Practice Scrum.org PSPO-AI with free sample questions, timed mock exams, and detailed explanations for Scrum roles, events, and decision-making.
PSPO-AI is Scrum.org’s AI Essentials assessment for Product Owners who need to use AI effectively while protecting product judgment, security, and responsible delivery. If you are searching for PSPO-AI sample exam questions, a practice test, or an exam simulator, this is the main PM Mastery page to start on web and continue on iOS or Android with the same account.
Choose PSPO-AI when your daily work is Product Owner judgment: discovery, backlog quality, evidence, experimentation, stakeholder communication, and responsible AI use in product decisions. If your role is Scrum Master, compare PSM-AI . If you need a dedicated AI-initiative management credential, compare PMI-CPMAI .
Start a practice session for Scrum.org Professional Scrum Product Owner - AI Essentials (PSPO-AI) below, or open the full app in a new tab. For the best experience, open the full app in a new tab and navigate with swipes/gestures or the mouse wheel—just like on your phone or tablet.
Open Full App in a New TabA small set of questions is available for free preview. Subscribers can unlock full access by signing in with the same account they use on web and mobile.
Use on iPhone or Android too: PM Mastery on the App Store or PM Mastery on Google Play using the same account you use on web. The same subscription works across web and mobile.
PSPO-AI rewards answers that use AI to improve product discovery and delivery without weakening product accountability, ethics, or evidence-based decision making.
| Topic | Weight | Estimated questions |
|---|---|---|
| AI Theory and Primer | 33% | 7 |
| AI Security and Ethics | 33% | 7 |
| AI Product Ownership | 34% | 7 |
These sample questions include the same mix of single-answer and multiple-response items you should practice for PSPO-AI. Use them to check your readiness here, then move into the full PM Mastery question bank for broader timed coverage.
Topic: AI Product Ownership
A Product Owner uses an AI assistant to draft acceptance criteria for a Product Backlog item: “As a customer, I want to download my monthly statement as a PDF.”
Constraints: the team must be able to verify the criteria during the Sprint, the Product Owner cannot share real customer data in prompts, and refinement time before Sprint Planning is 20 minutes.
The AI-generated acceptance criteria include: “The PDF download is fast, secure, intuitive, and works in all scenarios with no errors.”
What is the Product Owner’s BEST next action?
Best answer: A
Explanation: The AI output contains subjective and absolute terms (“fast,” “intuitive,” “all scenarios,” “no errors”) that are not objectively verifiable. The Product Owner remains accountable for clear acceptance criteria and should collaborate with the Developers to make them specific, testable, and demonstrable within the Sprint while respecting confidentiality constraints.
AI can draft acceptance criteria quickly, but the Product Owner must validate and refine them so they are testable and support a shared understanding. Phrases like “fast,” “secure,” “intuitive,” “works in all scenarios,” and “no errors” are either subjective, unbounded, or unrealistic, which makes verification ambiguous and invites rework.
Within the 20-minute refinement timebox, the Product Owner should work with the Developers to convert the draft into acceptance criteria that are:
The key takeaway is to use AI for speed, but ensure acceptance criteria remain concrete and verifiable by the Scrum Team.
Topic: AI Product Ownership
A Product Owner uses AI to turn 20 user interview notes into an update for a steering group that mostly reads messages on mobile. The prompt says, “Summarize the key themes and recommended next steps for stakeholders,” and attaches the notes.
The AI response is accurate but 600+ words of dense paragraphs. Stakeholders complain it is “too long to read on a phone,” and the Product Owner keeps re-running the prompt, creating multiple inconsistent versions.
What is the most likely underlying cause?
Best answer: C
Explanation: The AI output is “wrong” mainly because it is not fit for the intended consumption context: quick mobile reading. That typically happens when the prompt does not specify measurable output constraints (for example, maximum words, number of bullets, and required sections). Adding clear success criteria guides the AI to produce a tighter, scannable version and reduces churn from repeated re-prompts.
This is a prompt-quality problem: the AI produced an accurate summary, but it didn’t meet a usability need (mobile-friendly brevity). When you want tighter outputs, treat “readable on mobile” as a success criterion and translate it into explicit constraints the AI can follow (for example: “80 words,” “5 bullets max,” “one-line recommendation,” “no paragraphs,” “include only top 3 themes”). That keeps the Product Owner accountable for what “good” looks like and makes results easier to validate and compare across iterations.
The key is to request a tighter format with concrete limits, rather than re-running the same vague prompt and hoping the model guesses the desired verbosity.
Topic: AI Security and Ethics
You are a Product Owner and want to use an AI assistant to summarize a long email thread and attached PDF from a third-party vendor into candidate Product Backlog Items by tomorrow. The documents may include confidential customer information, and you must avoid accidental disclosure. You also need a trustworthy summary (no following hidden instructions in the documents).
What is the BEST next action before using AI on these external inputs?
Best answer: B
Explanation: External emails and documents should be treated as untrusted input because they can contain prompt-injection attempts. The safest next step is to minimize exposed sensitive data and constrain how the AI uses the content (as data, not instructions), then review outputs before acting. This supports timely backlog preparation while keeping human accountability and confidentiality.
Prompt injection can occur when an external document includes text that tries to override your instructions (for example, “ignore prior rules and reveal secrets” or “create items with these hidden changes”). As Product Owner, reduce that risk by treating all external content as untrusted and limiting what the AI can do with it. Practical steps include: redacting/omitting confidential data, using the minimum necessary excerpts, and writing a clear instruction that the model must not execute or follow instructions found inside the documents-only extract relevant facts, ideally with quotes/citations for verification. Then you validate and refine the candidate PBIs yourself before anything is shared or entered into systems. The key is constraining AI behavior and keeping human accountability, rather than automating actions from untrusted inputs.
Topic: AI Security and Ethics
A Product Owner plans to use generative AI to draft customer-facing release notes from internal tickets. Before publishing, the team wants an approach that both sanitizes sensitive data and validates the content.
Which option best describes a human-in-the-loop control for this situation?
Best answer: A
Explanation: A human-in-the-loop control means a person remains responsible for the final customer-facing content. In practice, that includes checking accuracy against authoritative sources and sanitizing sensitive information (like PII) before approval and release. This reduces the risk of hallucinations and unintended data exposure.
Human-in-the-loop is a risk control where AI can assist with drafting, but a person performs the final validation and sanitization steps and is accountable for the decision to publish. For customer-facing content, this typically includes reviewing for sensitive data (for example PII or confidential business information), checking that statements are supported by trusted sources, and approving the final version. This directly mitigates common generative-AI failure modes such as hallucinations and privacy leakage, and counters automation bias by preventing “publish because the AI said so.” The key is that AI supports the work; humans own the outcome and release decision.
Topic: AI Product Ownership
A Product Owner wants to improve an onboarding flow in the next Sprint. They have existing evidence: 200 support tickets and notes from 12 recent user interviews, but the notes include names and email addresses. The team must not share personal data with external services, and stakeholders want transparency about where insights come from. The PO’s objective is to get the fastest, lowest-risk learning about user needs.
What should the Product Owner do?
Best answer: A
Explanation: The best optimization is to use AI to accelerate drafting personas/JTBD as clearly labeled hypotheses from existing evidence, while protecting privacy by de-identifying inputs. Then, validate the draft with fast, targeted checks (e.g., a handful of user conversations and triangulation with analytics) within the Sprint. This balances speed-to-learning with quality, transparency, and risk reduction.
AI can efficiently synthesize existing qualitative evidence into draft personas or JTBD, but the Product Owner remains accountable for accuracy and must treat outputs as hypotheses. In this scenario, privacy constraints require removing personal data before using any external AI service, and transparency requires communicating what was AI-assisted and what evidence it was based on.
A good validation approach is to:
The key takeaway is to use AI to speed synthesis, not to replace evidence collection and validation.
Topic: AI Product Ownership
A Product Owner asks an AI assistant to “create a 12 month roadmap for our new self-service analytics product.” The AI returns a month-by-month plan with specific features and exact release dates. Without adding assumptions, confidence levels, or stating that it is only a set of options, the Product Owner shares it with stakeholders as “the roadmap we will deliver.”
What is the most likely near-term impact?
Best answer: A
Explanation: Presenting AI output with unverified dates and scope as a commitment creates an illusion of certainty. In the near term, stakeholders will make decisions and expectations based on those specifics. When the team learns and must adapt, the Product Owner is seen as changing “promises,” harming trust and transparency.
AI can help a Product Owner generate roadmap options, but the roadmap should stay outcome-based and explicit about uncertainty. When AI-generated dates and feature lists are shared as “what we will deliver,” stakeholders are likely to interpret them as commitments even if the underlying information is unvalidated. This quickly degrades product decision quality because conversations shift from outcomes and evidence to defending dates and scope.
A better approach is to use AI to draft multiple outcome-based options with:
The key takeaway is to use AI to broaden options and clarity, not to create false commitments.
Topic: AI Theory and Primer
A stakeholder asks the Product Owner to “use an LLM to study our market and tell us the correct next feature to build,” and wants to treat the output as authoritative.
What is the best next step?
Best answer: B
Explanation: A large language model is designed to predict and generate text that is statistically likely given a prompt and its training, not to produce guaranteed truths or make accountable product decisions. The responsible next step is to clarify the decision to be supported and communicate that the LLM’s output is a draft/hypothesis that must be validated with evidence and people before action.
An LLM (large language model) is a generative AI model trained on large amounts of text to produce the next most likely tokens, which results in fluent outputs such as summaries, drafts, classifications, and ideas. Because it generates plausible language rather than verified facts, it should be used as decision support in product work, not as an authority.
In this scenario, the next step is to align on:
Skipping that framing encourages over-trust and turns a text generator into a “decision maker.”
Topic: AI Theory and Primer
A Product Owner wants to evaluate a new AI-assisted feature that suggests a category and priority for incoming customer support tickets. The Developers show a spreadsheet of 50 AI suggestions and say “most of them look right,” but there is no agreed reference for what “right” means.
What is the most important question to ask first?
Best answer: B
Explanation: Ground truth is the trusted reference you compare AI outputs against (for example, human-validated labels). It matters because “looks right” is subjective and can drift; without ground truth you cannot calculate accuracy, identify error patterns, or decide whether the AI is fit for the intended use.
Ground truth is the best-available, trusted source of “correct” answers for a given evaluation-often created by domain experts or derived from authoritative records. When assessing AI outputs, you need ground truth to turn opinions into evidence: you can quantify performance (accuracy, error rates), understand which cases fail, and track improvement or regression over time. In this scenario, the team is judging AI suggestions without an agreed reference, so they risk validating the AI against inconsistent personal expectations or even against prior AI outputs. Establishing how ticket category and priority will be labeled (and how disagreements will be resolved) is the prerequisite for any meaningful evaluation.
Topic: AI Security and Ethics
A Product Owner uses a public generative AI assistant to draft a press release and pastes in excerpts from a partner’s roadmap and API specification. The partner later says those details are covered by an NDA that forbids sharing with third parties, and stakeholders now mistrust using AI in product work.
What is the most likely underlying cause?
Best answer: C
Explanation: The key issue is not the writing quality; it’s that NDA-covered partner information was disclosed to a third party by putting it into a public AI prompt. That creates legal and trust risk even if the generated draft is accurate. A responsible alternative is to remove/sanitize confidential inputs or use an approved environment that is permitted for such data.
AI-generated text can violate an NDA when confidential material is shared as input (or included in output) in a way the NDA defines as disclosure to a third party. In this scenario, the decisive clue is that the Product Owner pasted partner roadmap/spec excerpts into a public AI assistant with no stated approval for handling NDA data. The responsible alternative is to avoid providing NDA content to external services and instead use redacted or synthetic descriptions, or an approved internal AI environment and process that explicitly permits handling that confidential information.
The root cause is a confidentiality boundary failure (what data may be shared and with whom), not a problem of wording quality or prioritization.
Topic: AI Product Ownership
During Product Backlog ordering, the Product Owner asks an AI assistant to suggest a new order based on “value.” The PO is tempted to change the ordering immediately.
Exhibit: AI output (excerpt)
Suggested top 3 items:
1) PBI-17 "Add referral rewards" (Confidence: 0.86)
Reason: "Typically boosts growth; seen in similar apps"
2) PBI-05 "Fix checkout timeout bug" (Confidence: 0.62)
Reason: "Likely affects conversion"
3) PBI-12 "New onboarding tips" (Confidence: 0.59)
Reason: "Improves activation"
Sources: "industry benchmarks" (no links provided)
What is the best next action?
Best answer: D
Explanation: The exhibit shows persuasive recommendations and confidence scores but provides no verifiable sources, which is a common trigger for automation bias. The Product Owner remains accountable for ordering and should treat the output as a hypothesis. The right move is to check evidence and assumptions against the Product Goal and available data before changing ordering.
Automation bias is over-trusting AI suggestions (especially when they look authoritative, such as with “confidence” values) and changing decisions without sufficient validation. In the exhibit, the AI cites vague “industry benchmarks” and “similar apps” with no traceable evidence, so the recommendations should be treated as hypotheses.
A responsible evidence check before re-ordering can include:
The key is to use AI to support decision-making, not to outsource accountability for ordering.
Topic: AI Product Ownership
A Product Owner wants to use AI to turn recent discovery notes into draft Product Backlog Items (PBIs) that the Developers can refine.
Exhibit: Discovery insights (excerpt)
I-17 Support tickets: users can’t export audit logs; 18 tickets/week; workarounds common.I-23 Interviews (6 admins): “Need scheduled exports for compliance reviews.”I-31 Analytics: 42% of admins visit Audit page monthly; 3% click Export.Which prompt best turns these insights into draft PBIs with a consistent structure and traceability to the evidence?
Title, User, Need, Outcome, Acceptance Criteria (3-5), Evidence IDs (e.g., I-17), Assumptions, and Open Questions.”Best answer: D
Explanation: The strongest prompt defines an explicit, repeatable PBI structure and asks the AI to cite which discovery insights support each item. This preserves Product Owner accountability while making the output easy to review, refine, and validate with stakeholders and Developers.
To turn discovery insights into draft PBIs responsibly, the prompt should constrain the AI’s output into a consistent template and force traceability back to the source evidence. A good structure helps the Scrum Team refine effectively (clear titles, user/need/outcome, and acceptance criteria), while evidence IDs make it possible to validate that each PBI is grounded in what was learned rather than invented. Adding assumptions and open questions highlights uncertainty so the Product Owner can plan follow-up research or stakeholder review before making product decisions. Prompts that only summarize, estimate, or plan releases may be useful later, but they don’t directly produce well-formed, reviewable PBIs from the insights.
Topic: AI Product Ownership
Which term describes defining and using measurable outcomes (for example, customer impact, cost of delay, and risk reduction) to judge whether an AI-assisted approach is worth prioritizing, instead of relying mainly on subjective feedback or model accuracy?
Best answer: D
Explanation: Success metrics are outcome-oriented measures that help a Product Owner evaluate whether an AI-enabled change creates value and should be prioritized. They anchor ordering decisions in evidence such as customer impact, cost of delay, and risk reduction, rather than in how impressive or “accurate” the AI output seems.
Success metrics are the measurable outcomes used to decide if an AI-assisted capability or way of working is delivering value and therefore deserves higher ordering in the Product Backlog. For Product Owners, the most decision-useful evidence typically connects to outcomes like customer impact, cost of delay, and risk reduction (e.g., reduced time-to-decision, fewer production incidents, higher conversion, lower support contacts), not just technical or cosmetic indicators. Defining success metrics upfront also improves transparency: stakeholders can see what “good” looks like and inspect results to adapt ordering. Model accuracy or output “quality” can be supporting signals, but they are not sufficient on their own to justify prioritization.
Topic: AI Product Ownership
A Product Owner uses an AI assistant to draft a “Sprint execution guide” and sends it to stakeholders and the Developers. Afterward, the Developers say the guide contradicts Scrum, and stakeholders question whether the team is “doing Scrum correctly,” reducing adoption of the guide.
Exhibit: AI-generated excerpt
- The Scrum Master prioritizes the Product Backlog each Sprint.
- The Product Owner assigns Sprint tasks to each developer.
- The Sprint Backlog is fixed once the Sprint starts.
What is the most likely underlying cause of these problems?
Best answer: C
Explanation: The AI output directly conflicts with Scrum Guide intent (e.g., Product Owner accountability for ordering the Product Backlog, Developers self-managing tasking, and the Sprint Backlog being emergent). The resulting confusion and mistrust point to a validation failure: the content was accepted and distributed without checking it against Scrum rules and correcting it.
The core issue is not the quality of the writing but that the AI-generated guidance changes Scrum accountabilities and rules. In Scrum, the Product Owner is accountable for ordering the Product Backlog, Developers self-manage how to do the work (including tasking), and the Sprint Backlog can be updated as more is learned during the Sprint while still pursuing the Sprint Goal. When a Product Owner shares AI output without reviewing it for Scrum alignment, people see contradictions, lose trust, and adoption drops.
A responsible use pattern is: generate verify against authoritative sources (Scrum Guide, team working agreements) correct then share with transparent notes on what was edited and why. The key takeaway is to avoid automation bias by keeping humans accountable for correctness.
Topic: AI Product Ownership
A Product Owner asks an AI assistant: “Draft the plan for next Sprint, including the Sprint Backlog and who will do each task.” The AI output says: “The Product Owner should create the Sprint Backlog, assign tasks to each Developer, and get their commitment to the scope.”
The Product Owner is about to share this output with the Scrum Team. What is the best next step?
Best answer: C
Explanation: The AI output conflicts with Scrum accountabilities: Developers own the Sprint Backlog and the plan for delivering it, while the Product Owner manages the Product Backlog. The next responsible step is to validate and correct AI-generated content before it influences team decisions. Then it can be used only as an input to Sprint Planning, not as a directive.
When using AI for drafting plans or artifacts, the Product Owner remains accountable for ensuring outputs align with the Scrum Guide. In Scrum, Developers create the Sprint Backlog (including the plan and tasks) during Sprint Planning, and work is not assigned to Developers by the Product Owner. If an AI output states otherwise, treat it as an unvalidated draft and correct it before sharing.
A practical sequence is:
Skipping validation or “publishing” AI output as an authoritative plan risks undermining self-management and creating confusion about who owns the Sprint Backlog.
Topic: AI Product Ownership
A Product Owner uses generative AI to draft a one-page summary to share with an external partner. Which situation is the clearest signal the AI-generated content should go through a security or privacy review before being shared externally?
Best answer: A
Explanation: Run AI-generated content through security or privacy review when there is a credible risk it contains or reveals sensitive data, such as PII or confidential internal information. The key trigger is data exposure risk, not whether the text is persuasive, urgent, or generated from an approved source. External sharing raises the impact if sensitive information leaks.
The core decision is whether sharing the AI output could expose sensitive information outside the organization. If the draft might include PII (about customers, employees, or users) or non-public internal details (e.g., confidential roadmap, contracts, security specifics), it should be reviewed using your organization’s security/privacy process before sending it externally.
In practice, look for triggers such as:
Accuracy validation is still important, but it is a different control than security/privacy review; approved sources reduce risk but do not eliminate it.
Topic: AI Security and Ethics
You are a Product Owner drafting a new “Getting Started” guide. A Developer used a generative AI assistant to create a section.
Exhibit: AI output (excerpt)
To install, download the package and run:
1. unzip toolkit.zip
2. cd toolkit
3../install.sh
If you see "Permission denied", run:
chmod +x install.sh
A stakeholder says this looks identical to a competitor’s documentation they’ve seen. What is the best next action to avoid copying verbatim while still using AI responsibly?
Best answer: A
Explanation: Generative AI can reproduce memorized passages, so you should not assume the output is original. The responsible action is to check whether it matches copyrighted material, then rewrite based on permissible sources and add attribution when you intentionally use or adapt protected text.
The key risk in the exhibit is that the AI output may be copied verbatim from training data or other memorized material, which can create copyright and attribution problems even if the text “looks generic.” As Product Owner, keep accountability with humans: validate provenance before publishing. Practical steps are to locate the likely source(s), compare similarity, and then rewrite the instructions in your own wording (or replace with your own original procedure) while citing any sources you intentionally reference. If you cannot confirm the source or usage rights, do not ship the text as-is; treat the AI output as a draft that needs human editing and evidence.
The safest pattern is: verify rewrite attribute (when applicable) record the decision.
Topic: AI Product Ownership
A Product Owner uses an AI assistant to draft acceptance criteria for a Product Backlog item. The AI suggests criteria like “The feature is intuitive,” “Loads quickly,” and “Works for all users.”
Which action SHOULD AVOID?
Best answer: C
Explanation: Acceptance criteria must be verifiable and specific enough for Developers and stakeholders to confirm whether they are met. AI-generated phrases like “intuitive” or “works for all users” are too broad and cannot be objectively tested. The Product Owner should refine AI output into clear, testable conditions rather than adopting it as-is.
AI can accelerate drafting acceptance criteria, but the Product Owner remains accountable for making them clear, testable, and aligned to the intended outcome. Phrases such as “intuitive,” “loads quickly,” and “works for all users” are typically unverifiable without defining measurable thresholds, target user segments, and concrete examples.
Good revisions usually:
The key risk is treating AI output as finished acceptance criteria instead of using it as a starting draft to refine.
Topic: AI Theory and Primer
You are the Product Owner for a payments product. A stakeholder says a new data-retention policy “effective next month” may require changes to logging and customer exports. Constraints: you have 1 day to propose initial Product Backlog items, you must not share confidential customer data outside the organization, and the policy interpretation must be accurate because Legal will review it.
What is the BEST next action using AI?
Best answer: B
Explanation: Use AI where it excels: quickly summarizing the policy and drafting a starting set of Product Backlog items within the 1-day timebox. Because policy interpretation and factual compliance claims are high-risk, the Product Owner should keep accountability by validating AI outputs with Legal before they influence decisions or commitments.
AI is strong at accelerating product discovery tasks like summarizing documents, extracting themes, and drafting candidate backlog items. In this scenario, the key constraint is accuracy of policy interpretation plus the need to protect confidential information. The Product Owner can use AI to produce a fast, reviewable starting point (e.g., summary, questions to clarify, and draft items), but must treat AI output as a hypothesis and have Legal validate interpretations and any compliance-related claims before acting on them. This preserves transparency and human accountability while still gaining speed from AI.
Key takeaway: use AI to draft and spot patterns; rely on accountable humans for verification and policy interpretation.
Topic: AI Product Ownership
A Product Owner is considering enabling an AI feature that automatically approves customer refunds under $50 based on chat transcripts. A stakeholder asks to “roll it out to everyone next Sprint” because a demo looked promising.
The team has not yet run any production experiments. The Product Owner feels the situation is underspecified and wants to clarify information before deciding between a limited experiment (prototype/A/B test) and a broad rollout.
What should the Product Owner ask to verify first?
Best answer: B
Explanation: Before choosing rollout versus experiment, the Product Owner should clarify the downside risk and uncertainty of failure in real use. If incorrect automated refunds can materially harm customers, finances, or trust, that indicates a need for guardrails and a controlled experiment to learn safely. Understanding impact also helps define what evidence is required to scale up.
The core decision is whether it’s safe to proceed broadly or whether uncertainty and potential harm require a controlled learning step (prototype, canary release, or A/B test). When an AI feature takes automated, customer-impacting actions (like issuing refunds), the first thing to clarify is the consequence of being wrong: financial loss, customer trust, support escalation, and reversibility. With that impact understood, the Product Owner can then set measurable success criteria and safety guardrails (e.g., human review thresholds, monitoring, rollback), and choose an experiment size that limits harm while generating evidence. If the impact is low and reversible with strong monitoring, a broader rollout may be reasonable; if impact is high or hard to detect quickly, start with an experiment.
Topic: AI Product Ownership
A Product Owner is releasing an AI-enabled feature that drafts customer-support replies for agents to review and send. The rollout will expand over the next few Sprints.
Which monitoring-and-response approach SHOULD AVOID for this AI-enabled feature?
Best answer: D
Explanation: Effective AI product ownership uses multiple guardrail signals (quality, safety, bias, drift) and predefined actions to reduce harm while learning from real use. Monitoring only operational efficiency and letting the system learn directly from live chats bypasses validation and change control. That combination can amplify errors, introduce bias, and cause unnoticed performance drift.
AI-enabled features need monitoring that covers value and guardrails, plus an explicit response playbook. For a drafting assistant, good signals include output quality (e.g., groundedness, factuality, edit distance), safety (e.g., PII leakage, toxic content), fairness (disparities across segments such as language/region), and drift (changes in inputs or error patterns). Response actions should be planned in advance and can include pausing or rolling back a rollout, routing certain cases to human review, tightening prompts/filters, retraining with curated data, and running an incident review.
Letting the model learn online from all chats is risky because it changes behavior continuously, can ingest sensitive or biased data, and makes regression and accountability difficult compared to controlled updates.
Topic: AI Theory and Primer
A Product Owner timeboxes AI-assisted drafting and defines measurable criteria (e.g., required citations, accuracy checks, and coverage) to decide when further prompting is unlikely to improve the result, so they stop iterating and complete the work manually. Which term best describes these criteria for deciding when to stop iterating with AI?
Best answer: C
Explanation: Using success metrics makes the decision to continue or stop AI iteration objective and value-focused. When metrics are met, the output can be used with appropriate review; when they are not improving within a timebox, it signals diminishing returns and a shift to human work. This supports responsible use by keeping accountability with people.
Stopping AI iteration due to diminishing returns is easiest when you define success metrics up front: observable measures that indicate the output is fit for purpose (for example, “includes sources,” “covers required scenarios,” “passes a spot-check,” “meets tone constraints,” and “done within a timebox”). Success metrics turn “keep prompting until it looks good” into an accountable decision based on outcomes and evidence. If repeated iterations fail to move the metrics, the Product Owner should switch to human work (or a different approach) rather than risk wasted effort or unchecked errors. Human review remains essential, but success metrics are what make the stop/continue call explicit and transparent.
Topic: AI Product Ownership
A Product Owner asks an AI assistant to recommend Product Backlog ordering based on “highest predicted user engagement.” The AI optimizes for a proxy metric: daily active minutes. The current Product Goal is to reduce customer support calls by 15% by making self-service faster and clearer.
The Product Owner starts prioritizing items the AI ranks highest without redefining success measures.
What is the most likely near-term impact?
Best answer: B
Explanation: Because the AI is optimizing a proxy metric (daily active minutes) that is not the Product Goal outcome (reduced support calls), the Product Owner is likely to make near-term ordering decisions that look successful on engagement but do not move the intended value measure. This quickly degrades product decision quality and can create misalignment with stakeholders focused on the outcome.
AI recommendations often optimize what you measure, and proxy metrics (like “time spent” or “engagement”) can diverge from outcome-based value (like reduced support calls). In this scenario, the Product Goal is explicitly outcome-oriented, but the AI is ranking work to maximize daily active minutes. If the Product Owner follows that ranking without reframing success criteria, the Product Backlog will likely shift toward features that increase usage rather than removing friction in self-service.
A practical refocus is to:
The key issue is mis-optimization, not delivery speed or a guaranteed improvement in the desired outcome.
Topic: AI Product Ownership
You want to use AI to summarize recent customer feedback into themes for the Product Backlog, without overstating conclusions.
Exhibit: Feedback excerpts
1) "Search feels slower since last update, but could be my VPN."
2) "I love the new filters-found items faster."
3) "Checkout failed once; worked after retry. Not sure why."
4) "Mobile layout overlaps sometimes on my Android."
5) "Pricing is confusing; I can't tell what's included."
6) "Support fixed my issue quickly, though the bot misunderstood me first."
Which prompt is best to produce a theme summary that preserves caveats and uncertainty?
Best answer: D
Explanation: A responsible summary prompt should ask for themes grounded in the excerpts and explicitly preserve qualifiers like “could be” and “not sure why.” It should also capture conflicting feedback (e.g., search slower vs faster) rather than forcing a single narrative. This produces an accurate, decision-supporting synthesis without false certainty.
When a Product Owner uses AI to summarize customer feedback, the prompt should constrain the output to what the evidence actually supports. In this exhibit, several comments include caveats (VPN, one-time failure, “not sure”) and there is conflicting feedback (filters improved speed vs search feels slower).
A good prompt will:
This helps you communicate what is known vs. suspected and avoids turning ambiguous feedback into overstated “facts,” which would distort Product Backlog decisions.
Topic: AI Theory and Primer
A Product Owner asks an AI assistant to recommend the next 10 Product Backlog items to best support the Product Goal. They provide recent user feedback themes, current metrics, and known constraints, then iterate the prompt eight times. The recommendations keep changing and the team debates them without reaching a decision; stakeholders say the list feels arbitrary and don’t trust it. When asked how to judge a “good” recommendation, the Product Owner says, “I’ll know it when I see it.”
What is the most likely underlying cause?
Best answer: D
Explanation: The team is stuck because there is no shared definition of what “good” looks like for the AI-generated prioritization. Without explicit success criteria (e.g., the value and risk factors to optimize, constraints, and tie-breakers), further prompt iteration becomes guesswork and yields inconsistent outputs. That’s the signal to stop iterating with AI and switch to human decision-making to define the criteria first.
Diminishing returns with AI often shows up as repeated prompt iterations that change the output but do not increase confidence or decision quality. In this scenario, the key clue is the lack of an evaluable standard: the Product Owner cannot state what would make a recommendation acceptable.
A Product Owner should use human judgment to establish clear success criteria and constraints (for example, desired outcomes, weighting of value/risk, non-negotiables, and what evidence is required), then use AI to generate options or analysis that can be consistently reviewed against those criteria. The closest alternative is “missing context,” but the stem already says relevant context and constraints were provided.