prd-review

name	prd-review
description	Use this skill whenever a user wants to review a PRD, audit a product spec, run a PRD quality check, or get structured feedback on a product requirements document. Triggers include: "review this PRD", "check my PRD", "audit this spec", "is my PRD good", "PRD review checklist", "give feedback on my product doc", or when a user pastes a PRD and asks for critique.

PRD Review Checklist

Covers 2-pagers and 6-pagers. Mark each item ✅ (Yes) · ⚠️ (Partial) · ❌ (No)

Field
Document Title
Author / Reviewer
Review Date

---

0. Opening & Framing

Amazon PRFAQ principle: if the opening doesn't land, leadership stops reading. Check this first.

#	Check	Status	Notes
0.1	Headline / title is a customer benefit statement — not a feature description or internal project name
0.2	First paragraph names the customer, the problem, and the transformation delivered — an unfamiliar exec understands why this matters in under 30 seconds
0.3	Document is free of buzzwords and hype: "revolutionary", "game-changing", "AI-powered", "next-gen", "paradigm shift"
0.4	Narrative is written from the customer's perspective throughout — not the engineer's or the company's

Notes:

---

1. Problem & Customer

#	Check	Status	Notes
1.1	Problem is clearly articulated (job-to-be-done) and validated with real customer data or research — not internal assumptions
1.2	Problem statement is a pure pain statement — does NOT hint at or imply a solution
1.3	Negative impact is quantified: time lost, cost, frequency, or scale of affected users
1.4	Customer persona defined with specific roles, industries, and needs; buyer vs. user distinguished if different
1.5	Urgency is clear — explains why now and the cost of inaction
1.6	Existing alternatives called out as insufficient, with a specific reason why they fall short
1.7	MOAT is defined and defensible — data flywheel, workflow lock-in, proprietary training, or switching cost
1.8	Solution is differentiated from ChatGPT, copilots, and commodity AI wrappers
1.9	Validated demand signal present — customer quotes, paid pilots, waitlists, or demonstrated willingness to pay

Notes:

---

2. Solution & AI Design

#	Check	Status	Notes
2.1	Core value proposition is crisp (1–2 sentences) and directly addresses the stated problem
2.2	Solution addresses every problem raised in the Problem section — no orphaned pain points left unresolved
2.3	Solution is described in benefit language (what the customer experiences) — not feature lists or system architecture
2.4	Visual user flow included: input → processing → output
2.5	Core functional requirements listed in user story format
2.6	Use of Agentic AI / ML / LLMs justified over rule-based or simpler approaches; unstructured data types identified
2.7	AI drawbacks addressed: hallucination, attribution accuracy, agent failure modes, and explainability

Notes:

---

3. Metrics & Success

#	Check	Status	Notes
3.1	North Star metric defined — the single number that proves the product is working
3.2	Primary metric is customer-recognized — something the customer would care about, not an internal proxy
3.3	Every primary metric has a baseline (current state value) AND a SMART target (specific, time-bound)
3.4	Both leading indicators (early signals you can watch) and lagging indicators (ultimate business outcomes) are present
3.5	Secondary metrics included: adoption, time saved, cost reduction, or task completion rate
3.6	Anti-metrics / guardrails stated — what should NOT get worse (e.g. false attribution rate, user trust, latency)
3.7	Launch criteria clearly defined — what specific thresholds must be met to ship?
3.8	Metric ownership and tracking method defined — who owns each metric and how will it be measured?

Notes:

---

4. Prioritization & Roadmap

#	Check	Status	Notes
4.1	Agentic workflow broken into components (e.g. ingestion, extraction, summarization) with per-component risk assessment
4.2	ML feasibility, data availability, and accuracy expectations specified
4.3	Features prioritized by risk, cost, and value; MVP scope narrowed with rationale
4.4	Roadmap phases defined: MVP → MVP1 → Launch → Iteration, with feature sets and durations
4.5	Roadmap items linked back to priorities and risks

Notes:

---

5. AI Evaluation & Prompt Strategy

#	Check	Status	Notes
5.1	AI performance evaluation approach described (datasets, ground truth sources, benchmarking cadence)
5.2	Accuracy targets defined relative to a human or rule-based baseline — not just in absolute terms
5.3	Prompting strategy described — few-shot, CoT, or conditional prompting where applicable
5.4	Plan for prompt iteration and improvement based on user feedback

Notes:

---

6. Responsible AI

#	Check	Status	Notes
6.1	Risks assessed across Accountability, Transparency, Fairness, and Reliability
6.2	Human-in-the-loop and fallback paths planned
6.3	Sensitive use cases identified with mitigation strategies

Notes:

---

7. Pricing & Business Case

#	Check	Status	Notes
7.1	Development costs broken down (infra + manpower) and ongoing operational costs listed
7.2	Revenue scenarios defined and pricing model selected with rationale
7.3	Accuracy vs. cost tradeoffs explained
7.4	Directional pricing recommendation included

Notes:

---

Overall Verdict

Rating	⬜ Strong   ⬜ Good — minor gaps   ⬜ Needs Work   ⬜ Not Ready
Top Strengths	1. · 2. · 3.
Top Gaps	1. · 2. · 3.
Next Step

---

PRD Review Checklist v3.0 — Reviews Project · Updated 2026-04-01 based on Amazon PRFAQ best practices

---

✦ Proposed Additions — v2 (2026-04-24)

These items were added based on feedback patterns seen 5+ times across reviewed documents.

| ✦ NEW | Metrics & Success → 3.2 — Does each primary metric include a baseline (current state) and a SMART target? | | | (Added v2 — seen in 5 docs. Example: "When a document scopes the product too narrowly (e.g. "back-office teams in CRE ...") | | ✦ NEW | Problem & Customer → 1.1 — Is the problem backed by at least one concrete data point or direct customer evidence? | | | (Added v2 — seen in 5 docs. Example: "When a document scopes the product too narrowly (e.g. "back-office teams in CRE ...") |

---

Checklist v2 — pending approval from Mahesh