure.

A full worked example

Question: “How would you improve Spotify engagement?”

Discover (diverge): Segment by use-case: commute, focus, social, workout. Interview users who opened Spotify three or fewer times last week, not power users. Probe: what was the last session worth opening, and what made them close it? Hypothesis: engagement drop is a transition-moments problem; Spotify works when context is clear and fails when it is ambiguous.

Define (converge): The real problem is not lack of content. Spotify has no way to help users answer “what do I want right now?” without active effort. Job-to-be-done: decide what to listen to in under five seconds. Viability check: Spotify’s revenue is tied to active listening time, and a five-point improvement in 7-day return rate for the occasional-listener cohort is worth real retained subscriber revenue at scale.

Develop (diverge): Four different solution vectors: (1) a context-aware card on home that reads calendar and time-of-day; (2) a “resume where you were” persistent rail for habit continuity; (3) a “what’s the vibe” three-tap entry replacing the search bar for undirected sessions; (4) a social signal feed showing what connections are playing live. Generated to force real trade-offs, not to find the answer yet.

Deliver (converge): Kill (1): calendar permissions have low acceptance rates and ambient sensing trust remains low even when the technical lift is trivial. Kill (4): social features at Spotify have consistently underperformed outside the 18-24 cohort. Prioritize (2): shortest path to habit formation, measurable in 7-day return rate for a session of 10-plus minutes. Success metric: 7-day retention in the occasional-listener cohort up eight or more points within 90 days. The reason I waited: every company’s first instinct on a retention drop is to add a push notification. Diamond 1 revealed the problem is decision friction, not awareness. Notifications would have made it worse.

Use it, do not recite it

The most common failure: candidates describe all four phases, then jump to solutions. Saying “first I would do discovery” without naming what you learned or how it changed your problem statement is not doing Diamond 1.

Two traps. The reverse diamond: a stakeholder pre-decides the solution and you execute. If an interviewer hands you a specific feature to build, push back by asking what problem it solves. The conveyor-belt error: treating the framework as one-pass when findings in Develop can reopen Define. Interviewers at Google and Airbnb specifically probe for this.

The 2026 reframe

Develop used to be gated by feasibility. It is not anymore. Computer vision, NLP, and personalization at scale are commodity. The real gates are viability (will people pay, is the segment large enough?) and lovability: does the solution meet people where they work, anticipate needs without being obnoxious, and require the minimum steps to get the job done? A PM who walks through double diamond without a viability test in Define and a lovability standard in Deliver is using a 2018 version.

The Smashing Magazine critique (2023) is worth knowing: most orgs pre-decide the solution and the PM executes. Name the tension; apply the framework anyway. Skipping Diamond 1 costs more later than it saves now.

For the viable/lovable lens on AI features, see feasibility is free and lovable, not just usable. For how double diamond relates to adjacent frameworks, see the product sense 6-step.