Choose an opportunity you can defend

Public-source research checked on 2026-10-10. This is a proposed exercise, not a completed audit or evidence of results. The example is wholly fictional.

The decision

Should you investigate one task, build a small prototype, or stop this candidate? A popular phrase, a low difficulty score and a competitor's advertised price answer different questions. None proves that your proposed page has a useful gap or can earn revenue.

Inputs

Bring one audience, one task, a target language and market, a time limit, and the economic model you want to examine. An existing site is optional. Do not buy a tool subscription to complete this exercise. Missing data stays UNKNOWN.

Actions

1. Write the job: “When [situation], [person] needs to [finish task], because [consequence].” Gather a small bounded set of direct descriptions or observations, including a case where the current solution works. Link a repost to its original source; it is not a second independent need. GOV.UK distinguishes user needs from assumptions and proposed solutions. (S01)
2. Write query variants for beginners and experienced users. Record when and how you inspect results: engine, exact query, market, language, device, personalization limits. A general web-search tool's ordering is not a Google rank observation. The result set is a dated supply sample, not a market census. (S02, S03)
3. Open a few relevant alternatives. Separate advertised features from features you actually try. Use invented inputs, check one complete output and record the conditions tested. Do not upload personal files, create public objects or pass a payment step. “Not tested” differs from “missing.” These are exercise safeguards, not a search ranking rule.
4. Keep a demand ledger: query, likely task, source, unit, location, period, retrieval time and uncertainty. Do not sum overlapping synonyms into unique people. Trends is sampled and normalized; its zero may mean insufficient volume, and autocomplete does not rank popular demand. A vendor estimate remains an estimate. (S07)
5. Write one gap that survives the alternatives, plus its strongest competing explanation. If existing tools already solve the proposed task, record NO_GO for this version. A useful outcome can be declining to build.
6. Examine economics as a conditional model. For advertising, estimated revenue = page views / 1000 × page RPM; RPM is not CPC or final receipts. For a purchase model, keep visits, qualified starts, completed purchases, refunds and costs separate. Unknown rates make the outcome unknown. A sensitivity table is a planning aid, not a forecast. (S08)

Worked output — fictional

SpokeNote proposes a printable cycling packing list. In this invented scenario, Alternative A already provides a printable checklist and Alternative B lets users edit it. No user evidence establishes an unmet task; search volume and page RPM are unknown. The decision is “do not build the generic printable version.” The next bounded question is whether a particular packing task repeatedly fails in existing tools. This is invented reasoning, not a claim about any real search result or product.

Deliverable and evidence

Fill the opportunity template with the task, direct evidence, alternative checks, estimated-versus-observed demand, economic assumptions, strongest counterexample, decision and restart condition. Attach only shareable public sources or voluntarily supplied, suitably anonymized observations.

You are done when another reader can explain why the chosen action follows from the evidence, including a NO_GO. Do not use a score to hide a missing premise.

Failure limits

Low difficulty does not prove a gap; a forum result does not make a query easy; a price on a competitor page does not verify sales. Search volume is not obtainable clicks, visitors or page views. Failure to get a metric is not zero demand. Reopen the candidate only when a specified new task or data source changes the decision.

Bounded agent task

“Organize these supplied public sources into one opportunity card. Deduplicate original cases, distinguish actual tests from advertised features, state every unknown, and recommend investigate, prototype or stop. Do not invent demand, tool output, rankings or revenue, and do not register, purchase, post or upload user files.”

Public sources

• Learn about user needs (S01; checked 2026-10-10).
• SEO Starter Guide (S02; checked 2026-10-10).
• Performance report (S03; checked 2026-10-10).
• FAQ about Google Trends data (S07; checked 2026-10-10).
• Revenue per thousand impressions (S08; checked 2026-10-10).

Back to lesson 1