What this lesson is

Lesson 04 was your first AI PRD — a 24-minute draft from a one-line ask. This lesson is the production-grade version: the canonical template every AI PRD needs once a feature stops being a flag-gated v0 and starts being a thing real customers depend on.

Eight sections. Most PMs skip three of them on their first try, and each skipped section maps cleanly to a class of post-launch incident. The point of having a canonical template is not bureaucracy — it's pattern matching. When the PRD is laid out the same way every time, an engineer, a legal counsel, a data lead and a CFO can each read the section that concerns them in five minutes and not miss anything.

This is the lesson you screenshot, paste into your Notion sidebar, and refer to every time someone asks you to ship an AI feature. It is the reference document.

Who this is for:
you've shipped a v0 AI feature behind a flag, a small set of users have used it, and now leadership wants the v1 launch. Lesson 04 got you to draft. This one gets you to review-ready production PRD.

The 8 sections of a canonical AI PRD

Lesson 04 had seven sections. Production adds an eighth — Rollout plan — because v1 isn't "deploy on Tuesday." It's shadow → canary → 100%, with a launch gate between each step. Below are the eight, in the order they appear in the doc.

🎯
1. Problem & users
User pain in user words, top 3 intents, business KPI plus a trust metric.
🧭
2. Scope (in/out)
Intents in scope, intents refused, named handoffs for everything else.
📝
3. Prompts & behavior
Role · tone · format · refusal strings. The behavioural contract.
🗂️
4. Data & retrieval
Allowed fields, freshness, PII boundaries, retrieval config, audit logs.
🚦
5. Eval & launch gate
Golden set, rubric, pass-rate threshold, zero-tolerance subsets.
🚀
6. Rollout plan
Shadow → canary → 100%, with a gate and a rollback rule between each.
🛡️
7. Risks & fallbacks
Each named risk gets a fallback, an owner, and a detection signal.
📊
8. Metrics & success
North-star, guardrails, cost per task — what we watch on the dashboard.
Why eight, not seven: evals tell you whether the model can do the task. Rollout tells you whether it does the task on real traffic without breaking. Skip rollout and your golden-set pass rate is a press release, not a launch criterion.

Section × skip-if-missing × owner × failure mode

The fastest way to internalise the 8 sections is to know what breaks when you cut each one. This table is what Chidi pins to his Notion at NairaSwift before writing the v1 PRD. Notice the right-most column — every skipped section has a named, recurring failure that production AI teams have learned the hard way.

Section Skip if missing? Primary owner Common failure
1. Problem & users Never PM Beautiful feature solving the wrong problem; v2 is a complete rewrite
2. Scope (in/out) Never PM Users type out-of-scope intents; model improvises; legal incident
3. Prompts & behavior Never PM (eng implements) Tone drift across releases; user trust score quietly falls
4. Data & retrieval Never PM + Data lead Model leaks PII or cites stale data; DPDP/GDPR exposure
5. Eval & launch gate Never PM + Eng "We'll know it's working when…" — every exec meeting becomes a vibe check
6. Rollout plan Only if v0 already at 100% PM + Eng Day-one regression on edge case nobody saw in QA; no rollback path
7. Risks & fallbacks Never PM + Security Prompt injection, cost spike, vendor model upgrade — discovered on Twitter
8. Metrics & success Never PM + Analytics Launch happens; nobody knows if it worked; feature quietly deprecated in 9 months
The single skippable row:
Rollout, but only if you already shipped a v0 behind a flag and v1 is just removing the flag at the same percentage. Otherwise, every section is a precondition for shipping. Reviewers should reject any AI PRD missing one of these eight headings.

Scenario — Chidi at NairaSwift Lagos

💸
Real-world scenario
v0 dispute-explainer worked behind a flag. Now write the PRD that earns the 100% rollout.

Chidi is a senior PM at NairaSwift, a Lagos-based remittance startup. Six weeks ago he shipped a v0 dispute-assistant behind a feature flag — when a user files a "money sent but not received" dispute, an AI explains the likely cause (correspondent-bank delay, mismatched name, FX cutoff) and the next step. v0 is at 5% of dispute traffic, has a 67% thumbs-up rate, and saved roughly ₦4.2M in agent time last month. Leadership wants v1 at 100% before the next compliance audit. Chidi has 10 weeks.

v0 was a Notion page and a Slack thread. v1 needs an actual PRD that survives review by engineering, the head of compliance, the CFO, and a sceptical VP of CX who already lost trust in chatbots once. Chidi opens the canonical 8-section template. The rest of this lesson is what each section looks like by Friday.

Why this matters: Chidi's situation is the most common AI-PM moment of 2026 — v0 worked, v1 needs to survive. The PRD you write here is the one that 50 stakeholders will read; the one Lesson 04 taught is the one only you read.

How Chidi fills each of the 8 sections

Below is the actual sequence — eight step-blocks, one per PRD section, with what Chidi puts inside. Read it as a worked example, not a template to copy verbatim. The sections are universal; the contents are NairaSwift-specific.

SECTION 1Problem & users
Owner: Chidi · Length: 1 page · Source data: 6 weeks of v0 telemetry + 50 dispute tickets

Chidi opens with the user pain in user words, not PM words: "NairaSwift senders file ~14k disputes/month. 61% are not actually disputes — they're users panicking because the recipient hasn't received naira yet, and we haven't told them why. Average time-to-resolution is 38 hours. CSAT after a dispute resolution is 18 points lower than the baseline."

Then the goal, with three numbers: deflect 45% of dispute filings via the AI explainer in Q3; thumbs-up rate ≥ 75%; zero false-reassurances on actually-failed transfers (the safety constraint). He names the top three intents from v0 telemetry — correspondent-bank delay, name-mismatch hold, weekend FX cutoff — and notes they cover 78% of v0 traffic. Everything else gets a hard handoff. Personas are spelled out as: Sender (anxious, mobile, English or Pidgin), CX agent (using the AI as a co-pilot, not a replacement).

SECTION 2Scope (in/out)
Owner: Chidi · Length: half page · Reviewed by: Compliance, CX

Two columns. In scope (v1): explain the three named intents above; offer a "wait X hours" expectation; offer "talk to an agent" handoff; offer "track this transfer" link. Out of scope: initiating a refund, cancelling a transfer, KYC-related questions, anything mentioning fraud, disputes against a specific merchant. Each out-of-scope intent gets a verbatim handoff string written in this section, not in the prompt section — this is so Compliance can review the exact words a user will see. The doc says: "If the AI handles an out-of-scope intent in production, that's a P0 incident and the rollout is reversed."

SECTION 3Prompts & behavior
Owner: Chidi · Pointer to: prompts/dispute_explainer_v1.txt in the eng repo

The full prompt anatomy block (see the next section) goes here. Chidi keeps the prose system prompt in the PRD so non-engineers can read it, and the canonical version lives in git as prompts/dispute_explainer_v1.txt with a version pinned next to the model name. This section also names the explicit behaviours: never quote an exact arrival time (only ranges), never mention any correspondent bank by name unless it's in the retrieved data, never say the word "guaranteed", always include the transfer reference. The refusal strings from Section 2 are referenced — not duplicated.

SECTION 4Data & retrieval
Owner: Chidi + Data lead · Reviewed by: Compliance, Security

A field-by-field allow-list. The model can see: transfer ID, amount in NGN and destination currency, sender country, recipient country, status code, status timestamp, correspondent bank code (not name), expected arrival window. The model cannot see: full names, phone numbers, BVN/KYC data, linked cards, any other transfers from the same sender. Freshness: status updated within 5 minutes. Retrieval is keyed off transfer ID only — no semantic search across user history. Audit log: every retrieved record is hashed and stored for 90 days with the prompt and response, for compliance review. PII boundary: if the retrieval somehow returns a name field, it is stripped before reaching the prompt.

SECTION 5Eval & launch gate
Owner: Chidi + Eng lead · Frozen by: end of week 2

Golden set: 200 real disputes from v0, hand-labelled with the ideal one-paragraph response. Two graded axes — correctness (did the AI cite the right status code and arrival window?) and helpfulness (did the user understand and stop escalating?) — each scored 1–5 by an LLM-as-judge with weekly human spot-checks. Launch gate in one sentence: "Ship to next rollout step when ≥ 82% of golden set scores ≥ 4 on both axes, AND a 30-item 'failed-transfer' subset has zero false-reassurances." The fail subset is the zero-tolerance gate — even one "your money is on the way" answer on a transfer that actually bounced means no rollout.

SECTION 6Rollout plan
Owner: Chidi + Eng lead · Detailed in lesson 14

Four steps with explicit gates. Shadow (week 4): AI runs on 100% of disputes, output never shown to users, scored against agent answers. Gate: golden set ≥ 82%. Canary 5% (week 5): live to 5% of senders. Gate: live thumbs-up ≥ 70%, zero P0 fallback firings, cost ≤ ₦12 per dispute. Canary 25% (week 7): Gate: thumbs-up ≥ 73%, agent escalation rate within 10% of baseline. 100% (week 9): Gate: thumbs-up ≥ 75%, full compliance sign-off. Rollback rule named for each: any gate miss for 24 hours triggers automatic step-down to the previous percentage.

SECTION 7Risks & fallbacks
Owner: Chidi + Security · Format: risk · detection · fallback · owner

A four-column table inside the PRD. Each row: a named risk, how we detect it, the fallback, the owner. Examples: Hallucinated arrival time → format rule restricts to retrieved windows only → log monitor alerts on any answer with a date, owner: Eng. Prompt injection from transfer reference field → strip control chars and reject inputs > 64 chars at retrieval, owner: Security. Vendor silently upgrades model → pinned version + weekly automated re-eval, ≥3-point regression auto-rollbacks, owner: Eng. Cost spike from longer prompts → per-conversation token cap of 1500, hard fallback to templated reply, owner: PM. Compliance audit finds PII leak → audit log + retrieval allow-list + quarterly review, owner: Compliance. Twelve rows total. Each one named, no row of "various edge cases."

SECTION 8Metrics & success
Owner: Chidi + Analytics · Pointer to: dispute-ai dashboard in Looker

Three tiers. North-star: dispute-deflection rate (target 45%). Quality guardrails: thumbs-up rate ≥ 75%, refusal rate stable at 8–12% (too low means the model is over-helping; too high means it's punting), agent-escalation latency unchanged. Cost & ops: cost per dispute ≤ ₦12, p95 latency ≤ 4s, fallback firing rate ≤ 1%. Each metric names the dashboard panel, the alert threshold, and what the on-call PM does if it breaches. Crucially, the section ends with "What success means in 90 days": a one-paragraph statement of the outcome, signed by Chidi and the VP of CX. That paragraph is the PRD's closing argument.

What Chidi's finished v1 PRD looks like

Below is the at-a-glance preview — page one of the Notion doc. If a reviewer reads only this, they should already understand what's shipping, what isn't, the launch gate, and the rollout plan. This is the screenshot Chidi pastes into the v1 launch Slack channel.

Notion · NairaSwift · Customer Experience
PRD v1.0 — Dispute Assistant (AI)

Owner: Chidi · Target 100%: Week 9 · Launch gate: ≥ 82% golden set · ≥ 75% thumbs-up · 0 false-reassurance · ≤ ₦12/dispute

1
Problem & users
14k disputes/month · 61% are panic, not failure · 3 intents cover 78% of volume
2
Scope (in/out)
In: 3 named intents + handoff. Out: refunds · KYC · fraud · merchant disputes (verbatim refusals)
3
Prompts & behavior
Role · tone · format · refusal — pinned in prompts/dispute_explainer_v1.txt
4
Data & retrieval
8-field allow-list · 5-min freshness · transfer ID keyed · 90-day audit log · no PII
5
Eval & launch gate
200-item golden set · LLM-judge + human spot-check · gate = 82% + 0 false-reassurance
6
Rollout plan
Shadow → 5% → 25% → 100% · gates per step · 24-hour auto-rollback rule
7
Risks & fallbacks
12 rows · risk · detection · fallback · owner · pinned model + cost cap + audit log
8
Metrics & success
North-star: 45% deflection · guardrails: thumbs ≥ 75%, refusal 8–12% · cost ≤ ₦12/dispute
The reviewer test:
if a stakeholder reads only this preview, they should be able to predict what each full section says without surprise. If anyone is surprised by the contents of section 6 (rollout) or 7 (risks), the preview headline isn't doing its job. Re-write the hint until it does.

The PRD's lifecycle — draft to in-flight

The PRD is not written in one sitting. It moves through six states, and the canonical template is what makes those state transitions reviewable. Skip a state and the document silently degrades into a wishlist again.

Draft PM, 8 sections Review w/ eng prompts · data · risks Eval design freeze rubric, gate Stakeholder rev CX · legal · CFO Approved v1.0 frozen In flight w0–w1 w1 w2 w2–w3 w3 w4–w9
The most-skipped state: Eval design at week 2. PMs jump from "review with eng" straight to "stakeholder review" because they're worried about losing momentum. Skipping eval design means stakeholders sign off a PRD whose launch gate is still negotiable — and a negotiable gate is no gate at all.

Inside section 3 — what "Prompts & behavior" actually contains

This is the section reviewers most often glaze past, and the one that determines whether the model behaves the way the PRD claims. Below is exactly what Chidi pastes into section 3 — a colour-coded prompt anatomy block that compliance, eng and CX can each read in 60 seconds.

ROLE You are NairaSwift's dispute assistant. You explain transfer status using only retrieved data. You do not initiate refunds, cancel transfers, or discuss KYC.

TONE Calm. Direct. Never apologise before the facts. Never use the word "guaranteed". Match the user's language register (English or Pidgin) but stay neutral.

FORMAT One paragraph, ≤ 60 words. Always include: the transfer reference, the current status code, the expected arrival window (range only, never an exact time). End with one of two follow-ups: "Track this transfer" or "Talk to an agent".

REFUSAL If the user asks for a refund, mentions fraud, asks about KYC, or names a merchant, respond exactly: "That one needs an agent — I'm connecting you now." Do not attempt to explain or reassure. Do not name the correspondent bank.
Why this lives in the PRD, not just in git:
compliance and the VP of CX cannot read a .txt file in a code repo, but they need to sign off on the exact tone and refusal language. Putting the colour-coded version in the PRD makes the behavioural contract reviewable by non-engineers — and any change to the live prompt requires a PRD revision, not just a pull request.

The AI connection

How real AI teams use the canonical template

At Anthropic, Stripe, SouthBank India and PhonePe, AI PRDs follow a remarkably similar 7–9 section shape. The PRD is the input to four downstream artifacts: a versioned prompt file in git, a retrieval config, an eval CI job that blocks deploys, and a rollout config in the feature-flag service. PMs who write tight section 3 (prompts), 4 (data), 5 (gate) and 6 (rollout) save eng teams entire sprints of clarification meetings. The canonical template is what turns "AI feature" from a Slack thread into a deployable artifact. prd-template launch-gate rollout-config prompt-repo eval-ci audit-log

Mini challenge — pressure-test your last AI PRD

Take a real PRD you've written and grade it against the canonical template

1. Map your PRD onto the 8 sections

For each of the eight sections, paste the heading from your PRD that matches. If a section has no match, write "MISSING" in red. Most PMs find 2–3 missing on their first audit. Don't fix them yet — just count.

2. Predict the failure for each missing section

Using the table from the deep-dive, name the specific failure mode each missing section invites. Be concrete: "skipped Rollout → no canary, day-one regression on edge case" not "could be risky."

3. Re-write the launch gate as one sentence

Open section 5 and reduce your launch gate to a single sentence with three numbers: pass rate, golden-set size, zero-tolerance subset. If it doesn't fit in one sentence, the gate is still negotiable. Tighten it until a junior engineer could implement it without asking you.

Interview questions

Easy What are the 8 sections of a canonical AI PRD?
+
In order: (1) Problem & users, (2) Scope (in/out), (3) Prompts & behavior, (4) Data & retrieval, (5) Eval & launch gate, (6) Rollout plan, (7) Risks & fallbacks, (8) Metrics & success. Sections 3, 4, 5, 7 don't exist in classical PRDs. Section 6 (rollout) is the upgrade from the seven-section v0 template — it covers shadow → canary → 100% with named gates.
Easy What's the difference between scope and metrics?
+
Scope is what the feature tries to do; metrics are how you know it succeeded. Scope lists intents in and out — it constrains the model and gives compliance a refusal list. Metrics are dashboard panels with thresholds and on-call actions: north-star (deflection rate), guardrails (thumbs-up, refusal rate, latency) and cost. Confusing them is a classic miss: PMs put "thumbs-up rate ≥ 75%" in scope, which is a metric, or list "explain transfer delays" in metrics, which is scope. Scope is a contract; metrics are the scoreboard.
Medium How do you handle stakeholders who want to skip the eval section?
+
First, name what skipping it costs. "Without an eval section, every exec meeting becomes a vibe check on whether to ship." Then offer a smaller eval, not no eval — a 50-item golden set built in a week, with a single combined rubric, is enough to freeze a launch gate. Skipping the section entirely means there is no number that tells you when to roll back, so any production regression becomes a political negotiation. If the stakeholder still pushes, write the trade-off explicitly into the Risks section: "Risk: shipping without an eval. Detection: none. Fallback: full feature kill-switch. Owner: VP of CX." Most stakeholders fold the moment they have to sign their name to that row. The eval section is non-negotiable; what's negotiable is its size and timeline.
Medium Why does Rollout deserve its own section instead of being a sub-section of Eval?
+
Because eval and rollout answer different questions. Eval asks "can the model do this task on a frozen golden set?" Rollout asks "does it still work when real users try to break it on Tuesday afternoon?" The two have different owners, different gates and different rollback rules. A passing eval is necessary but not sufficient — golden sets miss long-tail user inputs, latency under load, prompt injection attempts, and vendor-side model drift. Promoting rollout to its own section makes it a first-class artifact: shadow → canary → 100% with explicit step gates and a 24-hour auto-rollback rule. Burying it inside eval is how teams ship at 100% on day one and discover edge cases on Twitter.
Hard How do you write the "risks" section without sounding scared?
+
The instinct is either to under-write it (three vague bullets, "we will monitor") or to over-write it (twenty paragraphs of caveats that read like a CYA letter). Both kill the PRD. The answer is a four-column table: risk · detection · fallback · owner. Each row is a sentence, not a paragraph. The tone you're aiming for is the same as a flight pre-flight checklist — neutral, specific, named owners — not "things that worry me." Three rules: (1) every risk has a named fallback, never "we'll figure it out" — that signals fear; (2) every risk has a detection signal, never "we'll know if it happens" — that signals you don't actually know; (3) every risk has a single owner, not "the team" — that signals nobody. Twelve rows of (risk · detection · fallback · owner) reads as competence; three rows of "potential issues" reads as nervousness. The structural move that turns the section from scared to confident is replacing every adjective with a number or a name. "Cost may spike" becomes "Cost > ₦18/dispute for 1 hour → fallback to templated reply, owner: PM." Same risk, completely different tone. The reviewers are not looking for fewer risks; they are looking for evidence you've thought about each one. The table format gives them that evidence in the most scannable form possible. Bonus: end the section with one line — "any production firing of a fallback is a P1 incident with a 7-day post-mortem." That single sentence shifts the section from "things that might go wrong" to "things we have already decided how to handle," which is the tonal difference between scared and ready.

Test your knowledge

Question 1 of 5
Which of these is the only PRD section a v1 launch may legitimately skip?
Question 2 of 5
Why does the canonical AI PRD have 8 sections instead of the 7 from a v0 mini-PRD?
Question 3 of 5
Why does the "Prompts & behavior" section live in the PRD and in git?
Question 4 of 5
A reviewer says: "your Risks section sounds nervous." What's the structural fix?
Question 5 of 5
Chidi's launch gate reads: "Ship to next step when ≥ 82% of golden set scores ≥ 4 on both axes, AND the failed-transfer subset has zero false-reassurances." Why two thresholds, not one?
0
/5
Your AI PRDs will pass review.

Common mistakes writing a canonical AI PRD

Renaming sections to sound smarter
"Behavioural Architecture" instead of "Prompts & behavior", "Quality Posture" instead of "Eval & launch gate". The point of a canonical template is shared vocabulary. Use the eight names verbatim — reviewers should find each section in five seconds.
Single-number launch gates
"Pass rate ≥ 80%." Averages hide rare-but-catastrophic errors. Always pair the average with a zero-tolerance subset — false-reassurances, fraud answers, PII leaks. Two thresholds, AND'd together.
Risks section as prose
Three paragraphs of "potential issues to monitor" reads as fear. Convert to a table: risk · detection · fallback · owner. Every cell has a number or a name. The structure is what makes the section sound competent.
Metrics without on-call actions
"We'll watch thumbs-up rate." What threshold triggers what response? Each metric needs three things: a target, an alert threshold, and the named action when it breaches. Otherwise, the dashboard is decoration.

Key takeaways

  • Canonical AI PRDs have 8 sections: problem & users · scope · prompts & behavior · data & retrieval · eval & launch gate · rollout · risks & fallbacks · metrics & success. Use these names verbatim.
  • The 8th section — Rollout plan — is what separates v0 from v1. Eval says the model can do the task; rollout says it survives real traffic. Skip rollout and you ship at 100% on day one with no rollback path.
  • Section 3 (Prompts & behavior) lives both in the PRD and in git. The PRD version is for compliance and CX; the git version is for engineering. Any change requires both.
  • Launch gates are two thresholds: an average pass rate AND a zero-tolerance subset for the highest-cost error class. Single-number gates hide rare disasters.
  • Risks & fallbacks is a table with four columns — risk · detection · fallback · owner. A number or a name in every cell. Prose risks read as fear; tables read as readiness.
  • Every metric in section 8 needs a target, an alert threshold, and a named on-call action. Without the third, the dashboard is decoration and the metric is theatre.
Africa & India's Best AI Academy
🤖 AI-first 🎓 Placement Assistance

You now write AI PRDs that ship.
Next — design the rollout that proves them.

Join the Best AI Program: The AI-Era Agentic Engineer — 5 months, fully online, with placement assistance. Week 1 is free.

Explore The AI-Era Agentic Engineer →