Product Owner & AI Exercises

Fill in the blanks to test your knowledge.

1

When AI coding tools increase development speed, the bottleneck in product delivery shifts to PO ___ quality — the clarity and accuracy of what the team is asked to build

Before AI: implementation was the constraint.

After AI: PO quality is the constraint — can developers implement without ambiguity?

Key dimensions: clarity, prioritization speed, and sprint review accuracy.

Risk: a PO who is the bottleneck in an AI-accelerated team has visible professional risk.

2

The AI use case where multiple user interview transcripts are analyzed simultaneously to find pain points, workarounds, and underlying needs is called ___ synthesis

synthesis uses AI to analyze multiple user interview transcripts simultaneously.

It finds: pain points by frequency, workarounds (signs of unmet needs), emotional signals (frustration = high-value problems), and feature requests with their underlying needs.

PO validation step: compare AI output to what you actually heard in the room.

3

The prioritization framework scored as Reach × Impact × Confidence ÷ Effort is called the ___ framework

The framework scores features as: Reach × Impact × Confidence ÷ Effort.

AI improvement: score multiple backlog items simultaneously.

Best feature: AI identifies where the score contradicts your intuition — forcing you to examine the assumption.

PO role: validate that Reach and Impact inputs reflect actual data.

4

A well-formed product problem statement must be solution-___ — it describes the problem without prescribing the answer

A product problem statement must be solution-.

It describes who has what problem in what context — not how to fix it.

AI risk: AI defaults toward solutions implied by the business context provided.

Test: could two different engineers read this and design two different valid solutions?

5

AI feature behavior is ___ rather than deterministic — it succeeds most of the time with a certain accuracy, which changes how requirements must be written

Traditional behavior: deterministic — when X happens, system does Y.

AI behavior: — succeeds most of the time with a certain accuracy level.

Requirements must now specify: accuracy thresholds, failure mode handling, human oversight design, and feedback loops for model improvement.

6

The explicit specification of when an AI's output must be reviewed by a human before taking effect — who reviews, what they see, and how decisions are recorded — is called human-in-the-___ design

Human-in-the- design specifies:

- When the AI acts automatically vs. when a human reviews first

- Who the reviewer is and what authority they have

- What information the reviewer sees (AI output + reasons + data used)

- How reviewer decisions are recorded for audit and model improvement

This must be a first-class requirement, not an afterthought.

7

The three types of acceptance criteria required for AI features are process-based, accuracy-based, and ___-based

AI features require three types of acceptance criteria:

1. Process-based: AI runs on 100% of inputs, all decisions logged

2. Accuracy-based: precision/recall targets on a validation dataset

3. -based: AI explanations are comprehensible to reviewers, workflow time is within limits

Traditional pass/fail criteria cover type 1 only. UAT must test all three with actual users.

8

A product owner's foundational understanding of what makes software features easy vs. hard to build, and how data models constrain features, is called ___ literacy

literacy — understanding the end-to-end software development process — enables a PO to:

- Make better prioritization decisions (knowing the cost of each feature)

- Have credible conversations with technical teams

- Assess trade-offs accurately without relying on developers

In the AI era: more valuable than ever, as AI tools compress implementation and surface requirements quality as the bottleneck.