Product manager interview flashcards
Flashcards for product manager interviews: rounds and rubrics, product sense, metrics and A/B tests, estimation, strategy, execution and take-homes.
270 cards. Read the question, answer it in your head, then open the card to check.
-
Which six stages do most product manager hiring processes share?
First the application is screened and a recruiter calls. One or two screens then sample a few skills, some roles add a written or project task, a final loop covers the rest, and the decision is taken from every interviewer's feedback. Employers skip, merge and reorder these stages, so treat them as a pattern.
From the lesson How product manager interviews work
-
What separates an interview loop from a phone screen?
A screen is usually a single interview that samples a few skills and decides whether you go further. The loop is the last stage: several interviews, each run by a different interviewer judging a different part of the role, and the decision draws on all of them. Prepare the screen as seriously as the loop.
From the lesson How product manager interviews work
-
Roughly how long do the interviews of a published product manager process add up to?
From employers' own pages, about four hours in one process and more than five and a half in another, where a 60-minute screen is followed by five interviews of 55 minutes. Plan the loop day like a full working day, with breaks and food.
From the lesson How product manager interviews work
-
Which eight kinds of round do product manager loops draw on?
Product sense and design, metrics and analytics, estimation, strategy, execution and prioritisation, technical, behavioural and leadership, and written or take-home work. Most loops use four or five of them, and one can sit inside another, such as a case inside a leadership interview.
From the lesson How product manager interviews work
-
Which six things should you note from an employer's own page about its hiring process?
The stages and their order, the length of each round, who interviews you, what is assessed, the rules (AI tools, notes, time limits on written work) and when you hear back. Anything the page leaves out becomes a question for your recruiter.
From the lesson How product manager interviews work
-
How do you decide which skill to prepare first for a process?
Score one practice answer per skill from 1 to 4, then multiply the rounds that test the skill by its gap, 4 minus the score; the biggest product goes first. Anything with a fixed deadline, such as a take-home, goes into the calendar before that, and a skill you score 1 on deserves an early first try.
From the lesson How product manager interviews work
-
What makes an interview structured?
Every candidate for the role gets comparable questions, each answer is scored signal by signal against one rubric from notes of what was said, and the decision reads all the interviewers' feedback together instead of resting on one person's impression.
From the lesson What product manager interviewers assess
-
Which seven signals recur in product manager rubrics?
User focus, structure, judgement, data, execution, leadership and communication. Each one needs evidence an interviewer can quote, such as a named user group together with the reason it matters most.
From the lesson What product manager interviewers assess
-
What turns 'I'm very data-driven' into evidence of the data signal?
A named metric with a value today, a target and a guardrail, for instance the share of new users who complete sign-up during their first visit: about 40 % now, expected to reach 50 % within a month.
From the lesson What product manager interviewers assess
-
What is the difference between an output and an outcome?
An output is what the team produced, such as a feature shipped. An outcome is what changed because of it, for whom and by how much, such as abandoned mobile carts falling from 31 % to 24 %. Outputs show effort; outcomes show value, which is what interviewers score.
From the lesson What product manager interviewers assess
-
What should you say when you cannot share a former employer's exact results?
Give the direction and a rough size, such as about a third fewer refund tickets, and say that the exact figure is confidential. If the result was never measured, say so and name what you would measure now.
From the lesson What product manager interviewers assess
-
What do the four steps of the track's rubric scale mean?
1: not shown. 2: partly shown. 3: shown. 4: shown at the level above the role. A score counts only when it quotes the sentence from the answer that earned it.
From the lesson What product manager interviewers assess
-
In the gym-booking sample answer, what lifted execution from 1 to 2, and what would reach 3?
Saying what ships first and where: the waiting list at two of the busiest gyms. A stop rule would reach 3, for example ending the trial if under a third of the offered places get taken.
From the lesson What product manager interviewers assess
-
Which three things grow as a product manager moves up a level?
Scope, from a feature to a product area to a portfolio; ambiguity, from a problem handed to you to choosing which problems matter; and influence, from one team to teams across the company, all brought along without managing them.
From the lesson Product manager roles, levels and expectations
-
Should you judge your level by your years of experience or your job title?
By neither. Judge it by evidence of scope, ambiguity and influence: three years spent setting a direction across four teams is staff-level evidence for that part of the role, and titles mean different things at different employers.
From the lesson Product manager roles, levels and expectations
-
In one public job family, how many teams does each level from product manager to principal span?
A product manager works on a feature set within one team, a senior on a product area across one or two teams, a staff product manager on a domain across three to five, and a principal on company-level problems across five or more.
From the lesson Product manager roles, levels and expectations
-
How is a product manager judged differently from a program or project manager?
A product manager is judged on outcomes: did the product change something for users and the business? Program and project managers are judged on delivering agreed work to scope, time and cost. In a product loop, end every story with what it changed.
From the lesson Product manager roles, levels and expectations
-
What does a technical product manager loop add to the usual product questions?
Questions on technical depth, such as the technical product life cycle. The role suits technical products like platforms, APIs and infrastructure.
From the lesson Product manager roles, levels and expectations
-
What does a strong answer add at each level, from associate to principal?
Associate: a sound method and empathy for users. Product manager: metrics and trade-offs. Senior: strategy, sequencing and how teams split the work. Principal: bets over several years and how other teams are brought along.
From the lesson Product manager roles, levels and expectations
-
What is the usual failure in pitching an answer to a level, and how do you avoid it?
Pitching below it: nothing said is wrong, but nothing shows the scope the role needs. Aim at your level's row, add one sentence from the row above, and ask the recruiter which level the loop is for when the invitation does not say.
From the lesson Product manager roles, levels and expectations
-
What are the six steps of the track's method for open product questions?
Clarify the goal and constraints; pick one user group and say why; name its most important problems; generate a few distinct solutions; decide on criteria said aloud; then set success metrics, guardrails and the main risks.
From the lesson A method for open-ended product questions
-
Why choose the goal and a user group before generating ideas?
Every idea is judged against them. Without a goal and a user, features cannot be compared with each other, so user focus and judgement stay unshown however good the ideas are.
From the lesson A method for open-ended product questions
-
When is a clarifying question worth asking?
When a different answer would change your plan: the goal (users, use or revenue), whether the product is new or existing, the platform and market, and constraints such as time or rules. Ask two to four, then state the rest as assumptions and move on.
From the lesson A method for open-ended product questions
-
How does the track split a 45-minute product round?
Clarifying takes about 5 minutes, users and their problems 10, solutions another 10, the decision 5, and metrics with risks a final 5. That leaves 10 minutes for the interviewer's follow-ups, and a round of another length keeps the same shares.
From the lesson A method for open-ended product questions
-
Why check in with the interviewer at the end of each step?
A one-line check such as 'I'll go with people who live alone; does that work?' takes seconds, catches a misunderstanding while it is cheap to fix and keeps you on time, which interviewers also read as judgement.
From the lesson A method for open-ended product questions
-
Why does a list of good features still score low?
Nothing in it says whom the features serve, which problem they solve or why one beats another. Rebuild it from the goal down: goal, user, problem, a few solutions, one decision with its reason, and a metric with a guardrail.
From the lesson A method for open-ended product questions
-
What is the diagnostic that starts a study plan?
One unseen prompt for each round of your loop, answered aloud in 15 minutes and scored from 1 to 4 the next day, quoting the sentence behind each score. Waiting a day stops you grading what you meant instead of what you said.
From the lesson A study plan for product manager interviews
-
How does the study plan weight each round?
By the number of loop rounds that test it times its gap, 4 minus the diagnostic score, and never below 1, so that a round you already do well still comes back.
From the lesson A study plan for product manager interviews
-
What does the learning research suggest instead of rereading your notes?
Test yourself by answering prompts aloud, spread each round's answers across the weeks and mix the rounds within a week. Research weighing ten study techniques rated practice testing and spreading study over time highly, and rereading and highlighting low.
From the lesson A study plan for product manager interviews
-
What shape does the track recommend for four weeks of preparation?
Week 1: the method and one round a day. Week 2: every round mixed. Week 3: mocks. Week 4: a mock loop and fixes, ending on a light day followed by rest so that you arrive fresh.
From the lesson A study plan for product manager interviews
-
What goes on one line of an error log?
The prompt, the round, what went wrong and the fix to practise. The lines change the scores and so the days that follow; a log that never changes the plan is rereading by another name.
From the lesson A study plan for product manager interviews
-
In the study plan's example, what changed after strategy went wrong in two mocks?
Strategy's score fell from 2 to 1, so its weight rose from 2 to 3, and the replanned last days gave it three answers instead of two, taking one from product sense, which the mocks found no fault with.
From the lesson A study plan for product manager interviews
-
Why practise answers out loud rather than only reading or writing them?
Delivering an answer while someone watches is a separate skill. In a randomised trial with 48 students, those solving a task at a whiteboard in front of an interviewer did less than half as well as those working in private. Writing prepares the content, not the delivery.
From the lesson Practising product manager interviews out loud
-
What goes into a solo mock?
An unseen prompt from a round of your loop, a clock set to the round's length, the answer spoken aloud and recorded on your own device, and a score the next day that quotes one sentence for each signal. The slips go into a log, and the one seen most often picks the next drill.
From the lesson Practising product manager interviews out loud
-
Why score a recorded mock the next day rather than straight away?
Straight after answering, memory fills in what you intended. The next day, the recording lets you judge the words you really used.
From the lesson Practising product manager interviews out loud
-
Which roles does a peer mock need?
An interviewer, who picks an original prompt, keeps time and asks two follow-ups written on a card; a candidate, who has not seen the prompt and scores the answer before hearing feedback; and, if possible, an observer, who holds the rubric and notes evidence.
From the lesson Practising product manager interviews out loud
-
What does useful feedback after a mock contain?
A single strength and a single fix, each backed by a quotation and the minute it came, such as 'at minute 22 your choice had no trade-off'. A remark like 'good, a bit long' cannot be checked against the recording and leaves nothing to drill.
From the lesson Practising product manager interviews out loud
-
How does the error log pick the next drill?
Take the slip that appears most often across mocks, even when it shows up in different rounds, and practise that one skill on its own until it stops appearing. 'No trade-off said' in three rounds is one skill to drill, not three round problems.
From the lesson Practising product manager interviews out loud
-
May you use AI tools to prepare for, and during, product manager interviews?
It depends on the employer and the stage. One may welcome responsible use while you prepare but expect your own work in interviews unless help is explicitly allowed; another may assess how you use AI in a round of its own. Read the employer's page and ask the recruiter before the day, never during the interview.
From the lesson Practising product manager interviews out loud
-
Why group users by situation and need rather than by an age band?
One age band can hold a nurse home from a night shift, a home baker and a man shopping for parents far away, each wanting something different, so no idea can be judged against all three. A segment shares one situation and need; age, city and income help later to count and reach it.
From the lesson Users, needs and pain points
-
What template does the track use to write a job to be done?
'When …, I want to …, so I can …': the situation, the progress the person wants and why it matters to them. A job names that progress, never a feature.
From the lesson Users, needs and pain points
-
Which three sides can a job to be done have?
A functional side, such as groceries reaching a parent's door; an emotional side, such as feeling that you look after your parents; and a social side, such as the parents not wanting to feel a burden. Forget one and the product does the job badly, however fast it is.
From the lesson Users, needs and pain points
-
What does a strong problem statement say?
Who is affected, what stands in their way and what it costs them today, in words the user would recognise and based on evidence rather than on a solution. 'Users want a reorder button' fails because it names a feature, so it can never be wrong.
From the lesson Users, needs and pain points
-
Which five criteria does the track use to choose a target segment?
Size, pain, frequency, reach and fit with what the company already does well, each scored from 1 to 5 and weighted by the company's goal. The same three segments rank differently under another goal, so state the goal and the criteria before choosing.
From the lesson Users, needs and pain points
-
How do you rank a segment's pain points?
Multiply how often each pain happens by how badly it hurts on a scale of 1 to 5, then name the top two or three and say why the first leads. A rare but dangerous pain, such as the wrong medicine reaching an elderly parent, can overrule the score.
From the lesson Users, needs and pain points
-
How do you handle numbers in a product round when you have no data?
Label each one as an assumption and say how you would check it, for example by counting second orders to the same address within two days. In the room the evidence is your reasoning, not data you pretend to have.
From the lesson Users, needs and pain points
-
What sets a 'design a product' question apart from improving or critiquing one?
Nothing exists yet. The prompt names a group of people and a situation, so the user and the problem you choose carry the whole answer.
From the lesson Designing a product for a user group
-
What makes the solutions in a design answer distinct?
They differ in approach, not in features. A directory, a managed service and a neighbours' board are three approaches; search, filters and reviews are three features of one directory.
From the lesson Designing a product for a user group
-
Which criteria should you name before choosing between design solutions?
The effect on the chosen problem, the effort, the risk and the fit with the goal, said aloud before the choice, so the interviewer can follow why one approach beat the other two.
From the lesson Designing a product for a user group
-
What is a minimum viable product in a design answer?
The smallest version real users can try that shows whether the riskiest assumption holds. For a managed home-repair service that assumption is whether new tenants will let a vetted worker in at a fixed price, so version one offers three jobs in two neighbourhoods and leaves the rest out.
From the lesson Designing a product for a user group
-
Why is 'version one has every feature, with fewer screens' a weak first version?
A smaller copy of everything tests no assumption in particular and takes nearly as long to build. Name the one assumption the first version tests, and say what it leaves out and why.
From the lesson Designing a product for a user group
-
How do the metrics of a design answer fit together?
One primary metric tied to the goal, here first customers booking a second job within 60 days; supporting metrics that explain why it moves; and a guardrail that must not rise, such as damage complaints per 100 jobs. Give the primary metric a number to beat.
From the lesson Designing a product for a user group
-
Where should an answer to 'How would you improve X?' start?
With what the product is for and its funnel: the steps people take towards that goal and how many reach each one. Features come only after the weak place has been found.
From the lesson Improving an existing product
-
What is a funnel step's conversion rate, and how do steps combine?
The share of people reaching a step who get through it. Overall conversion is the product of the step rates: 0.6 × 0.4 × 0.5 = 0.12, or 120 bus passes for every 1,000 people who open the app.
From the lesson Improving an existing product
-
Why does the same five-point gain help the weakest funnel step most?
Because the step rates multiply: five points lift a 40 % step by an eighth but a 60 % step by only a twelfth, and the total rises by the same share, 135 passes against 130. Gains of different sizes still need the arithmetic.
From the lesson Improving an existing product
-
Steps at 60 %, 40 % and 50 % each gain five points. What is the new overall rate?
About 16.1 %, since 0.65 × 0.45 × 0.55 = 0.161: 4.1 points above today's 12 %, not 15 points more, because step rates multiply rather than add.
From the lesson Improving an existing product
-
Which five growth stages does the track use to frame a growth question?
Reaching new users, their first success, return visits, invitations to others, and payment. Start with whichever stage is weakest in the numbers.
From the lesson Improving an existing product
-
Why split a funnel into new and returning users before choosing what to fix?
The two groups often stop in different places, such as new riders at the ID upload and renewing riders at payment, so an average of both can point at a step that neither group finds hardest.
From the lesson Improving an existing product
-
What are the parts of a testable product hypothesis?
The change, the users, the metric with its start and target, and the reason, for example: one-tap renewal for monthly riders will lift renewals within two days of a reminder from 55 % to 65 %, because retyping details is where they stop. Add a guardrail, and decide before the test which result drops the idea.
From the lesson Improving an existing product
-
What do favourite-product and critique questions really test?
Whether you see a product through its users and its business: whom it serves, why what works works, what fails and for whom, and one change with its cost and metric. Liking the product scores nothing on its own.
From the lesson Critiquing a product you use
-
What are the five parts of a strong critique, in order?
Who the product is for and the job it does; what it does well, and why; what it does poorly, and for whom; why it might be that way; and one change with its trade-off and its metric.
From the lesson Critiquing a product you use
-
Which product should you choose for a critique question?
One you use often and whose users you can describe, such as an app you open every morning. Talk only about what any user can see, never confidential details of a product you worked on.
From the lesson Critiquing a product you use
-
Which four reasons can make a design choice that looks like a flaw deliberate?
A constraint (time, a platform's rules, the cost of data), the business model, a compromise between two user groups, or a change people are still getting used to. Give the likely reason before suggesting a fix.
From the lesson Critiquing a product you use
-
How do you make a usability complaint checkable?
Name the user and the task, then measure the complaint against a standard. WCAG 2.2 sets a minimum contrast ratio of 4.5:1 for text and a minimum target size of 24 by 24 CSS pixels, which turns 'too small for a thumb' into a number you can check.
From the lesson Critiquing a product you use
-
What makes a strong critique stronger than 'It's clean'?
Every sentence can be checked by someone holding the app, such as 'one line covers the next two hours, which is what most visits want'. 'It's clean' names no user and no task, so nothing could prove it wrong.
From the lesson Critiquing a product you use
-
Which four constraints change a product design for users in India?
Device (a basic or a smart phone, shared or owned), network (whether data is steady and affordable), language (which script people read and type, or whether they speak instead) and payment (which ways to pay work on their phone). Anchor each one with an official figure.
From the lesson Designing products for users in India
-
Why is a count of telephone subscriptions not a count of users?
A subscription is a connection: one person may hold two SIMs, a family may share one phone, and some connections are not active (6.6 % on the peak day). Turn the counts into estimates per person and label the assumptions you add.
From the lesson Designing products for users in India
-
How do urban and rural telephone subscriptions per 100 people compare in India's regulator data?
About 155.7 in cities against 61.5 in villages, a gap of 2.5 times. If a fifth of rural owners hold two SIMs, only about 47 villagers in 100 have an active phone of their own, so a village design starts from a shared or basic phone.
From the lesson Designing products for users in India
-
How can someone with a basic phone pay by UPI?
Through UPI123Pay's four routes: an app on the basic phone, a missed call, a voice menu (IVR) and sound-based payments between phones that are close together. The central bank put feature-phone subscribers at more than 40 crore.
From the lesson Designing products for users in India
-
What does each channel need from a user: an app, a voice menu, a missed call or assisted use?
An app needs a smartphone, data and reading. A voice menu works on any phone without data or reading. A missed call costs the caller nothing and brings a call back or a text. Assisted use needs a trusted person, such as a family member or a shopkeeper.
From the lesson Designing products for users in India
-
When should you choose the channel in a design for users in India?
Before designing any screen: choose it from the group's device, network, reading and trust, measured or estimated from official data. Basic phones and little reading rule the app out.
From the lesson Designing products for users in India
-
Which WCAG 2.2 conformance level should a product set as its launch bar, and why?
Level AA, the middle of A, AA and AAA, because it holds the contrast and target-size criteria. An app that a screen-reader user cannot pay in has a bug as real as a crash.
From the lesson Accessibility, privacy and dark patterns
-
Name four accessibility checks a product manager asks for before a launch.
Text alternatives for images and icons (1.1.1); the whole flow usable from a keyboard (2.1.1); a 4.5:1 contrast ratio for text, or 3:1 if it is large (1.4.3); and tap targets no smaller than 24 by 24 CSS pixels unless spaced apart (2.5.8). A sign-in with no puzzle or memory test (3.3.8) is a fifth.
From the lesson Accessibility, privacy and dark patterns
-
What is a dark pattern, and why does it hurt the product that uses it?
A screen built to push people into a choice they would not otherwise make: an extra purchase, more data shared, a subscription they meant to leave. It lifts a number today and costs trust and refunds later, and India's consumer authority lists 13 such patterns that platforms must not use.
From the lesson Accessibility, privacy and dark patterns
-
What is drip pricing, and what is the honest design?
Revealing fees one step at a time, so that the first price shown is untrue even when each fee is fair. The honest design shows the full price from the very first step.
From the lesson Accessibility, privacy and dark patterns
-
What are the honest alternatives to confirm shaming, a subscription trap and interface interference?
A plain 'No, thanks' instead of a guilt-laden refusal; cancelling as quickly as joining; and two equally prominent choices instead of a pale 'No' beside a bright 'Yes'.
From the lesson Accessibility, privacy and dark patterns
-
What quick test tells you whether a screen is steering people?
Ask whether the user would still choose the same if every option were shown equally, with its full price, on the first step. If not, the screen is steering them.
From the lesson Accessibility, privacy and dark patterns
-
What does valid consent require under India's data protection law?
The Act asks for consent given freely, for a specific purpose, informed, unconditional and unambiguous, signalled by a clear affirmative action and covering only the data that purpose needs, with a notice before or alongside each request. Ask for one purpose at a time and make withdrawal as easy as agreeing; pre-ticked boxes and bundled purposes do not count.
From the lesson Accessibility, privacy and dark patterns
-
What does 'right 9 times in 10' mean for the users of an AI feature?
It is still wrong for about 100 of every 1,000 people who use it, so the design has to decide what the product does in those moments, not only when the answer is right.
From the lesson Designing AI product features
-
When should a task get rules, a model or a person?
Rules for fixed inputs. A model for varied inputs, such as typed messages, where an occasional error is acceptable and easy to correct, with a person on the costly or uncertain cases. A person where every decision must be explained and an error is costly.
From the lesson Designing AI product features
-
Which design moves shape the experience when an AI feature is wrong?
Make clear how well it works, narrow what it offers when in doubt, keep correction to one step (one tap to send, one to fix) and change its behaviour cautiously over time. Show no suggestion at all when the model is unsure.
From the lesson Designing AI product features
-
What do precision and recall mean for an AI suggestion feature?
Precision is the share of the suggestions shown that are right; recall is the share of the right suggestions that get shown. Raising the confidence threshold usually trades recall for precision.
From the lesson Designing AI product features
-
In the help-desk example, what happened when the threshold rose from 0.3 to 0.9?
Precision rose from 66.7 % to 94.1 % while recall fell from 71.2 % to 13.6 %. A person then handled 91.5 % of messages, and agent minutes per 1,000 messages grew from 2,345 to 3,262.
From the lesson Designing AI product features
-
Which three measures decide whether an AI feature is ready to launch?
Quality on a labelled test set, as precision and recall at the threshold you will ship; waiting time, at the median and for the slowest 5 %; and cost per request, from tokens and prices.
From the lesson Designing AI product features
-
How does a product reduce confident wrong answers and over-reliance on an AI feature?
Show what each answer rests on and hold back below the threshold; make the person's check a real step and sample accepted answers every week. Measure quality by language and segment too, and send only the fields the task needs.
From the lesson Designing AI product features
-
What should a north star metric count?
The value customers actually receive, such as good orders a week for a food-delivery app (on time, complete and not refunded), rather than app opens or downloads, which only show that people came.
From the lesson Product metrics and the north star metric
-
Why is monthly revenue a weak north star?
It counts what already happened and can rise while customers get less, for example through higher fees. Count the value that earns next month's revenue instead, and watch revenue beside it.
From the lesson Product metrics and the north star metric
-
What north star fits each kind of product: attention, transaction and productivity?
Attention: absorbed time, such as hours of podcasts played to the end each week. Transaction: the right thing bought and delivered, such as good orders a week. Productivity: a job done, such as documents shared and opened by someone else each week.
From the lesson Product metrics and the north star metric
-
What are input metrics, and how do they relate to the north star?
The few factors a team can move in its daily work that produce the north star, typically three to five. For the food app they multiply: ordering customers × orders per customer × share of orders that are good, so 5 % more of any one gives 5 % more good orders.
From the lesson Product metrics and the north star metric
-
How do you choose the guardrail for a team pushing an input metric?
Pick the metric its push would damage first, never the input it is paid to raise. More offer notifications are guarded by opt-outs per 1,000 customers; later promised delivery times by the median promised time.
From the lesson Product metrics and the north star metric
-
What are the five HEART categories, and how do you get from them to a metric?
Happiness, Engagement, Adoption, Retention and Task success. Go from a goal to a signal, what users would do or feel if the goal were met, and then to the metric that counts the signal. Report Engagement per user rather than as a total.
From the lesson Product metrics and the north star metric
-
What is stickiness, and why can two DAU/MAU figures for one app differ?
Average DAU divided by MAU: the share of a month's users who are active on a typical day. It depends on what 'active' means; the lesson's month of events gave 15.0 % when counting app opens and 6.0 % when counting orders, so read the definition before comparing.
From the lesson Product metrics and the north star metric
-
What is a vanity metric, and what should replace it?
A total that only grows, such as downloads since launch or registered users, so it cannot show whether a change worked. Use a decision metric that can fall: a rate, an average per user or a count within a fixed period.
From the lesson Product metrics and the north star metric
-
In what order do you choose a new feature's success metrics?
The goal first, then the signal that would show the goal was met, then the metric that counts that signal. A metric picked first tends to be whatever the dashboard already shows, like splits created, which says nothing about the goal.
From the lesson Success metrics for a new feature
-
What is the difference between a feature's adoption and its outcome?
Adoption counts the people who started using it; the outcome says whether it did its job. In week 1, 9 % of weekly active users tried split the bill, yet just 48 % of their splits were settled in full within 3 days, below the 55 % of ordinary requests: tried, but not yet working.
From the lesson Success metrics for a new feature
-
What is a leading indicator, and how does it earn that role?
An early behaviour that predicts a slow outcome, such as a second split within 7 days predicting use at day 90. It earns the role by predicting the outcome in past data, and it must be checked again once the outcome arrives.
From the lesson Success metrics for a new feature
-
What goes wrong when a team pushes a leading indicator directly?
A reward aimed at the indicator itself, such as a prize for a second split, can raise it without the habit it used to reveal, so it stops predicting the outcome.
From the lesson Success metrics for a new feature
-
How do you set a target for a feature that has no history of its own?
Use the behaviour it replaces as the baseline (ordinary requests paid within 3 days 55 % of the time), add the smallest gain worth the feature's cost, 10 points, and give the target a date: 65 % by week 8.
From the lesson Success metrics for a new feature
-
Why fix a feature's judgement window, target and guardrail limits before launch?
Judged too early, first-week curiosity or first-week dislike decides the verdict. Set the date, the target and each guardrail's limit, such as at most 4 failed payments per 1,000, before the launch: afterwards almost any number can be made to look like success.
From the lesson Success metrics for a new feature
-
What is the first thing to check when a metric drops?
Whether the drop is real. A changed definition, an event that stopped being logged after a release, data that has not fully arrived or filtered bot traffic can all move the dashboard while users behave as before.
From the lesson Diagnosing a sudden metric drop
-
Which two checks catch most drops that exist only in the data?
Compare the metric with a second source that counts the same thing another way, such as payments received against order events, and compare the drop with the metric's usual swing, such as the same weekday in earlier weeks.
From the lesson Diagnosing a sudden metric drop
-
What does the shape of a drop over time suggest?
A dip lasting one day points at an outage or a holiday, a step at a single change, and a slope at a change that reached more users each day.
From the lesson Diagnosing a sudden metric drop
-
Into which segments should you split a metric drop?
Platform, app version, region, new or returning users and acquisition channel. In the grocery example the whole fall sat on Android, down 27.0 %, and on the same days the new version turned 4.5 % of sessions into orders against 7.5 % for the old one.
From the lesson Diagnosing a sudden metric drop
-
How do you separate internal from external causes of a drop?
Internal causes are the team's own releases, experiments, price changes, outages and campaigns, checked against the change log and the experiment list. External ones are seasons, festivals, weather or a competitor, checked against a calendar and last year's weeks. Test first the cause that fits both where and when.
From the lesson Diagnosing a sudden metric drop
-
Why compare two app versions on the same days?
Time is one of the biggest sources of change in observed metrics, so comparing versions that ran side by side takes the calendar out of the comparison.
From the lesson Diagnosing a sudden metric drop
-
What is a mix shift, and how can it lower conversion when no segment got worse?
A change in the share each segment holds: when a low-converting group grows, the overall rate falls although every segment holds steady. In the lesson's shop, social-video visitors grew from 30.0 % to 58.3 %, pulling conversion from 4.80 % to 3.75 %, a mix effect of −1.13 points.
From the lesson Diagnosing a sudden metric drop
-
What does a funnel count?
The users who reach each step of a path, in order and within a stated window from the first step. Each step has its own conversion, its users over the previous step's, and an overall conversion, its users over everyone who started.
From the lesson Funnels, cohorts and retention
-
Why must a funnel's numbers always come with the window?
The window decides who counts: a user who starts a first workout on day 3 is lost in a 24-hour funnel but kept in a weekly one, so the same events give different rates.
From the lesson Funnels, cohorts and retention
-
Is the funnel step that loses the most users always the weakest step?
No. In the workout app, creating an account lost the most people, 25 of 80, while starting a workout had the lowest step rate, 56.4 %.
From the lesson Funnels, cohorts and retention
-
How do you read a weekly cohort retention table?
Each row is one cohort's curve, read from left to right; each column compares cohorts at the same age. The table is a triangle because recent cohorts have not reached their later weeks yet, so those cells are unknown, not 0 %.
From the lesson Funnels, cohorts and retention
-
What does a retention curve that flattens tell you?
That a core of users gets something from the product. Cohort 1 went 100 %, 42 %, 32, 28, 25 and 25, so about a quarter keeps coming back; a curve that keeps sliding towards zero keeps nobody for long.
From the lesson Funnels, cohorts and retention
-
How do you state retention and churn so that they mean something?
As rates for a named group, a period and a definition of active. Of 114 users active one week after joining, 87 came back the following week: retention of 76.3 % and churn, its complement, of 23.7 %.
From the lesson Funnels, cohorts and retention
-
What shapes do most product SQL interview questions take?
Counts of users or events, conversion by a segment such as channel, users joined to what they did, counts by day or week, each user's first or latest action, and the change from one week to the next. Each counts something per something, so name both before writing a line.
From the lesson SQL for product manager interviews
-
Which table should a product SQL query start from?
The table that holds the question's people, such as users, so that every join can be checked against it: the result should still have one row per person, or per group of people.
From the lesson SQL for product manager interviews
-
Why use COUNT(DISTINCT user_id) rather than COUNT(*) after a join?
A join can multiply rows: joining users to orders turned search's 6 users into 13 rows, one for each order plus one for the user who had none. COUNT(*) then counts orders, while COUNT(DISTINCT user_id) counts the 6 people.
From the lesson SQL for product manager interviews
-
How do you keep the users who did nothing in a product query?
Start from the users table and LEFT JOIN the activity to it, which keeps people with no matching row. An inner join, or counting from the orders table alone, drops them.
From the lesson SQL for product manager interviews
-
What do ROW_NUMBER() and LAG() answer in product SQL?
ROW_NUMBER(), partitioned by user and ordered by day, numbers each user's orders so that number 1 is the first one; LAG() reads the previous row, such as last week's count. Window functions are not allowed in WHERE, so number the rows in a CTE and filter afterwards.
From the lesson SQL for product manager interviews
-
Why does the most common first order matter more than the most-ordered item for a welcome offer?
A welcome offer targets what new buyers pick first. In the lesson's grocery data milk is ordered most overall, 9 of 29 orders, but rice is the most common first order, 4 of 15.
From the lesson SQL for product manager interviews
-
How should you state a SQL result in an interview?
In one sentence with its denominator and the checks behind it, for example: 5 of the 6 users who came from search ordered at least once, 83.3 %, with each user counted once and the user without an order kept in.
From the lesson SQL for product manager interviews
-
Why can an A/B test put a difference between its groups down to the change?
Users are assigned at random either to the change or to the current product and observed over the same days, so the change is the only systematic difference between the groups.
From the lesson Designing an A/B test
-
What must an A/B test design fix before the first user is assigned?
The hypothesis (the change, the users, the expected effect and why), one primary metric that decides, guardrails such as cancellations or load time, and the minimum detectable effect: the smallest lift worth shipping, set from cost and value, never from the data.
From the lesson Designing an A/B test
-
When should a test randomise something other than users?
When the change acts on something users share. If priority dispatch for treated customers takes riders that control customers would have had, control looks worse and the gain is overstated, so randomise cities or zones switched on and off by the hour, at the price of fewer units and less power.
From the lesson Designing an A/B test
-
Which four numbers set the users per group for a test on a rate?
The baseline rate, the minimum detectable effect, the significance level (by convention a two-sided 5 %) and the power (usually 80 %). For a 10 % baseline and a one-point lift, the formula gives 14,751 users per group.
From the lesson Designing an A/B test
-
What happens to the users needed when the effect you want to detect halves?
They almost quadruple, since the effect enters the formula squared: half a point on a 10 % baseline needs 57,763 users per group instead of 14,751.
From the lesson Designing an A/B test
-
How long should an A/B test run?
A whole week at least, then whole weeks, even when traffic fills the sample sooner, since weekend visitors often behave unlike weekday ones. Set the length before launch and look at the result a single time, when the test ends.
From the lesson Designing an A/B test
-
When is an A/B test the wrong tool?
When there are too few users to detect an effect worth shipping, when an announced change must reach everyone at once, when legal or ethical limits forbid holding back a control group, or when the question is why rather than what. Use a staged launch, research or a bolder change instead.
From the lesson Designing an A/B test
-
How do you compute and state the lift of an A/B test?
Treatment rate minus control rate: 10.9 % − 10.0 % = +0.90 points, which is 9.0 % in relative terms. Say which of the two forms you are quoting.
From the lesson Reading A/B test results
-
What does a p-value of 0.0375 mean, and what does it not mean?
If the change did nothing, a difference at least this large would turn up by chance about 4 times in 100. It is not a 96 % chance that the change works.
From the lesson Reading A/B test results
-
Why read the confidence interval before the p-value?
Its position says whether the lift is likely real, and its width says how sure you can be of the size: +0.05 to +1.75 points means probably above zero, of uncertain size.
From the lesson Reading A/B test results
-
What is a sample ratio mismatch, and what do you do when a test has one?
A split further from the configured one than chance allows, usually because something removed or added users in one group, such as a crash before logging, a redirect or bot filtering. Even 50.2 % to 49.8 % across 1.6 million users fails: stop, find the cause, fix it and run the test again.
From the lesson Reading A/B test results
-
How much does checking a test every day inflate false wins?
In the lesson's 2,000 simulated tests with no real effect, one look at day 14 gave 5.0 % false results; a look every day, stopping as soon as p fell below 0.05, gave 21.8 %.
From the lesson Reading A/B test results
-
Why do many metrics and segments chosen afterwards produce false wins?
Each extra look is another chance of a fluke: with 20 independent metrics, the chance that at least one looks significant is 64.2 %. Fix the length, the primary metric and any segments before the start.
From the lesson Reading A/B test results
-
Is a statistically significant lift always worth shipping?
No. Significance only says the lift is probably not zero; it says nothing about whether the lift repays its cost. Set the smallest lift worth shipping before the test and hold the interval against it: +0.05 to +1.75 against a 1-point bar reaches both sides, so ship only if the change is cheap to keep.
From the lesson Reading A/B test results
-
When two metrics move in opposite directions, what decides the call?
The product's goal names the metric that decides, and guardrail limits written before the test say how much harm a win may cost, so the call is settled before anyone has a favourite result.
From the lesson Metric trade-offs and launch decisions
-
What three-way rule turns a readout into a launch call?
Stop if the primary metric's interval lies below zero or a guardrail is clearly past its limit; ship if the primary interval lies above zero and every guardrail stays inside its limit; otherwise iterate.
From the lesson Metric trade-offs and launch decisions
-
Why was the streak freeze in the lesson's example an 'iterate' call rather than a ship?
It won on the primary metric, but the interval for day-28 retention, −0.88 to −0.02 points, reached past its −0.5 limit: the change might be harmful and the data could not yet tell.
From the lesson Metric trade-offs and launch decisions
-
What is a holdout group, and why keep one after a launch?
A random few users who stay on the old version after the launch, so that lasting effects, such as novelty fading or retention sliding slowly, can still be measured long after the test has ended.
From the lesson Metric trade-offs and launch decisions
-
What does a 5 % holdout of 200,000 learners buy, and what does it cost?
Its 10,000 learners can detect a change of about 1.40 points in day-28 retention, at a cost of about 210 fewer learners finishing three lessons each week. A 1 % holdout costs a fifth as much but only detects changes above 3 points.
From the lesson Metric trade-offs and launch decisions
-
How do you protect a target that can be gamed?
Pair it with a quality measure that the cheap way of raising it would make worse: tickets closed per agent with tickets reopened within 7 days, or sign-ups with accounts still active a week later.
From the lesson Metric trade-offs and launch decisions
-
What are the five parts of a written launch decision?
The decision, the evidence (intervals and guardrails), the risks, what will be watched after launch, and the result that would reverse the call, such as the holdout's retention ending up clearly above everyone else's.
From the lesson Metric trade-offs and launch decisions
-
What does an interviewer look for in an estimation question?
How you reach the number more than the number itself: a stated assumption can be followed, questioned and corrected, while a number from nowhere can only be accepted or rejected.
From the lesson Estimation questions, step by step
-
What are the five steps of the track's estimation method?
Clarify what is counted, split it into factors, give each factor a low and a high value, multiply, and check the result by a second route.
From the lesson Estimation questions, step by step
-
Why pin down the unit before estimating e-rickshaw rides?
A shared e-rickshaw carries about three riders a trip, so 216,000 vehicle trips make 648,000 passenger rides: one city, two answers three times apart. Fix the unit, the period and the area first, and keep the unit on every number you say.
From the lesson Estimation questions, step by step
-
What is the difference between a top-down and a bottom-up estimate?
Top-down starts from a known total, such as the population, and narrows it with shares and rates. Bottom-up starts from units you can count, such as vehicles or shops, and builds up from their capacity and use. Two routes that share no factor check each other.
From the lesson Estimation questions, step by step
-
What should you do when two estimate routes give different answers?
Within about two times of each other, say a round number between them. Further apart, a factor is wrong, so find it rather than averaging, for example by dividing one route's total by a factor of the other.
From the lesson Estimation questions, step by step
-
How do you give a range for an estimate, and which factor do you talk about?
Give every factor a low and a high, multiply each column, and quote the base with the range where the routes overlap: about 600,000 rides a weekday, roughly 300,000 to 900,000. Discuss the widest factor, because it moves the range most.
From the lesson Estimation questions, step by step
-
When should you round the numbers of an estimate?
Once, at the end, to one or two significant figures. Rounding 1.8 rides up to 2 halfway through already adds 11 % to the answer.
From the lesson Estimation questions, step by step
-
What must you check before using an official count as an estimation anchor?
What it counts. Subscriptions, accounts and SIM cards count connections, not people, and a figure per 100 people depends on the population base it was divided by, so say which base you used.
From the lesson Estimation questions, step by step
-
What do TAM, SAM and SOM each measure?
TAM is all the demand for the job the product does, whoever serves it today. SAM is the part the product can serve with what it has now: its cities, customers and channel. SOM is the share of the SAM it can win by a stated year, against competitors and with the sales it can afford.
From the lesson Market sizing: TAM, SAM and SOM
-
Why is 'our market is 40 million people' not a market size?
It counts people, not what the product could earn. A market size states demand in money or in units over a period, narrowed from the TAM to what the product can serve and what it can win.
From the lesson Market sizing: TAM, SAM and SOM
-
How do top-down and bottom-up routes size the same market?
Top-down starts from people or spending and narrows them with filters; bottom-up counts units on the supply side, such as salons, the visits each serves and the days it opens. In the salon example the two routes to the SAM landed 1.29 times apart, which counts as agreement.
From the lesson Market sizing: TAM, SAM and SOM
-
How should a SOM be argued?
From what the company can do, such as how many salons its sales team can sign and how many bookings each brings, never from a percentage chosen because it sounds modest.
From the lesson Market sizing: TAM, SAM and SOM
-
Which four traps inflate a market size?
Treating households as people, accounts or subscriptions as users, stating a value without its period ($1 million a month is $12 million a year), and quoting a growth rate without its time base.
From the lesson Market sizing: TAM, SAM and SOM
-
Why must you never add overlapping groups when sizing a market?
Adding 10 million working women to 25 million city dwellers counts the women who live in cities twice. Split the population into groups that do not overlap first, then add them.
From the lesson Market sizing: TAM, SAM and SOM
-
Why multiply average usage by a peak factor?
Use is uneven through the day. A million uploads a day average 11.6 a second, but with a stated peak factor of 3 the busiest hour runs at 34.7 a second, and that is the load the service must carry.
From the lesson Usage, revenue and capacity estimates
-
How does storage differ from load in a usage estimate?
Load comes and goes, but storage builds up: at 250 KB a receipt, a million uploads add 250 GB a day and about 91 TB a year, on top of every earlier year.
From the lesson Usage, revenue and capacity estimates
-
Why does the peak, not the average, decide whether a feature feels fast?
Waiting climbs steeply near capacity. With a server that finishes 40 uploads a second, an upload spends 200 ms in the service at 35 a second, eight times its 25 ms of work, against about 36 ms at 12 a second; above 40 the queue grows without limit.
From the lesson Usage, revenue and capacity estimates
-
What does Little's law say, and how do you use it?
L = λW: the people in a system at once equal the arrival rate times how long each stays, for a system in a steady state. Put both in the same unit first: 500 arrivals a minute staying 4 minutes gives 2,000 people at once.
From the lesson Usage, revenue and capacity estimates
-
How do you estimate the yearly revenue of a subscription app?
Users × share who pay × what each payer brings in over the period: 400,000 monthly users with 6 % paying $5 a month give 24,000 payers and $1,440,000 a year. Check the paying share against a public filing where one exists.
From the lesson Usage, revenue and capacity estimates
-
What is a product manager's capacity estimate for?
To judge whether a plan is feasible and what it costs, by showing how close the peak runs to what the service can finish. Sizing the servers is engineering's depth.
From the lesson Usage, revenue and capacity estimates
-
What separates a strategy answer from a list of good ideas?
Strategy is a choice with trade-offs: name the options, doing nothing included, judge them on the same criteria, decide, say what the choice gives up, and end with a cheap test. Doing the same things better is operational effectiveness, not strategy.
From the lesson Answering product strategy questions
-
What are the five forces, and what do they tell you?
Rivalry among established competitors, the power of buyers, the power of suppliers, the threat of new entrants and the threat of substitutes. Together they show how hard a market is to win, and a hard market rewards an advantage rivals cannot copy.
From the lesson Answering product strategy questions
-
Which six kinds of lasting advantage does the track list?
Network effects, switching costs, scale, brand, data and distribution. A strategy answer names the one the product really has and how it grows; for the bus-booking app it is distribution through small-town bus operators.
From the lesson Answering product strategy questions
-
What four responses can a product make to a competitor's move?
Ignore it while watching a metric that would show harm; match it by copying the feature or the price; leapfrog it with something clearly better; or differentiate on what you do that the rival cannot. First ask what was launched, for whom, and whether it threatens your goal.
From the lesson Answering product strategy questions
-
How should every strategy decision end?
With a cheap test: a signal, a threshold and a deadline that would change the decision.
From the lesson Answering product strategy questions
-
How does a question about the long term change a strategy answer?
The vision is turned into a sequence of bets, each of which makes the next one possible, instead of a single big plan.
From the lesson Answering product strategy questions
-
Which four questions come before choosing how to enter a market?
Is the market worth entering? Can the company win there with an advantage it brings? How should it enter: build, buy or partner? And how will it know early that the entry is failing? Each one can end the answer early.
From the lesson Entering a market: build, buy or partner
-
What does building your own way into a market give, and what does it cost?
Full control and a product that fits what the company already runs, but it is the slowest route, 14 months for the lesson's grocery chain, and the company pays for every lesson the market teaches it.
From the lesson Entering a market: build, buy or partner
-
What does buying a company already in the market give, and what does it cost?
Customers, people and a working product within months, but the highest cost up front, and joining two companies is a risk in itself.
From the lesson Entering a market: build, buy or partner
-
What does partnering give, and what does it cost, when entering a market?
Usually the cheapest and fastest start, through a company that already reaches the market, in exchange for a share of the revenue; the partner keeps part of the margin and part of the customer relationship.
From the lesson Entering a market: build, buy or partner
-
Why must you state the weights when ranking ways to enter a market?
They decide the ranking: for the grocery chain, partnering led with 8.7 when speed counted most, while building led by 7.5 to 7.1 when control counted most. Tie the weights to the goal and say what would flip them; a hard limit, such as going live within four months, comes before any score.
From the lesson Entering a market: build, buy or partner
-
What follows the choice of an entry mode?
A go-to-market plan, saying who buys first, through which channel, at what price and in what order, and a stop signal: a leading metric with a threshold and a date, agreed before launch, that stops or changes the plan when it is missed.
From the lesson Entering a market: build, buy or partner
-
What is the difference between monetisation and pricing?
Monetisation is how a product turns use into revenue: who pays, for what and when. Pricing comes after it: the amount, and the unit it is charged in.
From the lesson Pricing and monetisation
-
Which six monetisation models does the lesson compare?
A one-time purchase, a subscription, usage-based charging, freemium (most people use a tier at no charge and some pay for more), advertising, and a take rate on each sale a marketplace makes possible.
From the lesson Pricing and monetisation
-
Why must a pricing answer state its demand assumption?
Revenue is price times the payers who stay, so the assumed demand decides whether a rise earns or costs. For the photo-editing app, going from $5 to $6 adds $3,000 a month if each dollar loses 3,000 payers, but costs $9,000 if it loses 5,000.
From the lesson Pricing and monetisation
-
Where does revenue peak on a straight demand line?
Where the price elasticity of demand equals −1. Below that price, a 1 % rise loses under 1 % of payers, so revenue grows; above it, revenue shrinks. On the photo app's base line the peak is $6, and it moves from $4.60 to $7.75 between a steeper and a flatter line.
From the lesson Pricing and monetisation
-
How do you estimate LTV from the margin and churn?
LTV is roughly the monthly contribution margin divided by the monthly churn: ($6 − $1.20) ÷ 4 % = $120. Set it against CAC, what winning one payer costs, $60 in the example.
From the lesson Pricing and monetisation
-
Why is payback not simply CAC divided by the monthly margin?
Some payers cancel before their cost is covered, so count only the margin of those still paying. $60 ÷ $4.80 suggests 12.5 months, but at 4 % churn payback comes in month 17, and at 8 % LTV merely equals CAC, so a payer never pays back.
From the lesson Pricing and monetisation
-
What does honest pricing look like?
The whole price before anyone pays, a reminder before a trial turns into a charge, and cancelling as easy as starting, so that someone who did not mean to pay sees the charge coming and can stop it in one tap.
From the lesson Pricing and monetisation
-
What makes a platform two-sided?
It connects groups that need each other, such as riders and drivers or households and repairers, and has to get both sides on board while earning from the whole. Deciding which side pays, and how much, is part of its strategy.
From the lesson Platforms, marketplaces and network effects
-
What are direct and indirect network effects?
Direct: users gain from more users like them, as with a messaging service. Indirect: users on one side gain from more users on the other, as in marketplaces and app stores, which run on the indirect kind.
From the lesson Platforms, marketplaces and network effects
-
When can network effects be overcome?
When users can switch easily or use several rival services at the same time, called multi-homing. Platforms also usually need a critical mass to survive their launch.
From the lesson Platforms, marketplaces and network effects
-
What are three ways to start a marketplace that has neither buyers nor sellers?
Begin with one side, usually sellers recruited and sometimes paid to wait for buyers; begin in one small area, so each side sees high density; or begin with a single-player use, a tool worth using before the other side arrives.
From the lesson Platforms, marketplaces and network effects
-
What did the home-repair simulation show about density?
In one 6 km by 6 km area with 12 requests an hour, 4 repairers matched 26 % of requests, 16 matched 83 % and 32 matched 97 %. Beyond 32 the match rate stayed put while waits kept falling.
From the lesson Platforms, marketplaces and network effects
-
Which three measures show a marketplace's liquidity?
The match rate (requests that find a seller), the time to match, and booked time (the share of sellers' time that is sold). With 64 repairers the wait fell to 5.4 minutes, but each was booked only 15 % of the evening, so many would soon leave.
From the lesson Platforms, marketplaces and network effects
-
What limits a marketplace's take rate?
Sellers who keep too little leave, and a buyer and a seller who have met can arrange the next job directly, going around the platform, which is called disintermediation.
From the lesson Platforms, marketplaces and network effects
-
How does an execution question differ from a product sense question?
A product sense question asks you to choose the users, the problem and the solution. An execution question takes the product and its direction as given and asks how you would get results from them.
From the lesson Execution questions and goal setting
-
What are the five families of execution questions?
Goals, priorities, deadlines and launches, problems as they come (a key metric falling overnight, for instance) and working with others (design and engineering disagreeing, for instance).
From the lesson Execution questions and goal setting
-
What is an OKR?
An objective, which says where the team is going and may feel uncomfortably ambitious, with around three key results, each a number with a start, a target and a deadline. Key results describe outcomes, not activities.
From the lesson Execution questions and goal setting
-
How do you tell a key result from an activity?
Ask whether the team could finish the work while the number stays where it was. 'Launch filters' is an activity; 'raise searches that end in an order from 18 % to 22 %' is a key result, because it moves only when customers behave differently.
From the lesson Execution questions and goal setting
-
How is a key result graded, and how is its objective graded?
From 0.0 to 1.0, here by the share of the distance covered from start to target: 18 % to 20.8 % against a target of 22 % scores 0.70. The objective's grade is roughly the average of its key results, 0.73 for 0.70, 0.50 and 1.00.
From the lesson Execution questions and goal setting
-
What does it mean if a team scores 1.0 on its OKRs every quarter?
Its targets were safe. Grades of around 0.6 to 0.7 suggest ambitious targets mostly reached. Grades describe goals, not people: if they set pay, every team would choose targets it was sure to hit.
From the lesson Execution questions and goal setting
-
How does an objective become a first plan?
Each piece of work becomes a bet tied to the key result it should move, with one owner, a date, a risk with its early sign, and a number checked each week against a straight line from start to target. A number still flat in week 6 of 13 is off track: change the bet and keep the target.
From the lesson Execution questions and goal setting
-
How is a RICE score calculated, and what does it measure?
Reach × impact × confidence ÷ effort: reach in people or events a period, impact from 3 (massive) down to 0.25 (minimal), confidence as 100, 80 or 50 %, and effort in person-months. It measures total impact per unit of work, to compare many independent ideas.
From the lesson Prioritisation frameworks compared
-
When should you overrule a RICE ranking?
When something the score cannot see matters, such as another idea depending on this one. Overrule it openly and say why: the score is a guide, not a hard rule.
From the lesson Prioritisation frameworks compared
-
What does each MoSCoW category mean?
Must Have: without it you would cancel. Should Have: important but not vital. Could Have: matters less and is the first to go. Won't Have: agreed to be out this time. Keep Must Have work to about 60 % of the time box.
From the lesson Prioritisation frameworks compared
-
What are Kano's three main kinds of quality?
Must-be: missing, they cause dissatisfaction; present, they earn no praise (a booked court that is empty on arrival). One-dimensional: they please in proportion to how well they are done (fewer taps to book). Attractive: they delight when present and disappoint nobody when absent (a rain alert that moves the booking indoors).
From the lesson Prioritisation frameworks compared
-
How does WSJF rank jobs?
By cost of delay (value, plus time criticality, plus risk reduction or opportunity, each scored against the other jobs) divided by its duration, the job's size. Tournament sign-up, 23 ÷ 5 = 4.6, goes before coach booking, 20 ÷ 8 = 2.5, although coach booking is worth more.
From the lesson Prioritisation frameworks compared
-
Which prioritisation framework fits which decision?
Many ideas with some data: RICE. One release with a fixed date: MoSCoW. Learning what users take for granted: Kano. Value that drops the longer you wait: WSJF. Real decisions often pair one that sets what must be there with one that orders the rest.
From the lesson Prioritisation frameworks compared
-
Which four levers can move when a launch slips?
Cut scope and ship less on the date, move the date and ship everything later, add people, or lower the quality bar. For a fixed date, scope is the first lever to pull.
From the lesson Deadlines, scope cuts and launch readiness
-
Why does adding people rarely rescue a late software project?
Newcomers must learn the work before they can help, and every extra person adds communication, so people added late tend to make the project later still.
From the lesson Deadlines, scope cuts and launch readiness
-
In what order do you cut scope with MoSCoW?
Could Haves go first, then Should Haves, while the Must Haves, the minimum usable subset, stay. Give every item you cut a place in the next release.
From the lesson Deadlines, scope cuts and launch readiness
-
In the group-ordering example, why cut the host's spending limit as well as both Could Haves?
Dropping the two Could Haves left 20 engineer-weeks, exactly the capacity, with nothing spare for a surprise. Cutting the spending limit too left 16: four weeks of work with one week in hand.
From the lesson Deadlines, scope cuts and launch readiness
-
What turns launch readiness from a debate into a check?
Launch criteria written before anyone has a favourite answer, each with a limit and the rollout stage it must hold by, such as at least 99.5 % crash-free sessions before the 1 % stage.
From the lesson Deadlines, scope cuts and launch readiness
-
Why ship a feature behind a feature flag and roll it out in stages?
The code ships switched off, so turning the feature on is a separate decision that can be undone in seconds. The rollout grows from about 1 % of users to everyone, with a rollback trigger agreed before day 1, so switching it off needs no meeting.
From the lesson Deadlines, scope cuts and launch readiness
-
Which question does each partner own: engineering, design, data and the product manager?
The engineering lead owns how to build it and how long it takes; the designer, how it works for users; the analyst, what the data shows; and the product manager, why and what comes first. None of them reports to the product manager.
From the lesson Working with engineering, design and data
-
What is 'unity of effort', as opposed to 'unity of command'?
Bringing people who do not report to you towards one goal, instead of giving orders that are followed. It is how a product manager leads a designer, an analyst and engineers.
From the lesson Working with engineering, design and data
-
What five moves settle a disagreement between design and engineering?
Restate each view until its owner calls it fair, name the shared goal, agree which evidence would decide and by when, decide, then commit, those who disagreed included. Agreeing the evidence before anyone looks stops either side rejecting it later.
From the lesson Working with engineering, design and data
-
What do the four RACI roles mean?
Responsible people do the work, one accountable person decides, consulted people are asked before the decision, and informed people are told after it. Every decision has exactly one accountable person.
From the lesson Working with engineering, design and data
-
What goes into one line of a decision log?
What was decided, why, who decided, and the result that would reopen the decision.
From the lesson Working with engineering, design and data
-
What four parts make a no that keeps the trust of the person who asked?
The goal you share with them, the trade (what their request would push out), what would change the answer, and the smallest yes you can give right now.
From the lesson Working with engineering, design and data
-
What does an HTTP request from an app contain?
A method, such as GET, which asks for data; a path naming what is asked for; and headers, such as the user's session token. The reply opens with a status code.
From the lesson Technical questions for product managers
-
What do 2xx, 4xx and 5xx HTTP status codes tell an app?
2xx means success. 4xx means a problem with the request, so the app must change something, such as logging in again. 5xx means the server failed, and the same request may work later.
From the lesson Technical questions for product managers
-
Why is 'the app is slow' really a question about which stop?
A request travels from the phone through the network and the API server to the database and back, and every stop adds to the wait, so the fix depends on where the time goes.
From the lesson Technical questions for product managers
-
Why should a product manager be able to read an API description?
An API is the set of requests one program accepts from another, with the answers it promises, so its description shows what any app, website or partner can build. Even its error responses hold product rules, such as two patients competing for one slot.
From the lesson Technical questions for product managers
-
Why ask for p95 or p99 latency rather than the average?
An average is pulled up by a few slow requests yet describes almost none of them. In the lesson's 1,000 simulated requests the average was 191 ms and the median 114 ms, but p99 was 2,338 ms: 38 requests took over a second.
From the lesson Technical questions for product managers
-
If 38 of 1,000 requests take over a second, how likely is someone who opens the screen 10 times to meet one?
About 32 %: the chance of avoiding a slow request ten times running is 0.962 to the power of 10, roughly 0.68.
From the lesson Technical questions for product managers
-
What three parts make a plain explanation of a technical issue?
A picture the listener can imagine, the trade-off with its numbers, and the decision you need from them. The picture is optional; the trade-off and the decision are not, because they are what the listener acts on.
From the lesson Technical questions for product managers
-
Which four requirements should a product manager write as numbers before a design?
Scale (how many users and requests, and when they peak), latency (such as 99 % of reads within 200 ms), freshness (how soon new data must appear) and reliability (what share of requests must succeed). Each line rules options out.
From the lesson System design for product managers
-
What is an SLO?
A service level objective: a target for a measured indicator, for example 99 % of requests finishing in under 100 ms.
From the lesson System design for product managers
-
Which common parts make up most product features, and which product question does each answer?
An app, an API, a service that owns the feature's data, a database, a cache and a queue. The database says where the data lives, the cache what may be old, and the queue what may be late.
From the lesson System design for product managers
-
In the recipe-reviews example, what shapes the design: reads or writes?
Reads: 20,000 a second at the dinner peak against a database that serves 5,000, while writes are only about 1.7 reviews a second. Photos, not text, decide the storage.
From the lesson System design for product managers
-
Why does a cache that restarts empty endanger the database?
With 20,000 reads a second and a database limit of 5,000, at least 75 % of reads must hit the cache. An empty cache starts near 0 %, so at the peak the database is the first part to break.
From the lesson System design for product managers
-
Which three trade-offs come up in almost every design conversation?
Fresh or fast: how old may this screen's data be? Consistent or available: when parts cannot reach each other, show the last copy or refuse? Sharp or cheap: is a sharper photo worth four times the storage?
From the lesson System design for product managers
-
When does a design have to choose between consistency and availability?
Only while a network partition is present; the rest of the time a design can keep both.
From the lesson System design for product managers
-
What is technical debt?
A shortcut taken to ship sooner, such as copying rules into three places, that makes every later change cost more until it is repaid. A little debt repaid quickly speeds a team up; left alone, it charges interest on every change.
From the lesson Technical trade-offs for product managers
-
How do you put a number on the interest that technical debt charges?
Count the extra work it causes: if each slot change takes 4 extra days and there are 9 a quarter, the debt costs 36 engineer-days a quarter, so 15 days spent merging the rules pay back in about 5 weeks. Delivery measures such as change lead time and change fail rate, incidents and the bugs users see add to the picture.
From the lesson Technical trade-offs for product managers
-
How do building, buying and open source differ for a part that does not set your product apart?
Build: your engineers write and run it; it fits exactly and you pay for it every year it runs. Buy: a vendor runs it for a fee; it arrives fast, but its price, roadmap and data handling are theirs. Open source: no fee for the code under its licence, but your engineers run, update and patch it.
From the lesson Technical trade-offs for product managers
-
What should be checked on every component you take in, commercial or open source?
Its security, for as long as you use it. Reusing well-secured software lowers costs and avoids introducing new flaws.
From the lesson Technical trade-offs for product managers
-
How do you compare the cost of two AI options for a feature?
Price each per request and at its volume, then set the cost beside the wait and the quality. Tagging 3 million messages a month cost 6,075 dollars with option A and 297 with option B, which also had a p95 wait of 602 ms against 1,895 ms.
From the lesson Technical trade-offs for product managers
-
How is an AI feature's quality judged before and after launch?
Before launch, on examples whose right answer is known, such as how often each option picks the right team; after launch, by testing and monitoring that continue for as long as it runs.
From the lesson Technical trade-offs for product managers
-
Why do behavioural rounds ask what you did rather than what you would do?
Past behaviour is taken as evidence of future behaviour, and a real story can be probed: which fields went, who decided, how the result was measured. A habit ('I always…') or a plan for a case that never happened gives nothing to check.
From the lesson Behavioural rounds for product managers
-
What does the STAR method stand for?
Situation, task or goal, actions (the steps you took and the part you played) and result. The strong story in the lesson adds a fifth part, the lesson: what you now do differently.
From the lesson Behavioural rounds for product managers
-
Which seven story types cover most product manager behavioural questions?
A product you shipped, a hard prioritisation call, a metric you moved, a bet that failed, a disagreement with engineering or design, influencing without authority, and a decision taken with incomplete data.
From the lesson Behavioural rounds for product managers
-
How many stories cover the seven story types twice?
About eight, because one story often answers two types. A type with fewer than two stories marks the next story to find.
From the lesson Behavioural rounds for product managers
-
Which four faults make 'We prioritised the most important one and it went well' a weak story?
'I always' describes a habit, 'we picked' hides whose decision it was, 'it went well' has no number, and nothing was learned.
From the lesson Behavioural rounds for product managers
-
What does a scorable behavioural story contain?
One real situation, your own decision and why, a result against a starting number, such as timeouts falling from 14 % to 3 % in six weeks, and what you now do differently, with most of the words spent on your actions.
From the lesson Behavioural rounds for product managers
-
What do 'Tell me about yourself', 'Why product management?' and 'Why this product?' ask for?
Your path to this role; what you have already done that a product manager does; and what you know about the product's users.
From the lesson Behavioural rounds for product managers
-
How does a deep-dive round differ from a design round?
A design round asks you to invent something. A deep dive stays on one real product you know from the inside and asks about choices already made and results already known, so each answer can be tested against the next question.
From the lesson The product deep-dive round
-
What are the five parts of a ten-minute deep-dive walkthrough?
Context and goal, users and problem, decisions and the options you rejected, results, and what you would change.
From the lesson The product deep-dive round
-
Which part of a deep-dive walkthrough gets the most time?
The decisions and the options you turned down, about minutes 3 to 6 of 10. For the invoicing app that meant a payment link on every invoice rather than a lending product, and reminders sent by text message.
From the lesson The product deep-dive round
-
How are results stated in the invoicing-app walkthrough?
Against the goal set at the start: after six months wholesalers were paid in 29 days instead of 41, inside the target of under 30, and 38 % of invoices were paid by link.
From the lesson The product deep-dive round
-
Why should a deep-dive walkthrough finish on time?
It is only the first phase of the round. Ending at about ten minutes leaves the interviewer the choice of what to probe, and the probing decides the round.
From the lesson The product deep-dive round
-
What is a take-home assignment in a product manager process?
Written or analysed work done alone, within a stated time and on your own time, and read without you in the room, such as a writing assessment or a small project before the interviews.
From the lesson Take-home assignments for product managers
-
Which five kinds of take-home do product managers get?
A product spec or one-pager, a case-study deck of about ten slides, a data analysis with a short memo, a prioritisation exercise, and a written narrative of two to six pages.
From the lesson Take-home assignments for product managers
-
What does a reader look for first in any take-home?
The decision, in the take-home's own form: the recommendation on a deck's first slide, the problem, user and decision of a spec, the answer the data gives in an analysis, and the criteria stated before a ranking.
From the lesson Take-home assignments for product managers
-
Which phases of a time-boxed take-home are never squeezed?
The last two, cutting and checking: even a three-hour brief keeps 30 minutes for them. Write each assumption down with its reason in the second phase, since a reader can judge only what is on the page.
From the lesson Take-home assignments for product managers
-
What should you do about AI tools before starting a take-home?
Read the rules before you begin, because they differ by employer and by stage, and the brief's own rule comes first; if it is unclear, ask the recruiter in writing. Where the rules forbid it, never hand in a tool's output as if you wrote it.
From the lesson Take-home assignments for product managers
-
What does a product spec say, and what does its reader want first?
It describes the change before anything is built: the problem and who has it, the scope it covers and the scope it leaves out, and the measure that will show it worked. Readers look for the decision first, so a one-sentence summary of it opens the page.
From the lesson Writing a product spec for an interview
-
What are the nine parts of the track's one-page spec?
A one-sentence summary followed by eight sections: the problem and its users, goals and non-goals, requirements and metrics, then the risks and the questions still open. It fits in about 450 words, and every number carries its source.
From the lesson Writing a product spec for an interview
-
What is a non-goal in a product spec, and why include one?
A thing the change will deliberately not do although a reader might assume it would, such as leaving fines and loan periods untouched in a library-reminder feature. Together with open questions, non-goals show the writer's judgement most clearly.
From the lesson Writing a product spec for an interview
-
What makes a good open question in a spec?
Something not yet decided, together with who will decide it, for example whether the catalogue system can renew a loan on request, owned by engineering.
From the lesson Writing a product spec for an interview
-
How does the library-app spec state its metric?
With a start, a target and a period, plus a guardrail: late returns down from 20 % to 12 % of loans within a quarter, while watching how many members opt out of reminders.
From the lesson Writing a product spec for an interview
-
What is a PR/FAQ?
A document written as though the product were already out: it opens with a press release and follows it with answers to the questions people would ask. Some employers request it in place of a spec.
From the lesson Writing a product spec for an interview
-
Which five steps take a data take-home from the file to the memo?
Check, clean, analyse, conclude and write. The first two decide whether anything after them can be trusted, so they come before any rate is computed.
From the lesson Solving a data take-home exercise
-
Which four faults should you count in a take-home's data before computing any rate?
Duplicates (keep the first copy of each id), missing values (count them, then say whether you dropped or kept them), time zones (convert them before you cut the period) and test accounts (remove them by the marker their ids carry).
From the lesson Solving a data take-home exercise
-
How should a data take-home report the rows it removed?
Count each fault before removing anything, then give every removal one line: what, how many rows and why. In the lesson's ride-booking file, 25 of the 300 sessions had to go.
From the lesson Solving a data take-home exercise
-
Why do staff test accounts matter in a data take-home?
They behave like no real user. In the lesson's file every staff test ride ended in a booking, so they would have flattered the new screen in the one city where they were logged.
From the lesson Solving a data take-home exercise
-
What is Simpson's paradox, and why does it matter in a data take-home?
A gap that points one way inside every segment points the other way in the combined total, because the groups compared are made up of different mixes of segments. Look within segments before you trust the overall figure.
From the lesson Solving a data take-home exercise
-
In what order should a data take-home's memo go?
The recommendation first, then the evidence and its limits: what the numbers show, how sure they can be, and what they cannot show.
From the lesson Solving a data take-home exercise
-
Which three things does a panel score in a case presentation?
The recommendation (one decision, for whom, with its number, on the first slide), the reasoning (the evidence and the options weighed, with criteria said aloud) and how you handle questions.
From the lesson Presenting a case and handling questions
-
In what order should a case presentation run?
The recommendation first, then the problem and its evidence, the options weighed, the plan with its metrics, and last the risks.
From the lesson Presenting a case and handling questions
-
How many slides suit a 20-minute case talk, and why keep two minutes spare?
About ten, one every two minutes, with two time checks on the way. The spare two minutes absorb an interruption or a slow start instead of eating into the risks.
From the lesson Presenting a case and handling questions
-
How do you answer a challenge during a case presentation?
Sort it first. A fact is answered with its source and its limit; an assumption with the evidence for it and the result that would show it wrong; a judgement with the criteria you weighed and what the choice gives up.
From the lesson Presenting a case and handling questions
-
Why restate a panel's question before you answer it?
So that you answer the worry behind it rather than only its words. Then close by naming what would change your mind.
From the lesson Presenting a case and handling questions
-
What does a full mock loop test that single mocks cannot?
Whether your fifth answer is as clear as your first, and whether one weak round spills into the next. It runs five 45-minute rounds of different kinds in one day, with short breaks and a long one for a meal.
From the lesson A full mock product manager interview loop
-
What four things do you set up for a mock loop?
Interviewers (two or three people, each reading only their own prompts the night before); one unseen prompt per round with three follow-up cards; a room with no notes or phone, with audio recorded by everyone's agreement; and a clock that the interviewer keeps.
From the lesson A full mock product manager interview loop
-
How is each round of a mock loop scored?
Each round scores only the signals it reads, on the rubric's scale where 1 means missing and 4 means beyond what the role's level expects. A behavioural round, for instance, scores leadership, execution, judgement and communication.
From the lesson A full mock product manager interview loop
-
Why score a mock loop from the recording the next day?
Scored from memory, an answer gets the score you meant rather than the one you said. Quote the sentence that earned each score.
From the lesson A full mock product manager interview loop
-
Where does the long break go in a mock loop's day?
Before the last two rounds, the ones that test stamina, with the rounds in the order of a typical loop.
From the lesson A full mock product manager interview loop