Idea discovery

A prompt pattern for taking a one-paragraph domain description and working it through to a sequenced, evidence-tested set of product ideas — without reconstructing the process from scratch each time.

It’s deliberately domain-agnostic: the steps below have been run on maritime data businesses, but nothing in them is maritime-specific.

Getting the directive

get_playbook("idea-discovery") returns this page verbatim — read it and adapt it by hand, filling in the domain description.

Input

A one-paragraph description of the domain or application area to explore.

Constraints applied throughout

• Free/open data sources only — verify licences permit commercial reuse before relying on a source. A promising idea built on a data source you can’t legally use commercially is not a promising idea.
• Every idea needs a clear distribution channel — a press cycle, SEO, an existing community, or a developer channel. An idea with no plausible way to reach users is not worth qualifying further, however novel.
• Solo technical founder — favour products buildable as an app or API on serverless infrastructure. Ideas that need a team, capital, or bespoke infra to stand up don’t fit the constraint and should be discarded early, not qualified.

Steps

Web search at every stage. Pause for a decision at each transition — don’t run straight through to idea generation without confirming the frontier, assets, and landscape are right.

1. Research frontier — recent advances (roughly the last 3 years) in the domain’s literature. Skip this step entirely if the domain isn’t research-driven.
2. Assets — open datasets, records, feeds, and registries available in the domain; their licences; what can be inferred from them beyond their stated purpose.
3. Landscape — incumbents in tiers (raw data providers → enterprise intelligence → vertical tools), with pricing. Note who is served at what price point, and who isn’t served at all.
4. Gap hypotheses — propose gaps between what the landscape serves and what the assets/frontier make possible, then actively try to falsify each one with further searches. Report kills explicitly — a dead hypothesis is progress, not a wasted step.
5. Idea generation — batches of roughly 10 ideas. Each idea gets exactly three attributes: novelty, free data source(s) it relies on, and its distribution channel. No idea without all three.
6. Qualification — for ideas on the shortlist: product shape, user story, similar existing products, and the unserved niche it targets. Then pressure-test the load-bearing assumption — usually licensing, legal exposure, or moat — the one assumption that kills the idea if it’s wrong.
7. Sequencing — a build order that exploits shared infrastructure across ideas (a dataset pipeline or API client one idea needs is often reusable by the next). Name the cheapest first experiment explicitly.

Principles

• Cheap evidence before commitment — a falsifying search costs minutes; a shipped product costs weeks. Spend the cheap evidence first.
• Be honest when a space is saturated — a landscape step that finds five well-funded incumbents at every price tier is a valid, useful outcome. Don’t force a gap that isn’t there.
• Desk validation is not buyer validation — qualification and sequencing narrow the field on desk research alone. They are not a substitute for talking to a real buyer before building.