Product Documentation (PRD, PR/FAQ) Questions
Writing the artifacts that align teams and drive decisions: PRDs, one-pagers, and the PR/FAQ (working-backwards) format. Covers document structure and clarity, the press release lead, the external and internal FAQ, hard-question answers, success metrics and acceptance criteria, engineering-ready requirements and constraints, and reviewing or pressure-testing a draft. Also covers using written narrative to sharpen thinking and secure buy-in across functions, and running PR/FAQs as a team practice: intake and review process, versioning, onboarding and co-creation workshops. Assesses written product communication and documentation strategy.
A stakeholder insists on adding every requested feature and edge case into the press release. Describe how you would facilitate a conversation (role-play style) to keep the press release focused on the core customer benefit while respecting the stakeholder's input. What compromises and alternatives would you propose to capture remaining requests?
Sample Answer
Direct answer
Treat the press release as a promise to one customer about one benefit, and treat the stakeholder's list as evidence of what they care about, not as content to paste in. In the conversation I first show I have heard the concern, then apply one test ("would a customer read this sentence and understand why to care?"), and finally give every request a real home somewhere else. In a PR/FAQ (Press Release / Frequently Asked Questions: the working-backwards document where you write the imagined launch announcement and the hard questions before building anything), those homes are the FAQ, a "not in version 1" list, and a tracked backlog with a decision date.
Role-play (short excerpt)
Stakeholder (head of sales): "Enterprise buyers need SSO, audit logs, CSV export and the CRM sync. Put all four in the release or they will think we are not serious."
Me: "You are right that enterprise buyers judge seriousness, and I want them to see it. Help me understand: which of these has come up most in deals we lost?"
Stakeholder: "SSO and audit logs, mostly."
Me: "Then both go in the customer FAQ, where an enterprise buyer looks for exactly that, with a direct answer and a date. The release itself has one job: make the reader say 'that is for me' in the first paragraph. If it lists four capabilities, the one sentence about the benefit, less time spent building the weekly report, gets buried. Can I show you both versions side by side and you tell me which one you would forward to a customer?"
Stakeholder: "Fine, but I do not want these to disappear."
Me: "They will not. Each one gets an owner, a customer example and a review date, and I will bring the list back to you on that date."
Where each request goes (the compromises and alternatives)
| Kind of request | Home | Why |
|---|---|---|
| The one core customer benefit | Release headline and lead paragraph | This is what the customer must remember |
| Secondary capability many customers ask about | One line in the customer FAQ | Findable, but not competing with the benefit |
| Edge case or enterprise-only need | "What is not in version 1" FAQ plus a roadmap-later list | Honest about scope; shows it was heard |
| Internal detail (data model, migration) | Internal FAQ or the design doc | Customers do not need it |
| Idea nobody has evidence for yet | Parking list: owner, evidence needed, decision date | Prevents both arguing and forgetting |
Facilitation moves that keep it collaborative
- Restate the underlying worry ("you are afraid buyers will think we are not serious") before discussing the feature.
- Ask for evidence per item (lost deals, support tickets, usage) instead of debating importance in the abstract.
- Run the read test: give the draft to three people who were not in the room, take it back after 30 seconds and ask each to state the benefit. If they cannot, cut more.
- Let the stakeholder draft the FAQ answer for their request. Owning the wording usually ends the fight.
- If they still insist, name the decision-maker (the product owner accountable for the launch) and escalate the choice, not the argument.
Trade-offs and pitfalls
- The cost of cutting is a stakeholder who feels overruled; the cost of not cutting is a release nobody remembers. Cutting is worth it only when each cut item has a visible home.
- The FAQ can become a dumping ground. Cap it at the questions a customer or executive would really ask, and move the rest to the roadmap list.
- "Later" without an owner and a date is a polite no. Attach both, or say no plainly.
Describe the key sections you include when writing a Product Requirements Document (PRD) for a small feature. For each section explain its purpose and provide one example artifact or data point you'd attach (user-story, mock, metric baseline, acceptance criteria).
Sample Answer
Direct answer
A PRD (Product Requirements Document) is the written agreement on what we are building, for whom, and how we will know it worked. For a small feature I keep it to one to two pages: problem, users, goals and metrics, scope with non-goals, requirements with acceptance criteria, design, risks and open questions, and launch plan. Each section exists to answer one question a reviewer would otherwise ask in a meeting, and each gets one piece of evidence attached.
Sections, purpose, and the artifact to attach
| Section | Purpose | Artifact to attach |
|---|---|---|
| Problem and context | Why now; what hurts today | Support-ticket count or a two-line user quote |
| Target users | Who this is for, and who it is not | A one-line persona (a short description of a typical user) plus the segment size (how many such users exist) |
| Goals and success metrics | How we will know it worked | A metric baseline (today's number, its definition and date) and a target |
| Scope and non-goals | Stops scope creep | A short "not in this release" list |
| Requirements | What the system must do | User stories ("As a [user], I want [action], so that [outcome]") |
| Acceptance criteria | Testable definition of done | Given/When/Then checks ("Given [starting state], When [action], Then [expected result]") |
| Design | What the user sees | A mock (low-fidelity screen sketch) linked from the doc |
| Risks and open questions | What could go wrong, what is unknown | A table with owner and date to resolve |
| Launch plan | Rollout, support, rollback | Phase, audience %, and go/no-go owner |
Worked example: "save filters" on a job-search page
- Problem: 30% of repeat searchers re-enter the same three filters every visit (illustrative figure from an analytics query).
- Baseline: repeat searches per user per week = 2.1 over the last four full weeks, defined as searches by users with 2+ sessions.
- User story: "As a returning job seeker, I want my last filters remembered so that I can see new jobs in one tap."
- Acceptance criteria: Given a signed-in user who searched with filters X, When they return within 30 days, Then the same filters are preselected and a visible "Clear" control is shown.
- Target user: a persona of "returning job seeker who searches at least weekly"; segment size is 40,000 such users out of 120,000 monthly searchers (illustrative).
- Non-goal: named saved searches and email alerts (next release).
- Risks and open questions (owner, date to resolve): "Do remembered filters hide new job types a user would want?" Owner: PM, resolve by 10 June via a 5-user test. "Where are filters stored for signed-out users?" Owner: engineering, resolve by 12 June.
- Launch plan: 10% of returning users for one week, then 50%, then everyone; go/no-go owner is the PM, who checks that repeat searches per user per week did not fall below the 2.1 baseline before each step.
Pitfalls
- Listing solutions before the problem, so review debates buttons instead of value.
- A metric with no baseline, which cannot be judged later.
- Acceptance criteria like "works well", which cannot be tested.
- Copying a large-feature template onto a small feature until nobody reads it.
How would you structure the FAQ so that it helps engineers produce initial implementation estimates? Provide three example FAQ questions focused on scoping and constraints (e.g., expected data volume, performance targets, third-party dependencies) and one guideline about the level of technical detail that should be included versus deferred to design sessions.
Sample Answer
Direct answer
Structure the FAQ so it states constraints and their source, not solutions. Engineers estimate from four things: how much, how fast, what it depends on, and what is still unknown. So the FAQ gets a scoping section with a fixed set of question types (scope, data volume, performance, dependencies, open unknowns), and each answer gives a number or range, why it is believed, and a confidence level. One guideline rule keeps the level right: state the constraint and the outcome, defer the mechanism.
Three example FAQ entries (feature: saved searches with email alerts)
1. Data volume. "How much data will this create and keep?"
"We expect 2 million users, with a cap of 20 saved searches each, so the worst case is 40 million saved searches. We expect the average user to save 3, or about 6 million. Each search is stored for as long as the account exists. Source: survey and current usage of favourites. Confidence: medium; we will revisit after two weeks of beta."
Why it helps: the estimate for storage and indexing changes a lot between 6 million and 40 million rows, and the cap is the lever.
2. Performance target. "How quickly must an alert arrive, and how many run at once?"
"An alert email leaves within 15 minutes of a matching new listing, at the 95th percentile (P95: 95 of 100 alerts meet it). Average-case check load, assuming every saved search is re-checked every 15 minutes: 6,000,000 divided by 900 seconds, about 6,700 checks per second; at the 40 million worst case it is about 44,400 per second. A 15-minute re-check interval leaves almost no slack against a 15-minute P95 target (a listing that arrives just after a check waits nearly the whole interval), so meeting it may need a shorter interval (every 5 minutes triples the load to about 20,000 per second at 6 million) or matching on arrival; that choice belongs to the design session, but the estimate should show the load range. Confidence: low on the 15-minute target, and negotiable to 30 minutes if cost is high."
Why it helps: it sets the throughput engineers must size and names what is negotiable.
3. Third-party dependencies. "What outside services does this rely on, and what are their limits?"
"Email delivery through our existing provider, whose rate limit we have not confirmed for the volume above. Owner to confirm by end of next week. No new vendor is planned. If the limit is under 500 emails per second, batching or digests become part of scope."
Why it helps: an unconfirmed limit is a risk that widens the estimate, and it has a named owner and date.
The level-of-detail guideline
Include: numbers with a source, targets, caps, named dependencies, and open questions with owners. Defer to design sessions: schema, storage technology, queue choice, algorithm, service boundaries. Test for a line: if the sentence would still be true after the engineers pick a completely different implementation, it belongs in the FAQ; if it fixes how to build, it does not.
Structure of the scoping section
- In scope and out of scope for this release.
- Volume and growth (with caps).
- Performance targets, marking which are firm.
- Dependencies and their owners.
- Open unknowns and what would settle each.
Answers use ranges plus confidence (high, medium, low), so engineers can give a range estimate instead of false precision.
Pitfalls
- Precise-looking numbers without a source. Engineers will anchor on them.
- Writing "high performance" or "scalable": untestable words that cannot be estimated.
- PMs prescribing the database. It removes the engineers' freedom and puts the PM on the hook for the choice.
You are writing the success-metrics section of the PR/FAQ for an initiative to lift new-user onboarding completion. Which metrics would you commit to in the document, how would you define each precisely, and how do you write the section so the team can be held to it later?
Sample Answer
Direct answer
I commit to one primary metric, a small set of supporting metrics, and guardrails (metrics that must not get worse). Each has a precise definition, a baseline, a target and a date, plus a named owner and a review date. The section is written so the outcome can be judged by anyone later, without asking the author what we meant.
Metrics for "new-user onboarding completion" (onboarding means the first-time setup steps a new user completes; a cohort is a group of users who signed up in the same period and are tracked together)
| Role | Metric | Precise definition | Type |
|---|---|---|---|
| Primary | Onboarding completion rate | Users who finish all required steps within 7 days of signup divided by all users who signed up in that week | Leading indicator |
| Supporting | Step-2 (profile) completion in first session | Users who finish step 2 in the session they signed up divided by signups | Leading |
| Outcome | Week-4 retention of completers | Completers active in days 22 to 28 (the fourth 7-day block after signup: days 1-7 week 1, 8-14 week 2, 15-21 week 3, 22-28 week 4) divided by completers | Lagging |
| Guardrail | Support contacts per 1,000 signups | Onboarding-related tickets divided by signups times 1,000 | Must not rise |
Leading versus lagging: a leading metric moves within days and predicts the outcome we care about, while a lagging metric (retention) confirms it weeks later. Step-2 completion counts as leading because we expect (a hypothesis to verify on the baseline cohort, not a known fact) that users who finish the profile step in session one go on to finish onboarding more often, and we learn it on day one rather than in week four. Commit to both, or the team can inflate completion without creating lasting value.
Worked example (illustrative numbers)
- Baseline week: 10,000 signups, 4,200 complete within 7 days, so the completion rate is 4,200 / 10,000 = 42%.
- Target: stretch 50% (5,000 of 10,000) and committed 46% (4,600 of 10,000) by the end of the quarter, measured on the same definition. The team is held to the committed level; 50% is the level we aim for.
- Guardrail: 250 onboarding-related tickets on those 10,000 signups is 250 / 10,000 x 1,000 = 25 contacts per 1,000 signups. The commitment is that it stays at or below 25 while completion rises.
- Completers' week-4 retention baseline 30%, meaning 1,260 of 4,200; the target does not fall below 30%.
Second example: a social app's "introductions" metric
- Formula: users who send at least one introduction that the recipient accepts within 72 hours of signup, divided by all signups in the cohort.
- With 8,000 signups and 1,600 such users: 1,600 / 8,000 = 20%.
- Leading, because it predicts long-term connections. The lagging pair is the share of those users still messaging in week 4.
Writing it so the team can be held to it
- State the baseline with its date range and data source.
- Fix the definition now (cohort, window, exclusions such as test accounts). Changing it later requires a written note.
- Give a target and a date, and separate "committed" (the level we promise, e.g. 46% completion) from "stretch" (the ambitious level we aim for, e.g. 50%).
- Name the owner and the review date.
- State what happens on a miss: a review, not a blame session.
How it is gamed, and guards
Making steps optional inflates completion. The guard is the retention metric plus the requirement that "required steps" are listed in the document.
You are writing the PRD for a conversion-optimization feature that will be judged on numbers. Outline the document so a reviewer can see what is being built, for whom, how success will be measured and what could go wrong. Where do the hypothesis and the experiment plan live, and what do you leave out?
Sample Answer
Direct answer
I write a short PRD (Product Requirements Document) built around a hypothesis. The reviewer should see, in order: the problem and the opportunity in numbers, for whom, the hypothesis, what is being built, the success metrics with guardrails, the experiment plan, and the risks. The hypothesis sits right after the problem and users (item 4), so the reviewer meets the bet before the detail, and the experiment plan sits right after the success metrics (item 7), so the test is judged against numbers already agreed. I leave out design minutiae and solution detail that does not affect the test.
Outline
- Summary and decision needed (three sentences): what we are building, the expected impact, what we are asking approval for.
- Problem and baseline: current conversion rate with definition and date range, and where users drop off in the funnel (the sequence of steps to purchase).
- Target users: which traffic segment, and who is excluded.
- Hypothesis: "If we show a delivery-date estimate on the product page, then checkout conversion will rise, because uncertainty about delivery causes abandonment."
- What we are building: scope, non-goals, and requirements with acceptance criteria.
- Success metrics: primary metric (conversion), a target, and a guardrail (a second metric that must not get worse while the main one improves, such as return rate or page speed).
- Experiment plan: an A/B test (users randomly split into control and variant), split, duration, sample size rule, and the decision rule fixed in advance (the written-down condition for shipping, iterating or stopping).
- Risks and dependencies: what could go wrong and mitigations.
- Launch and rollback.
Worked example of the experiment section (illustrative)
- Baseline conversion is 3.0%. We commit to a 3.3% goal (10% relative lift, so 3.0 times 1.10 equals 3.3).
- Split 50/50, run for two full weeks to cover weekday and weekend cycles.
- Decision rule set before the test: ship if the variant beats control on conversion and return rate does not rise by more than the agreed tolerance (0.5 points in the written version below); otherwise iterate or stop.
- Sample size is computed with the data team before launch, not decided after seeing results. It depends on the baseline, the smallest lift worth detecting, and how sure we want to be. The usual two-proportion calculation (95% confidence, 80% chance to detect a real effect) gives roughly 53,000 users per arm for 3.0% to 3.3%, about 106,000 in total. Two weeks is enough only if about 7,600 eligible visitors arrive per day (106,000 / 14). If traffic is smaller, run longer or accept a larger minimum lift.
Hypothesis and experiment sections as they would read (illustrative)
Hypothesis: If we show a delivery-date estimate on the product page, checkout conversion will rise from 3.0% to at least 3.3%, because uncertainty about delivery causes abandonment.
Experiment: 50/50 split of product-page visitors for 14 days. Primary metric: checkout conversion. Guardrail: return rate within 30 days of purchase must not rise by more than 0.5 points versus control. Because that outcome matures 30 days after each user's order, we read conversion once, at the end of the 14 days, and read the guardrail once, 30 days after the last exposure (or use a shorter proxy such as 7-day returns if launch cannot wait). Decision rule: ship if conversion is above control at the pre-agreed sample size and the guardrail holds.
What I leave out
- Pixel-level design (linked mocks suffice).
- Implementation details, which belong in a design doc.
- Long market research, and metrics we will not act on.
Pitfalls
- Peeking at results early and stopping when they look good. Results bounce around by chance, so if you check daily and stop at the first good-looking moment, you will often stop on a random spike and ship a change that does nothing.
- A hypothesis with no mechanism ("it will improve things").
- Optimising conversion while return rate or support load quietly rises.
Unlock Full Question Bank
Get access to all 36 Product Documentation (PRD, PR/FAQ) interview questions and detailed answers.
Sign in to ContinueJoin thousands of developers preparing for their dream job.