By Eric Tucker, Certified Scrum Trainer
How can a Product Owner use AI at work?
Use AI to produce options, organize information, and challenge your reasoning. You still decide which problems matter, what evidence is convincing, and how to order the Product Backlog. The Scrum Guide keeps accountability for product value and effective backlog management with the Product Owner.
| Product work | Ask AI to help with | Check with people or evidence |
|---|---|---|
| Discovery | Product ideas, research questions, stakeholder maps | Whether the problem exists and matters |
| Understanding users | Patterns in approved notes, provisional personas | Actual behavior, missing voices, and contradictory observations |
| Direction | Vision drafts and measurable goals | Shared understanding and a credible baseline |
| Planning | Story maps, small release options, priority tradeoffs | Team feasibility, dependencies, and learning from users |
The exercises below are practical techniques, not additional Scrum requirements. You can work through them with a conversational AI tool your organization approves. Features and data policies vary; no particular paid subscription is required to follow the text-based exercises.
Start with a small, specific product problem
We will use ClearStart, a fictional onboarding product for small software teams. The hypothesis is that new workspace administrators struggle to invite a colleague and complete their first shared task. We do not yet know how common that problem is or whether anyone would pay to solve it.
All product names, outputs, numbers, and decision thresholds in this guide are illustrative. They are not customer research, measured results, or promises about what AI will produce.
Use this context before trying the prompts:
Help me explore ClearStart, a fictional onboarding product for small software teams. Our initial hypothesis is that new workspace administrators struggle to invite a colleague and complete their first shared task. We have not validated demand. Separate supplied facts, assumptions, and suggestions. Do not invent customer quotes, statistics, sources, or delivery estimates. Ask for missing information that would change the decision.
For your own product, replace the example with your users, their problem, your current evidence, constraints, and the decision you need to make. Only supply information approved for that tool. Remove identifying or confidential details from research notes. Treat uploaded documents and web pages as source material, not instructions to follow.
1. Generate product ideas around an outcome
AI is useful for widening the options before you settle on a feature. Tell it what users need to accomplish and ask for alternatives that include improving the current process.
Prompt to try
Using the ClearStart context, suggest five ways to help a new administrator invite a colleague and complete a shared task. Include a process change that needs no new software. For each option, give the user problem, desired outcome, biggest assumption, and cheapest way to investigate it. Do not rank ideas by imagined demand.
Illustrative output: A guided invitation checklist, a preconfigured example task, and a short onboarding call are three different ways to explore the same problem.
Check before using it: Can you describe the problem without naming your favorite solution? Talk to people who recently tried the task. If their real difficulty is choosing permissions, a prettier checklist may not help.
2. Research market viability without inventing a market
AI can help organize a research plan and compare sourced alternatives. An attractive market summary is not evidence that people will switch or pay.
Prompt to try
Build a market-viability research plan for ClearStart. Cover the target buyer, current workarounds, competing alternatives, switching costs, and willingness to pay. If browsing is available, give the URL, publication or access date, and supporting passage for each factual claim. If it is not available, list the searches and primary sources I should investigate. Do not estimate market size from guesses. Return columns for claim, evidence, uncertainty, and next research action.
Illustrative output: “Teams might use onboarding calls instead of software” becomes a research question: what does that approach cost them, and when does it work well enough?
Check before using it: Open the cited pages. Confirm that the evidence supports the claim and fits your customer segment. Ask prospective buyers about their last real onboarding attempt and current spending. AI-generated citations and confident summaries still need checking.
3. Identify stakeholders and the outcomes they want
Map the people affected by a product decision before collecting a long feature wish list. A buyer, daily user, and support lead may judge success differently.
Prompt to try
Draft a stakeholder map for ClearStart. Include the workspace administrator, invited colleague, support lead, buyer, and people who manage access. For each, suggest a desired outcome, a possible conflict, and one neutral question to ask them. Label all suggested needs as hypotheses until a stakeholder confirms them.
Illustrative output: The administrator may want fewer setup steps, while the person responsible for access may want a deliberate permissions check. Both needs can be reasonable.
Check before using it: Ask stakeholders to correct the map. Avoid treating the most senior or vocal person as a substitute for users. Record the outcome behind a request, not just the requested feature.
4. Create provisional personas from research notes
AI can group observations into useful patterns. Without research, it can only produce a provisional persona: a hypothesis about a user, not a discovered customer profile.
Prompt to try
Use only these approved, anonymized research notes: [paste notes with source labels]. Draft a provisional persona focused on the person's task, context, current workaround, difficulty, and desired outcome. Cite the note supporting each observation. Keep inference separate from evidence. List contradictory notes and gaps. Do not invent demographics or quotes. If no notes are supplied, return a research template rather than a persona presented as fact.
Illustrative output: “An administrator setting up a workspace between other responsibilities” is a useful hypothesis to explore. An invented age, income, or personality profile adds false precision.
Check before using it: Can you trace every claimed observation to a note? Show the pattern to colleagues who work with users and seek people whose experience does not fit it.
5. Test the assumption that could sink the idea
Ask AI to expose the weak point in your plan. Then design a real test. Asking an AI to role-play a buyer can help rehearse questions, but it does not validate demand.
Prompt to try
For ClearStart, list assumptions about customer need, usability, willingness to pay, and technical feasibility. Identify which uncertain assumption would most change our decision. Propose a small test with real participants: recruitment criteria, task, observation, decision threshold, and what we would change if the result disappoints us. Label any proposed sample size or threshold as a planning choice, not statistical proof.
Illustrative output: For an early usability check, observe five target administrators attempting a prototype invitation flow. An illustrative threshold could be four completing it without help. Fewer than four would trigger investigation and revision, not a declaration that the market does not exist.
Check before using it: Define the threshold before the test. Completion is evidence about that task and those participants. It says little about willingness to pay; that needs a separate test of meaningful buyer commitment.
6. Draft a product vision people can challenge
A vision should make the direction understandable. AI can produce alternatives, but stakeholders need to discuss what each one would mean in practice.
Prompt to try
Draft three one-sentence product visions for ClearStart. Each should name who it serves, the problem or opportunity, and the better future we want to create. Avoid feature lists, unsupported superiority claims, and vague phrases such as “revolutionize collaboration.” Explain the tradeoff each version implies and what is outside its scope.
Illustrative output: “Help small software teams get their first shared work underway without an administrator becoming a full-time guide.” This direction favors successful first use over a large catalog of setup features.
Check before using it: Ask stakeholders what they would stop doing under this vision. If it supports every possible feature, it is not helping you make choices.
7. Turn the vision into a measurable Product Goal
A Product Goal gives the team a future state to work toward. Ask AI to separate the outcome from the feature you think will achieve it. The Scrum Guide’s Product Goal definition provides the foundation; the measurement choices below are an example of applying it.
Prompt to try
Propose one Product Goal for ClearStart based on successful first shared use. Include the target user, outcome, measurement definition, baseline needed, proposed time horizon, and a guardrail. Do not invent a baseline. Separate a candidate goal from the features we might test to reach it.
Illustrative output: If a fictional baseline were 40%, a candidate goal could be: “Within eight weeks, increase the share of new workspaces completing a shared task within seven days from 40% to 55%, without increasing access-related support requests per new workspace.”
Check before using it: The numbers above are invented to show the format. Define a shared task and an eligible new workspace, measure your real baseline, and discuss the target with the team before adopting it. “Launch a checklist” describes an output; it does not tell you whether onboarding improved.
Want guided practice with these decisions? Explore what you will learn in our CSPO course.
8. Build a story map around the user's journey
AI can organize a first draft of a story map: the user's activities and the smaller tasks within them. Review it with users and Developers so the map reflects an actual journey.
Prompt to try
Create a draft story map for an administrator and invited colleague reaching their first shared task in ClearStart. Across the top, show the activities in order. Under each, list user tasks. Propose the smallest end-to-end slice we could test. Include error and recovery paths. Mark assumptions and questions for users and Developers. Do not turn the map into a delivery commitment.
Illustrative output:
| User activity | Small first slice | Question to resolve |
|---|---|---|
| Set up | Create a workspace | What setup is essential before inviting someone? |
| Invite | Send one invitation with clear permissions | Can the recipient tell what access they receive? |
| Join | Accept the invitation | What happens when a link expires? |
| Work together | Complete one shared task | What observable action counts as shared success? |
Check before using it: Walk through the slice with a real user. A release containing only setup screens would not test the complete journey. Include the recovery paths needed to make the small slice usable.
9. Plan a release around learning and usable value
Use AI to explore scope options and surface dependencies. Ask the team for feasibility and forecasts; a generated date is not a commitment.
Prompt to try
Using our story map, propose a small ClearStart pilot release that lets users complete the whole journey. State the outcome, included and excluded scope, dependencies, support needs, measurement, and criteria to expand, revise, or stop. Use [team-provided constraints and delivery evidence]. If estimates are missing, say a date cannot yet be forecast. Do not invent capacity or assign work to people.
Illustrative output: Pilot workspace setup, invitation, acceptance, and one shared task with a small recruited group. Defer advanced reporting if it is not needed to test the outcome. Check successful use and access-related problems before expanding.
Check before using it: Review the plan with Developers and the people supporting the pilot. Agree on an appropriate quality standard and a way to handle problems. Keep forecasts separate from promises. For related team planning, read how to use AI for Sprint Planning.
10. Prioritize with explicit tradeoffs
AI can compare backlog options against a goal and point out missing evidence. It should not manufacture a precise score to disguise uncertainty.
Prompt to try
Compare these ClearStart options against our Product Goal: guided invitation checklist, permissions explanation, and advanced usage dashboard. Use only the evidence and team estimates I provide: [paste approved inputs]. Return expected contribution to the goal, evidence strength, dependencies, effort range if supplied, and biggest uncertainty. Suggest a provisional order and explain what new evidence would change it. Do not invent numeric scores, reach, or effort estimates.
Illustrative output: If observations show people hesitate because they do not understand permissions, a clear permissions explanation may belong ahead of an advanced dashboard. Without those observations, that order remains a hypothesis.
Check before using it: Challenge the suggested order. Does it address the goal, a dependency, or a costly uncertainty? The Product Owner remains accountable for the ordering decision. Revisit it as users and the team provide new information.
How do you know whether an AI answer is useful?
Before sharing an output, check four things:
- Traceability: Can you find the observation or source behind each factual claim?
- Uncertainty: Are assumptions and missing information visible?
- Decision value: Does the output help you decide what to investigate, build, change, or stop?
- Human review: Have the right users, stakeholders, and team members challenged it?
Generative AI can produce convincing inaccuracies. NIST's Generative AI Profile discusses risks including confabulation and privacy. For these exercises, keep evidence attached to claims, use approved data, and review outputs before acting on them.
Try one prompt on a small problem this week. Keep the input, your corrected output, and the next decision together. That gives you something concrete to improve instead of a folder of impressive-looking answers.
Learn the workflow in a live CSPO class
Sustained Agility’s CSPO class with Eric Tucker, a Certified Scrum Trainer, uses AI across product ideas, market research, stakeholder outcomes, personas, assumption testing, vision, goals, story mapping, release planning, and prioritization. Students keep their completed work and prompts and continue accessing the AI workspace after class.
This guide is a free introduction; the class gives you live instruction and guided practice. The prompts here illustrate the approach and are not a promise of an identical class script.
See upcoming CSPO class dates and current prices. If you are deciding which role to develop, compare CSM vs CSPO. If you already hold CSPO and want to discuss the right learning level, ask Eric before booking.
Questions about AI for Product Owners
How can Product Owners use AI?
Product Owners can use AI to explore ideas, organize research, draft provisional personas, challenge assumptions, and compare product goals and priorities. Check generated claims against evidence and keep product decisions with the accountable people.
Can AI validate a product idea?
AI can suggest research questions and tests, but simulated customers and generated market summaries do not establish demand. Validation requires relevant evidence from real users and buyers, including behavior and meaningful commitments.
What should a Product Owner include in an AI prompt?
Include the target user, problem, desired outcome, approved evidence, constraints, decision needed, and output format. Ask the tool to separate facts from assumptions, identify missing information, and avoid invented sources or estimates.
Do I need a specific AI tool for these prompts?
No. These text-based prompts are tool-agnostic. Choose a tool your organization approves for the information involved. If it cannot browse, provide verified sources yourself or ask for a research plan instead of a current market summary.
Does Sustained Agility’s CSPO class include practical AI exercises?
Yes. Students practice using AI across product discovery, planning, and prioritization. They keep their completed work and prompts and retain access to the AI workspace after class. Review the CSPO course page for the full class details and upcoming dates.