Module 5

Requirements & PRDs

📖 2 Lessons⏱ 60-90 min 🧪 Lab: PRD Generation and QA✅ Quiz: 6 questions

Learning Objectives

Write production-ready PRDs with all required sections
Use AI to draft PRDs from discovery insights
Run QA loops on AI-generated requirements
Identify missing edge cases through structured review

Lesson 5.1

PRD Anatomy

Section	Purpose	Common Failure
Problem statement	What is wrong and why it matters	Too vague, no evidence
User impact	Who is affected, how badly	No quantification
Business impact	Revenue/churn/cost implications	Missing or aspirational
Requirements	What needs to be built	Solutions masquerading as requirements
Success metrics	How we know it worked	Vanity metrics, no baseline
Edge cases	What could go wrong	Missing entirely
Rollout plan	How we ship it safely	Big bang launch with no safety net

💡

Requirements vs Solutions: "Users need to connect data sources without IT help" is a requirement. "We will add a 3-step wizard" is a solution. Keep them separate.

⚠️

Every requirement must be independently testable. "Improve onboarding" is not testable. "Users can complete step 3 without contacting IT support" is.

Lesson 5.2

AI PRD Generation: The QA Loop

QA Check	What to Look For
Hallucination	Requirements not supported by discovery data
Edge case coverage	Common failure scenarios missed
Metric quality	Success metrics have measurable baselines
Scope creep	AI added requirements nobody asked for
Technical feasibility	AI assumed capabilities that do not exist

Lab 5

PRD Generation and QA

Scenario: FlowScale has decided to prioritize the Guided Onboarding Wizard to address the 62% drop-off at step 3.

Step 1

Create prompts/prd-generator.md with role, context reference, task definition, and quality rules including: every requirement must cite its discovery source; mark technical unknowns as [NEEDS ENGINEERING INPUT]; minimum 8 edge cases required.

Step 2 and 3

Run the PRD generator with all discovery data loaded. QA the output against: hallucination check, 8 edge case categories, metric baselines, and scope boundaries.

Edge cases to verify are covered:
- Network failure during setup
- User abandons mid-flow and returns
- Data source connection fails
- User skips optional steps
- Mobile browser experience
- Enterprise SSO vs email signup
- Re-onboarding (existing workspace)
- Browser tab closed mid-setup

Step 4

Write the final corrected PRD at deliverables/W05-prd-onboarding-wizard.md. Mark each correction inline: what was wrong and what you changed.

Deliverables

PRD generator prompt
Raw AI output
Completed QA checklist
Final corrected PRD with fix markers

How to Verify Completion

PRD contains all 7 sections
Every success metric has a baseline from FlowScale analytics
At least 8 edge cases documented
At least 2 QA issues found and fixed with inline comments
No requirement prescribes a specific implementation

Done when: An engineer could build the feature from your PRD without a clarification meeting.

Quiz

Module 5 — Knowledge Check

0 / 6 answered

Question 1 of 6

Which of these is a requirement, not a solution?

A Add a 3-step wizard with API key auto-detection

B Build a contextual help sidebar

C Non-technical users can connect a data source without contacting IT support

D Redesign step 3 of the onboarding flow

Correct. Option C describes an outcome without prescribing implementation. A, B, and D all describe how to build something — they are solutions, not requirements.

Question 2 of 6

A success metric reads "Improve onboarding completion." What is wrong?

A It is too technical for stakeholders

B It has no baseline, no target, and "improve" is unmeasurable without numbers

C Revenue is a more important metric

D It should be qualitative instead

Correct. A good version: "Increase step 3 completion from 41% to 60% within 60 days of launch." Baseline + target + timeline + tracked metric.

Question 3 of 6

The AI added "social login with Google" to the PRD. Discovery never mentioned this. This is an example of:

A A technical feasibility issue

B An edge case to document separately

C Scope creep — the AI added a feature not justified by discovery data

D An opportunity to add to the backlog

Correct. Every requirement must trace to evidence. Scope creep requirements that seem reasonable but have no discovery backing distract from validated problems and inflate engineering scope.

Question 4 of 6

Why is "user abandons mid-onboarding and returns" a critical edge case?

A It tests the database persistence layer

B It only affects enterprise SSO users

C Abandoned sessions are rare

D With 62% drop-off at step 3, returning users who face the full flow again are far more likely to abandon permanently

Correct. The edge case directly connects to the core problem. Without progress persistence, the wizard makes re-entry worse, potentially increasing permanent abandonment even though it was designed to help.

Question 5 of 6

A rollout plan with rollback criteria primarily protects against:

A Engineers shipping too slowly

B Continuing a broken launch because nobody pre-defined when to stop

C Stakeholders requesting more features post-launch

D The feature going over budget

Correct. Rollback criteria define the threshold in advance: "If step 3 completion drops below 41% within 7 days, roll back." Without them, teams continue with a failing launch because there is no pre-agreed signal to stop.

Question 6 of 6

Why should QA-corrected PRDs mark every fix with an inline comment?

A To show how much work the PM put in

B Because Git requires change annotations

C To create an auditable record of AI failures that improves prompt quality over time

D To allow the AI model to learn from its mistakes

Correct. QA fix comments reveal patterns across modules. When the AI consistently misses certain edge cases or inflates scope in the same way, that is the signal to add a specific constraint to the prompt template.

Module 5 Score

←

Module 4 — Strategy

→

Module 6 — Prioritization