Product Discovery: the method to validate product ideas before you build

Product Discovery precedes development: it consists of checking that a problem is worth solving, and that the envisioned solution actually solves it, before spending engineering time on it. Marty Cagan, a former product VP at eBay and Netscape, did much to popularize this approach from the 2010s onward, through his book Inspired and the work of his firm, Silicon Valley Product Group.

The four risks Discovery aims to reduce

The method, step by step

1. Clearly state the assumed problem, without starting directly from a solution.
2. Interview a sample of real or potential users to check that the problem exists and is a priority for them.
3. Sketch several possible solutions, without committing to the first idea that comes up.
4. Test these directions at low cost (prototype, mockup, limited experiment) before any development.
5. Decide, based on this feedback, whether to build, adjust or drop the idea.

A simple template to get started

• Observed problem: what behavior or frustration was observed, and for whom?
• Hypothesis: why do we think this problem is worth solving?
• Evidence sought: what signal would confirm or disprove this hypothesis?
• Test method: interview, prototype, survey, limited experiment.
• Decision: build, adjust the hypothesis, or drop the idea.

Common mistakes

• Confusing Discovery with simply collecting customer requests.
• Jumping straight to development as soon as an idea looks good on paper.
• Only talking to already convinced users, which biases the feedback.

Discovery has its own limits, though. It works well for improving an existing product based on observable feedback, much less so for genuine breakthroughs, where users struggle to articulate a need they have never been able to satisfy for lack of an existing solution. An organization that turns it into a mandatory gate for every decision, applied without judgment, risks delaying bets that deserved to be taken earlier, on the strength of the team's judgment rather than exhaustive validation.