The PM role
How product teams take an idea to launch
In this lesson8 sections
An unofficial lesson drawing on the methods in the Companies chapter of Cracking the PM Interview by Gayle Laakmann McDowell and Jackie Bavaro (2014). PracHub is not affiliated with the authors or publisher. Buy the book
Execution questions like "How would you take this from idea to launch?" test whether you know how work moves at the company in front of you. Companies follow different operating models, and the model decides what you write, who says yes and what you own in launch week. Most mix two or three of the models below.
One idea runs through this lesson: PracHub, the site you're reading, wants to pair two members for a live mock interview, each taking a turn as interviewer. Each section sends pairing down a different model's route, so read them as alternatives. Every figure is an assumption, not PracHub data.
A written narrative instead of slides
Some companies want a document, not a deck: a page for a feature or about six for a large bet, opening with the customer's problem. Amazon's six-page memo is the best-known example; its PR/FAQ (working backwards) starts from a mock press release and likely questions.
Prose exposes a weak step that slide bullets would hide. Reviewers check that each number leads to the next, from what you put in to what changes for the customer. The review often opens with everyone reading in silence, so the argument starts from the same facts, not from whoever talks first.
Pairing's narrative opens with the ask and its evidence in one line: fund two engineers and a designer for six weeks to test pairing for product sense and behavioral rounds with 300 members, because, say, a survey found 58% of 500 members had never done a spoken mock, mostly for lack of a partner at their level. Then it walks the chain one link at a time, each with the assumption a reviewer will push on:
Members practicing: suppose 20,000 in a month.
Partner requests: 2,000, if 1 in 10 asks.
Matches within two days: 700 sessions, if 70% of requests pair up.
Sessions held: 560, if 80% take place.
Weekly return: members who finish a session come back the next week more often than similar members who don't. This link is a guess, labeled as one, for the test to settle.
It closes with an FAQ: What if my partner doesn't show? (Two no-shows pause pairing for a month.) What does my partner see? (First name, target role, round and a one-line note on what I want to practice.) For Amazon itself, see our Amazon PM interview guide.
Writing a spec a tester can check
On this route pairing lives in a PRD (often an AI draft of the PM's notes, cut hard), sometimes posted as an RFC for comments. Judge each line by one test: could a tester mark it pass or fail? This one can: given two members who asked for a behavioral partner and share an hour of free time, when matching runs, both get the same proposed time in their own time zone within ten minutes.
Practice this on Engineering Spec With Acceptance Criteria for Apple's Online Apple Watch Shopping Page: write criteria for a single module, then stop.
The spec also says what pairing won't do, so nobody builds it by accident: no recording, group sessions or paid coaches in this version. Write those non-goals beside the goal they protect: members preparing for the same round find a partner and finish a spoken mock within two days of asking.
Staff book each other every Friday from the first build, and keep at it after launch. The goal and non-goals sort what they raise. Only something that stops a session, like a missing join link, goes on the fix list. A shared notes pad sounds useful, but a session still happens without one, so it goes on the parked list beside the non-goals.
To practice scoping, write only the goal, user, main use cases and non-goals for Databricks product spec: pick a capability and define scope, requirements and metrics, then stop. For a full PRD, try WhatsApp Plan-Making Feature for Group Outings: Scope an MVP and Write Its PRD.
Databricks product spec: pick a capability and define scope, requirements and metrics
A Databricks product manager exercise: choose any Databricks product and write a product requirements document for it.
WhatsApp Plan-Making Feature for Group Outings: Scope an MVP and Write Its PRD
A product design exercise set at WhatsApp in late 2021: people want to meet in person again, but most plans made in chats never happen.
The time-zone criterion shows how the spec's trio (PM, engineering lead and designer) splits decisions. Showing each member the time in their own zone is a what question, so the PM holds it even if engineering would rather start with one standard zone. How times are stored and converted is engineering's call, and how the proposed time looks on the match screen is usually the designer's, though on some PM-heavy teams, often where mistakes are costly, the PM decides the look too. Decision rights shift by team (what a product manager does covers the split).
Ship to a slice, then read the numbers
The default loop at most software companies: find a problem in usage data, build something, switch it on for a small share of users and check whether it helped. Pairing could start here, if the logs showed members opening a behavioral question's answer box and leaving without typing a word on 4 visits in 10. The rollout widens in steps, each with a bar agreed in advance:
| Who gets it | Move on when |
|---|---|
| Staff, in Friday sessions | Nothing open stops anyone booking or joining |
| A beta of 300 invited members | 70% of requests matched within two days |
| A random 10% of members, the rest held back | That 10% returns weekly more often than the held-back group; no-shows stay under 1 session in 10 |
| Everyone | Metric review two weeks after launch |
Count the whole 10%, including members who never asked for a partner: those who ask are keener anyway, so they'd beat the held-back group even if pairing did nothing.
Judge each stage by the success metric and a guardrail or two, without letting data run the product. A public leaderboard of top-rated partners might lift sessions 9% and still stay out, because it turns low-stakes practice into a contest. Keep moving, too: a four-week test on a button label costs more than the label is worth.
Practice this stage on Plan a Staged Launch of a New AI Feature at Perplexity, From Dogfood to Full Release, then on Strategic App Launch in a Month but Dogfood Misses Targets: Ship, Narrow or Delay?.
Plan a Staged Launch of a New AI Feature at Perplexity, From Dogfood to Full Release
A product manager question on how you would roll out a new feature at Perplexity.
Strategic App Launch in a Month but Dogfood Misses Targets: Ship, Narrow or Delay?
A product execution question about a strategically important app that, one month before launch, misses several internal dogfood targets including...
Newer paths: design sprints, time boxes and AI prototypes
Newer models test before a spec exists. A design sprint takes one problem to a prototype tested with about five users in a week. Shape Up, Basecamp's method, fixes the time (say six weeks) and cuts scope to fit. Dual-track teams test ideas with users while building the ones that passed.
On the quickest route, pairing starts as a clickable prototype an AI app builder makes in an afternoon. Two calls decide whether it goes further.
The first is the bar, written before anyone sees the prototype: at least 5 of 6 test members book a session from the match screen. A bar set afterward bends to fit whatever happened. Suppose 4 book and two ask to choose their partner's experience level. That's a miss, and a miss earns one more round: add the choice and show it to 6 new members. Suppose 5 book, clearing the bar. A second miss would mean rethinking or dropping the idea, not building it anyway.
The second is what the eval must prove before real engineering starts. An AI model proposes each partner, so the team pairs 200 sample requests by hand; the model's pick must rate as good as the hand pick in 90% of them, with no pair more than four time zones apart. A screen members like can't make up for a matcher that fails that test.
Hack weeks are the older form: pull in a few colleagues, build something nobody approved and let a working demo win support. They remain common; standing side-project time is rarer. Scrappy, technical teams prize a PM who builds the first version. Few large companies require PMs to code, but "I built a prototype and showed it to six users" is now a persuasive interview line.
Getting the idea adopted
On another route, pairing must win backers before anyone builds it, and merit alone rarely does that. Tie it to what the company already wants. If PracHub's goal for the year is more members practicing every week until their interview, pitch pairing as a reason to return next week, not a social feature:
"This year's goal is weekly practice. A booked mock with a partner is the firmest weekly commitment we can offer, and a 10% test will show within a month whether it raises weekly return."
The same goal decides what other teams will build for pairing. Reminders before a session are mobile's work, and mobile plans its own quarter. A reminder serves weekly practice directly, so it earns a place in mobile's plan on its own, and the pairing PM never has to borrow mobile's sprint. Pairing's team owns a narrower goal for the quarter, one it can move alone: 70% of requests matched within two days, and 80% of sessions take place.
Where teams set direction, often at engineering-led companies, nobody hands a PM the pairing brief. A new PM might spend the first weeks in the usage data and with the team, find the abandoned answer boxes, and bring an analysis, a recommendation and a meeting to decide. The team's agreement, not a sign-off from above, starts the build.
Where direction comes from the top, leadership picks the bet, and a build spanning many teams brings in a program manager to run it.
Launching across many teams
Suppose web builds booking, mobile builds reminders and email sends confirmations. A program manager or TPM is the hub, or the PM if there's neither. Three calls from pairing's launch show the job.
Reminders arrive an hour late after a clock change. The bug sits in mobile's code, so mobile owns the fix, but the hub owns the proof: it retests on the next build, clock change included, before calling it closed.
Mobile argues over sending reminders from the phone or the server. The hub asks what each means for time zones and for the launch date, then leaves the choice to engineers.
Email confirmations will miss the date. Slipping pairing or launching without them changes every team's plan, so the hub doesn't decide alone: it takes both options and their costs to leadership's weekly review, and every team plans around the answer. See how that job differs from a PM's.
Pairing also has to clear launch reviews of a few days to two weeks each, so plan them from day one. Start from what could go wrong for a member, then name the review that would catch it:
| What could go wrong | Review that catches it |
|---|---|
| A partner learns more about you than the FAQ promises | Privacy |
| Video drops at the evening peak | Infrastructure |
| A screen-reader user can't book or join, or gets no captions | Accessibility |
| A one-line note steers the matcher, or matches skew by name | AI safety |
| Someone uses matching to reach a particular person | Security |
Suppose 31 bugs are open a week out. With the engineering leads, the PM marks 9 to fix first (anything that breaks booking, shows the wrong time or exposes contact details), 15 for the week after launch and 7 not to fix, and cuts rebooking the same partner, which isn't ready.
Two weeks after launch the PM checks the narrative's targets. Suppose 61% of requests matched within two days against 70%, mostly missing evening slots in time zones with few members. The options: let those members match up to six zones away, after rerunning the eval, or recruit members there.
From then on, pairing questions from across the company come to you: support's reports of reminders landing in spam become bugs you chase, rebooking goes on next quarter's roadmap, and the next sprint's order is yours.
Sprints, backlogs and the product owner seat
On an agile team, pairing's PM is often the product owner too, ordering the backlog sprint by sprint. In the second sprint a reschedule button dropped below the no-show report: no-shows threaten the guardrail, and members can already cancel and book again.
The PM then wrote the report as a story with a criterion, so engineers knew when it was done: "As a member whose partner never joined, I want to report it in one step, so the matcher stops pairing me with people who skip." Done means: given a member reported as a no-show twice, when matching next runs, they get no partner for a month and see why.
To practice, answer Agile vs. Scrum Explained, and How a PM Runs the Backlog, Sprints and Ceremonies, then Contrasting the Product Manager and Product Owner Roles on an Agile Team.
Agile vs. Scrum Explained, and How a PM Runs the Backlog, Sprints and Ceremonies
A Capital One product manager interview question that asks the candidate to explain Agile and Scrum and how they apply them in practice.
Contrasting the Product Manager and Product Owner Roles on an Agile Team
A product management interview question that asks the candidate to explain how the product manager and product owner roles differ.
Reading a company's model before you interview
The posting and the interviewers usually give the model away:
| What you notice | Likely model | Prepare |
|---|---|---|
| A writing exercise | Narrative | A one-page memo |
| "Write PRDs" | Spec and trio | A spec with acceptance criteria |
| "Own the backlog", "Scrum" | Agile | Stories you wrote |
| "Experimentation", a metrics round | Data loop | A test and how you'd read it |
| "Prototype", "zero to one" | Prototype path | What you built and what users did |
| TPMs on the panel, "cross-functional launches" | Program-run | A launch plan with reviews |
To check, ask: "How did the last thing your team shipped get approved?" When asked to take an idea to launch, name your model first:
"I'll assume a team that writes narratives and tests on a slice of users, so I'd start with a one-pager on the problem and its impact chain."
Some interviewers ask for your own model instead. Name the one your last team used and how an idea got approved there, then answer Present Your PM Operating Model From Customer Input to Engineering Delivery.