PSPO-AI Product Discovery Prompts: Evidence, Risk, and Human Decisions

The easiest product prompt asks for ideas. The more valuable prompt helps a Product Owner inspect evidence without pretending the model has customer knowledge it was never given.

Scrum.org positions PSPO-AI Essentials around practical AI use in market understanding, personas, product vision, prototyping, testing, routine work, and collaboration, with explicit attention to security and ethics. The prompts below preserve those boundaries.

Before prompting: create an evidence boundary

• Remove personal, confidential, client, credential, and commercially restricted information.
• State which supplied material the model may use.
• Ask it to distinguish observations from interpretations.
• Prevent unsupported claims by requiring a source reference for each theme.
• Name the decision that remains with the Product Owner and team.

Prompt 1: synthesise interview observations

Use only the anonymised interview notes below. Group observations by customer problem. For each group, cite the note identifiers, show contradictory evidence, and mark themes supported by fewer than three observations. Do not create personas or recommend features.

This prompt narrows the task to synthesis. It stops the model from turning sparse interviews into a confident market conclusion.

Prompt 2: expose assumptions in an opportunity

Review this opportunity statement. List assumptions about the customer, problem frequency, current alternative, business outcome, feasibility, and ethical risk. Label each assumption as evidenced, partly evidenced, or unsupported using only the supplied material.

The output becomes an assumption inventory, not a validation result. The team still decides which uncertainty deserves investigation.

Prompt 3: design a small learning experiment

Propose three reversible ways to test the riskiest assumption without building the complete solution. For each, state the participant group, evidence produced, cost, potential harm, stop condition, and what result would fail to reduce uncertainty.

Review recruitment fairness, consent, privacy, and whether the experiment could mislead participants. Fast learning is not a reason to lower ethical standards.

Prompt 4: challenge a proposed Product Backlog item

Act as a critical reviewer. Identify unclear outcomes, hidden assumptions, affected users, dependencies, operational concerns, and missing acceptance evidence. Ask questions only; do not rewrite or estimate the item.

Questions can prepare refinement, but the model should not become the authority that defines value or decides readiness.

Prompt 5: compare product options without fake precision

Compare the three options against the stated outcome, evidence quality, reversibility, expected learning, delivery risk, and potential harm. Do not calculate a total score. Show where the available evidence cannot support a comparison.

Removing the total score prevents arbitrary weights from producing a false winner. Leaders can see the trade-off and own the decision.

Prompt 6: prepare a stakeholder decision brief

Using only the approved evidence summary, draft a decision brief with the outcome sought, options considered, supporting and contradictory evidence, material risks, unresolved questions, and the decision owner. Leave the recommendation blank.

Leaving the recommendation blank is deliberate. The model can organise information while the accountable people discuss trade-offs. After the meeting, add the chosen action, rationale, dissent, and review date to the human-owned record.

Review every output through four checks

Check	Product Owner question
Evidence	Can I trace each material statement to a reliable source?
Risk	Who could be harmed if the output is wrong or biased?
Security	Was the input permitted, minimised, and handled appropriately?
Accountability	Which human decides, records the rationale, and monitors the result?

Record the decision after the prompt

Keep the prompt, approved input summary, model or tool used, important output, verification performed, decision owner, chosen action, and review date. This record is more useful than saving a polished answer with no evidence trail.

The free PSPO-AI prompt and risk checklist provides a reusable review page. To practise these decisions in an official instructor-led setting, see Professional Scrum Product Owner - AI Essentials training.