# Product Manager Interview Preparation: [Your name, target companies and recruiting cycle] > A preparation workbook for product manager (PM) interviews, written for MBA candidates recruiting for product > management internships and full-time roles, including associate PM and senior PM tracks at technology companies, > product roles in consumer and industrial firms, and AI product roles. It covers the seven question families that > PM loops use: product sense, execution and metrics, estimation, strategy, behavioural, technical fluency and AI > products. Each family has an approach described in original words, an original practice prompt with a model answer > outline, and, where numbers matter, computed metrics (funnel conversion, cohort retention, experiment sample size, > AI quality and cost). Named frameworks from other authors are mentioned by name only. Source basis: Rodden, > Hutchinson and Fu (2010) for user-centred metrics, Kohavi, Tang and Xu (2020) for online experiments, Gupta, Lehmann > and Stuart (2004) for customer value and Reichheld (2003) for recommendation-based loyalty, with fictional examples > (Northfield Outdoor Co. and invented products). Replace every [bracketed] value and every row marked *Example*. This > is a learning aid, not career, legal or investment advice. ## Document control | Field | Value | |---|---| | Candidate | [Your name] | | Target companies and teams | [Company A, consumer app / Company B, enterprise platform / Company C, AI assistant] | | Role and level | [PM intern / associate PM / PM / senior PM] | | Interview stage | [Recruiter screen / first round / onsite loop] | | Interview dates | [YYYY-MM-DD] | | Prior product exposure | [None / adjacent (consulting, engineering, design) / direct] | | Reviewed by | [Second-year student, PM mentor or club lead] | | Version | [0.1] | | Last updated | YYYY-MM-DD | ## Contents - [1. How PM interviews work](#1-how-pm-interviews-work) - [2. Product sense](#2-product-sense) - [3. Execution and metrics](#3-execution-and-metrics) - [4. Experiments](#4-experiments) - [5. Estimation](#5-estimation) - [6. Strategy](#6-strategy) - [7. Behavioural](#7-behavioural) - [8. Technical fluency](#8-technical-fluency) - [9. AI products](#9-ai-products) - [10. Practice prompts](#10-practice-prompts) - [11. Preparation plan and mock log](#11-preparation-plan-and-mock-log) - [12. Common mistakes](#12-common-mistakes) - [13. My notes and next steps](#13-my-notes-and-next-steps) - [Sources and further reading](#sources-and-further-reading) - [Answer key](#answer-key) ## 1. How PM interviews work A PM loop usually has four to six interviews after a recruiter screen. The names vary by company, so ask the recruiter and recent interns for the current format, and check each company's careers pages. The families below cover what almost every loop tests. | Family | What it tests | Typical prompt shape | Time to practise | |---|---|---|---| | Product sense | Empathy for users, creativity, prioritisation, taste | "Design a product for..." / "Improve product X" | 25 to 35 minutes | | Execution and metrics | Goal setting, metric choice, diagnosis, trade-offs | "How would you measure success of..." / "Metric X fell 10%, why?" | 15 to 20 minutes | | Estimation | Structured reasoning with numbers | "How many... per year?" | 10 to 15 minutes | | Strategy | Market, competition, business model, sequencing | "Should company X enter market Y?" | 20 to 25 minutes | | Behavioural | Leadership, conflict, influence without authority, failure | "Tell me about a time..." | 3 to 4 minutes per story | | Technical fluency | How software, data and AI systems work, trade-offs with engineers | "Explain how... works" / "Design the system for..." | 15 to 20 minutes | | AI products | Model capabilities and limits, evaluation, cost, risk | "Build an AI feature for..." / "How do you know it works?" | 20 to 25 minutes | MBA candidates are often assessed against people with engineering or product backgrounds. The advantage you bring is business judgement, structure and leadership; the gap interviewers probe is technical fluency and hands-on product work. Close the gap with a product teardown, a small side project or a course, and say what you learned from it. ```mermaid flowchart LR A["Prompt"] --> B["Clarify: user, goal, constraints, scope"] B --> C["Structure: state the approach in one sentence"] C --> D["Generate: users, problems, options or metrics"] D --> E["Prioritise with explicit criteria"] E --> F["Go deep on one choice"] F --> G["Risks, metrics and next steps"] G --> H["Summarise in 30 seconds"] ``` ## 2. Product sense ### 2.1 The approach Product sense prompts ask you to design a new product or improve an existing one. A strong answer moves from a mission to a user segment, from the segment to a real problem, from problems to solutions, and from solutions to a choice with reasons. Several published checklists describe this sequence. The best known is the CIRCLES Method, a named framework by Lewis C. Lin (from his book *Decode and Conquer*, first published in 2013); learn the sequence in your own words rather than reciting any acronym, because interviewers listen for thinking, not labels. A sequence in plain words: 1. **Clarify the goal.** Why would the company build this? Engagement, revenue, a new segment, a strategic defence? Ask one or two questions, then state your assumption. 2. **Choose a user segment.** List three to four segments, then pick one with a reason (size, pain, fit with the company's strengths). 3. **Find the problems.** Describe the user's job and where it goes wrong. The jobs-to-be-done view (Christensen, Hall, Dillon and Duncan 2016) helps: what progress is the user trying to make, in what circumstance? 4. **Prioritise problems.** Use explicit criteria such as frequency, severity and how well the company can solve it. 5. **Generate solutions.** Three ideas of different ambition: a quick fix, a solid feature, a bold bet. 6. **Choose and describe one.** Walk through the user's experience step by step. Say what you would build first. 7. **Define success and risks.** One primary metric, two guardrail metrics, the main risk and how to test it. The Kano model (Kano and colleagues, 1984) is useful when you prioritise features: some features are expected (their absence causes dissatisfaction), some scale with satisfaction, and some delight because users did not expect them. ### 2.2 Prioritisation worksheet | Problem *(Example)* | Frequency (1 to 5) | Severity (1 to 5) | Company fit (1 to 5) | Score (product) | |---|---|---|---|---| | First-time hikers do not know which gear to rent | 4 | 4 | 5 | 80 | | Gear pick-up times clash with work hours | 3 | 3 | 3 | 27 | | Users cannot tell whether rented gear fits before the trip | 3 | 4 | 4 | 48 | | [Your problem] | [ ] | [ ] | [ ] | [ ] | Multiplying scores makes a weak dimension count heavily; adding them is gentler. State which you use and why. ## 3. Execution and metrics ### 3.1 Choosing metrics Start from the product's goal and work down a metric tree: a primary (north star) metric that reflects value delivered to users, input metrics the team can move, and guardrail metrics that must not get worse. The HEART framework (Rodden, Hutchinson and Fu 2010) groups user-centred metrics into Happiness, Engagement, Adoption, Retention and Task success, and pairs each with goals, signals and metrics. The AARRR "pirate metrics" funnel named by Dave McClure (acquisition, activation, retention, referral, revenue) is another common checklist. | Level | *Example:* Northfield Outdoor Co. trail-gear rental in the app | Why | |---|---|---| | Goal | Help people start outdoor activities without buying gear | Mission-level | | Primary metric | Completed rentals per month | Value delivered; grows with users and frequency | | Input metrics | Booking start rate, booking completion rate, 90-day repeat rate | Levers the team owns | | Guardrails | Damage claims per rental, customer support contacts per rental, return rate of new-gear purchases | Growth must not raise cost or cannibalise sales | ### 3.2 Funnel: computed example *Example*: one month of the rental flow. | Stage | Users | Step conversion | Overall conversion | Excel formula (step) | |---|---|---|---|---| | Visited rental page | 200,000 | | 100.00% | | | Started booking | 26,000 | 13.0% | 13.00% | `=26000/200000` | | Completed booking | 9,100 | 35.0% | 4.55% | `=9100/26000` | | Picked up gear | 8,190 | 90.0% | 4.10% | `=8190/9100` | | Booked again within 90 days | 2,457 | 30.0% | 1.23% | `=2457/8190` | Reading it: the weakest step is start to completion (35%), so it is the first place to look. If completion rose from 35% to 40% with the other steps unchanged, completed bookings would rise from 9,100 to 10,400 (`=26000*40%`) and repeat bookings from 2,457 to 2,808 (`=10400*90%*30%`). ### 3.3 Retention: computed example *Example*: weekly active users as a share of a sign-up cohort of 10,000, before and after an onboarding change that recommends a starter gear kit. | Week | Cohort A (old onboarding) active | Retention A | Cohort B (new onboarding) active | Retention B | |---|---|---|---|---| | 0 | 10,000 | 100.0% | 10,000 | 100.0% | | 1 | 4,200 | 42.0% | 4,800 | 48.0% | | 2 | 3,300 | 33.0% | 4,000 | 40.0% | | 4 | 2,600 | 26.0% | 3,400 | 34.0% | | 8 | 2,150 | 21.5% | 3,050 | 30.5% | | 12 | 1,980 | 19.8% | 2,950 | 29.5% | Retention for each cell is active users divided by the cohort size (`=4200/10000`). Two things matter: the level at week 1 (activation) and whether the curve flattens (a stable core of users who found lasting value). Cohort B starts higher and flattens near 30% instead of 20%, which suggests the change improved activation and lasting value, not just a first-week spike. Before you claim it, check that the cohorts are comparable (season, acquisition channel) or, better, test it in an experiment (section 4). ```vega-lite { "$schema": "https://vega.github.io/schema/vega-lite/v5.json", "description": "Weekly retention of two fictional sign-up cohorts", "width": 420, "height": 220, "data": { "values": [ {"week": 0, "cohort": "A (old onboarding)", "retention": 100.0}, {"week": 1, "cohort": "A (old onboarding)", "retention": 42.0}, {"week": 2, "cohort": "A (old onboarding)", "retention": 33.0}, {"week": 4, "cohort": "A (old onboarding)", "retention": 26.0}, {"week": 8, "cohort": "A (old onboarding)", "retention": 21.5}, {"week": 12, "cohort": "A (old onboarding)", "retention": 19.8}, {"week": 0, "cohort": "B (new onboarding)", "retention": 100.0}, {"week": 1, "cohort": "B (new onboarding)", "retention": 48.0}, {"week": 2, "cohort": "B (new onboarding)", "retention": 40.0}, {"week": 4, "cohort": "B (new onboarding)", "retention": 34.0}, {"week": 8, "cohort": "B (new onboarding)", "retention": 30.5}, {"week": 12, "cohort": "B (new onboarding)", "retention": 29.5} ] }, "mark": {"type": "line", "point": true}, "encoding": { "x": {"field": "week", "type": "quantitative", "title": "Weeks since sign-up"}, "y": {"field": "retention", "type": "quantitative", "title": "Weekly active (% of cohort)"}, "color": {"field": "cohort", "type": "nominal", "title": "Cohort"} } } ``` ### 3.4 Other metrics to compute quickly | Metric | Formula | *Example* | Result | Excel formula | |---|---|---|---|---| | Stickiness | Daily active users / monthly active users | 46,000 / 184,000 | 25.0% | `=46000/184000` | | Simple customer lifetime value (monthly) | ARPU x gross margin / monthly churn | 9.00 x 70% / 5% | USD 126.00 | `=9*0.7/0.05` | The simple lifetime value formula assumes constant churn and margin and no discounting; see mba/mba-customer-lifetime-value-card for the discounted version based on Gupta, Lehmann and Stuart (2004). ### 3.5 Diagnosing a metric drop When a metric falls, structure the diagnosis before you guess: 1. **Is it real?** Check data pipelines, logging changes, definition changes and dashboards. 2. **When and how fast?** A step change suggests a release or an outage; a slow drift suggests competition or seasonality. 3. **Where?** Segment by platform, app version, country, channel, new versus returning users. 4. **Internal or external?** Releases, pricing, marketing changes versus competitors, holidays, news, regulation. 5. **Which funnel step?** Decompose the metric into its steps and find the one that moved. 6. **Act.** Roll back, fix, or run an experiment, and say how you would confirm the cause. ## 4. Experiments Online controlled experiments (A/B tests) are the standard way to estimate the causal effect of a product change. Kohavi, Tang and Xu (2020) describe the practice in detail, including trust checks such as sample ratio mismatch. The statistics are in academic/stats-proportion-tests and academic/stats-hypothesis-testing; the plan format is in product/ab-test-plan. Sample size per arm for comparing two proportions, with significance level $\alpha$ (two-sided) and power $1-\beta$: $$ n = \frac{\left(z_{1-\alpha/2}\sqrt{2\bar{p}(1-\bar{p})} + z_{1-\beta}\sqrt{p_1(1-p_1)+p_2(1-p_2)}\right)^2}{(p_2-p_1)^2} $$ | Symbol | Meaning | |---|---| | $p_1$ | Baseline conversion rate in control | | $p_2$ | Conversion rate you want to be able to detect in treatment | | $\bar{p}$ | Average of $p_1$ and $p_2$ | | $z_{1-\alpha/2}$, $z_{1-\beta}$ | Standard normal quantiles for significance and power | *Example*: test a simpler booking form. Baseline completion 35%, smallest effect worth detecting +2 percentage points (to 37%, so $\bar{p}$ = 0.36), $\alpha$ = 0.05 two-sided, power 80%. With these inputs the formula gives 9,040.7, rounded up to 9,041; a smaller effect or a lower baseline needs many more users, because $n$ scales with $1/(p_2-p_1)^2$. | Item | Result | Excel formula | |---|---|---| | $z_{0.975}$ | 1.9600 | `=NORM.S.INV(0.975)` | | $z_{0.80}$ | 0.8416 | `=NORM.S.INV(0.8)` | | Users per arm | 9,041 | `=ROUNDUP((NORM.S.INV(0.975)*SQRT(2*0.36*0.64)+NORM.S.INV(0.8)*SQRT(0.35*0.65+0.37*0.63))^2/0.02^2,0)` | | Total users | 18,082 | `=2*9041` | | Days at about 867 booking starts a day (26,000 a month) | 20.9, so run 21 days (three full weeks) | `=18082/(26000/30)` | Running whole weeks covers weekday and weekend behaviour. Decide the duration in advance and do not stop early when the result first looks significant; repeated peeking raises the false positive rate. ## 5. Estimation Estimation prompts test structure, not knowledge. Say the equation before you fill it, use round numbers, and finish with a sanity check and the most sensitive assumption. *Example*: "How many outdoor gear rentals happen in a year in a metropolitan area of 3 million people?" | Step | Assumption | Value | Excel formula | |---|---|---|---| | Households | 3,000,000 people / 2.5 per household | 1,200,000 | `=3000000/2.5` | | Households doing overnight outdoor trips | 30% | 360,000 | `=1200000*30%` | | Of those, households that rent rather than own some gear | 20% | 72,000 | `=360000*20%` | | Trips per renting household per year | 3 | | | | Share of trips with a rental | 50% | | | | Rentals per year | 72,000 x 3 x 50% | 108,000 | `=72000*3*50%` | | Revenue at an average of USD 45 per rental | | USD 4,860,000 | `=108000*45` | Sanity check: about one rental per 28 people a year, plausible for an activity that a minority does a few times a year. The most sensitive assumption is the 20% renting share: halving it halves the answer, so it is the one to research first. All shares here are illustrative assumptions, not market data. ## 6. Strategy Strategy prompts ask whether to enter a market, build or partner, respond to a competitor, or change a business model. A strong answer covers: the company's goal and strengths; the market (size, growth, segments); competition and substitutes (see mba/mba-five-forces-card); the company's right to win (see mba/mba-resource-based-view-card and mba/mba-competitive-advantage-card); the business model and economics (see mba/mba-pricing-card); risks; and a recommendation with a first step that tests the riskiest assumption cheaply. | Question | *Example:* Should Northfield Outdoor Co. launch a gear subscription? | Your case | |---|---|---| | Goal | Grow repeat revenue and reach first-time customers | [ ] | | Market | Rental demand of about 108,000 rentals a year locally (section 5 estimate) | [ ] | | Competition and substitutes | Buying, borrowing from friends, other rental shops | [ ] | | Right to win | Store network for pick-up and returns, brand trust, repair skills | [ ] | | Economics | Subscription price versus gear depreciation, cleaning and damage | [ ] | | Risks | Cannibalising gear sales, damage costs, seasonality | [ ] | | Recommendation and first test | Pilot in two stores for one season with a capped number of members | [ ] | ## 7. Behavioural Prepare six to eight stories and map each to the themes PM interviewers probe. Use a situation, task, action and result structure, and end with what you learned. Keep each answer to about two to three minutes, with numbers. | Theme | Story title | Result with a number | Learning | |---|---|---|---| | Influence without authority | *Example:* Persuaded a regional sales team to adopt a new pricing tool | Adoption from 2 to 11 of 14 branches in a quarter | Start with the users who feel the pain most | | Conflict with a partner team | [ ] | [ ] | [ ] | | Data-driven decision | [ ] | [ ] | [ ] | | Failure or a wrong call | [ ] | [ ] | [ ] | | Shipping under constraints | [ ] | [ ] | [ ] | | Customer insight that changed a plan | [ ] | [ ] | [ ] | "Why product management?" and "Why this company?" need a short, specific answer: a moment that drew you to building products, evidence that you have done product work, and a link to this company's products and users. ## 8. Technical fluency Topics to be able to explain in plain language, with one trade-off each: | Topic | Explain | Trade-off to mention | |---|---|---| | Client, server and APIs | How an app asks a server for data through an API | Latency versus freshness of data; caching | | Databases | Tables, indexes, why queries can be slow | Consistency versus availability at scale | | Releases | Feature flags, staged rollouts, rollback | Speed of shipping versus risk | | Data and analytics | Event logging, data pipeline, dashboard | Measurement accuracy versus privacy and cost | | Machine learning basics | Training, evaluation on held-out data, inference | Model quality versus latency and cost | | Technical debt | Why shortcuts slow later work | Short-term delivery versus long-term speed | A useful exercise: pick a feature in an app you use and describe what happens from the tap to the screen update, which services are involved, and what could fail. Use system-design templates in the library if you want more depth. ## 9. AI products ### 9.1 What interviewers look for - **Problem first.** Is AI the right tool? Would rules, search or a better form solve the problem more cheaply? - **Capabilities and limits.** Models can be wrong with confidence, can vary between runs and can be manipulated by inputs. Design for failure: human review, citations to sources, easy correction, clear escalation. - **Evaluation.** How you will know the feature works: an offline evaluation set with human ratings, then online metrics in an experiment (see ai-data/llm-evaluation-plan). - **Cost and latency.** Cost per request and per successful outcome, response time. - **Risk and governance.** Privacy, safety, fairness, intellectual property, and applicable law. AI rules such as the EU AI Act (Regulation (EU) 2024/1689, as amended) apply in stages; as of October 2026 check the current consolidated text on EUR-Lex before you cite a date, and see mba/mba-ai-risk-governance-card. ### 9.2 Computed example: an AI assistant for customer support *Example*: Northfield's support team receives 120,000 tickets a month. An AI assistant answers some tickets fully (deflection) and routes others to people. | Metric | Computation | Result | Excel formula | |---|---|---|---| | Answer correctness on a review sample | 172 correct of 200 reviewed | 86.0% | `=172/200` | | 95% confidence interval (normal approximation) | 86.0% +/- 1.96 x standard error | 81.2% to 90.8% | `=0.86-1.96*SQRT(0.86*0.14/200)` and `=0.86+1.96*SQRT(0.86*0.14/200)` | | Escalation precision | 90 true escalations / (90 + 30 false) | 75.0% | `=90/(90+30)` | | Escalation recall | 90 / (90 + 10 missed) | 90.0% | `=90/(90+10)` | | F1 score | 2 x 0.75 x 0.90 / (0.75 + 0.90) | 0.818 | `=2*0.75*0.9/(0.75+0.9)` | | Tickets deflected at 30% | 120,000 x 30% | 36,000 | `=120000*30%` | | Cost before, at USD 6.50 per human-handled ticket | 120,000 x 6.50 | USD 780,000 | `=120000*6.5` | | Cost after, with USD 0.40 of AI cost on every ticket | 84,000 x 6.50 + 120,000 x 0.40 | USD 594,000 | `=84000*6.5+120000*0.4` | | Monthly saving | 780,000 - 594,000 | USD 186,000 | `=780000-594000` | Interpretation: with 200 reviewed answers the correctness estimate is uncertain by almost five points either way, so a launch threshold of, say, 85% cannot yet be confirmed; review more answers or segment by ticket type. A missed escalation (recall) is usually worse than an unnecessary one (precision), because a frustrated customer with an unresolved problem costs more than a few extra minutes of agent time. The saving is only real if deflected tickets stay resolved: track the re-contact rate as a guardrail. ## 10. Practice prompts 1. **Product sense.** Design a product that helps first-time campers prepare for their first overnight trip. 2. **Product sense.** Improve the experience of returning rented equipment for Northfield Outdoor Co. 3. **Metrics.** Northfield launches a gear subscription. What is the primary metric, and which guardrails do you set? 4. **Metrics diagnosis.** Booking completion fell from 40% to 30% in a week while booking starts rose from 6,000 to 6,100 out of 50,000 visitors. How many completed bookings were lost, and how do you investigate? 5. **Estimation.** Estimate the number of electric bicycles charged at workplaces in a city of 2 million people on a weekday. 6. **Strategy.** A fictional note-taking app with 5 million users is considering adding an AI meeting summary as a paid add-on. Should it, and how should it price it? 7. **Experiment.** In the booking test in section 4, the treatment shows +3 points after four days with p = 0.04. The team wants to ship. What do you say? 8. **AI product.** The support assistant in section 9 is wrong on refund policy questions 25% of the time. What do you do? 9. **Behavioural.** Tell me about a time you changed a decision because of data. 10. **Technical.** Explain to a non-technical executive why the app's search results take two seconds to load and what could make them faster. ## 11. Preparation plan and mock log | Relative timing | Focus | Output | Done | |---|---|---|---| | T-12 weeks | Product exposure | Two product teardowns, one small side project or course | [ ] | | T-10 weeks | Product sense | Five design prompts, two improve prompts | [ ] | | T-8 weeks | Metrics and experiments | Sections 3 and 4 rebuilt, five metrics prompts | [ ] | | T-7 weeks | Estimation | Ten drills from mba/mba-market-sizing-drills | [ ] | | T-6 weeks | Strategy | Three strategy prompts with recommendations | [ ] | | T-5 weeks | Behavioural | Stories mapped in section 7 and mba/mba-fit-stories | [ ] | | T-4 weeks | Technical fluency and AI | Section 8 explanations aloud, section 9 example rebuilt | [ ] | | T-3 to T-1 weeks | Mock loops | At least six mocks with PMs or second-year students | [ ] | | Date | Interviewer | Prompt | Score (1 to 5) | What to fix | |---|---|---|---|---| | *Example:* YYYY-MM-DD | Second-year student, former PM | Improve gear returns | 3 | Jumped to solutions before choosing a user segment | | [YYYY-MM-DD] | [Name] | [Prompt] | [ ] | [ ] | ## 12. Common mistakes 1. Jumping to solutions before clarifying the goal and choosing a user segment. 2. Designing for "everyone", which produces generic features. 3. Listing frameworks by name instead of applying them; reciting an acronym is not structure. 4. Choosing vanity metrics (downloads, page views) as the primary metric instead of value delivered. 5. Forgetting guardrail metrics, so a "win" hides higher costs or worse experience elsewhere. 6. Diagnosing a metric drop by guessing causes instead of checking data quality, timing and segments first. 7. Stopping an experiment early because the result looks significant. 8. Estimation without a stated equation, or without a sanity check at the end. 9. Strategy answers without a recommendation, or a recommendation without a first test. 10. Behavioural stories with "we" throughout, no numbers and no learning. 11. Treating AI as the answer before establishing the problem, and ignoring evaluation, cost and failure handling. 12. Running out of time: no summary at the end. Keep 30 seconds to summarise. ## 13. My notes and next steps | Topic | What I still get wrong | Next action | By | |---|---|---|---| | *Example:* Metrics | Forget guardrails | Add two guardrails to every metrics answer for the next five mocks | YYYY-MM-DD | | [Topic] | [ ] | [ ] | [YYYY-MM-DD] | Products I will use as examples: [Product 1, why] / [Product 2, why] ## Sources and further reading Primary works: - Rodden, K., Hutchinson, H. and Fu, X. (2010). Measuring the user experience on a large scale: user-centered metrics for web applications. *Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI 2010)*, 2395-2398. - Kano, N., Seraku, N., Takahashi, F. and Tsuji, S. (1984). Attractive quality and must-be quality. *Journal of the Japanese Society for Quality Control (Hinshitsu)*, 14(2), 39-48. - Reichheld, F. F. (2003). The one number you need to grow. *Harvard Business Review*, 81(12), 46-54. - Gupta, S., Lehmann, D. R. and Stuart, J. A. (2004). Valuing customers. *Journal of Marketing Research*, 41(1), 7-18. - Kohavi, R., Tang, D. and Xu, Y. (2020). *Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing*. Cambridge University Press. - Christensen, C. M., Hall, T., Dillon, K. and Duncan, D. S. (2016). *Competing Against Luck: The Story of Innovation and Customer Choice*. HarperBusiness. - Regulation (EU) 2024/1689 (Artificial Intelligence Act), https://eur-lex.europa.eu (check the consolidated text). Books (further reading, by name only): Lin, *Decode and Conquer* (the CIRCLES Method); Cagan, *Inspired*; Olsen, *The Lean Product Playbook*. Open resource: OpenStax, *Introductory Statistics 2e* (CC BY-NC-SA 4.0), https://openstax.org, for the proportion tests behind section 4. ## Answer key 1. Goal: grow first-time customers. Segments: solo beginners, families, groups of friends; pick families (largest spend, highest anxiety). Problems: what to bring, how to set up, what if the weather turns. Solutions: a checklist generator (quick), a "first trip kit" rental bundle with a video guide (solid), a guided first trip with a ranger partner (bold). Choose the kit bundle; primary metric first-trip completion and 90-day repeat rate; guardrails support contacts and damage claims; risk: low margin on bundles, test with a pilot. 2. Map the return journey (travel to store, queue, inspection, deposit refund). Prioritise the queue and slow deposit refunds. Options: drop-off lockers, photo-based self-inspection, instant partial refund. Metric: return time and satisfaction; guardrail: damage disputes. 3. Primary: active subscribers who rented in the month (value delivered), or subscription revenue retained after three months. Guardrails: gear sales to subscribers (cannibalisation), damage cost per rental, support contacts. 4. Expected completions at the old rate: 40% x 6,100 = 2,440; actual 30% x 6,100 = 1,830; lost 610 (`=0.4*6100-0.3*6100`). Investigate: data check, release log for the booking form, segment by platform and app version, payment provider errors, step-level drop within the form; roll back if a release coincides. 5. Equation: commuters x share cycling x share electric x share charging at work. Example: 2,000,000 x 50% workers = 1,000,000; 5% cycle = 50,000; 30% electric = 15,000; 40% charge at work = 6,000 charges a weekday. Sanity check and name the sensitive assumption (cycling share). 6. Recommend yes if the summary solves a frequent pain for heavy meeting users. Segment by meetings per week, price as an add-on per seat with a usage cap to control model cost, measure attach rate and retention of payers, guardrails on summary accuracy and privacy complaints. First test: a paid beta for a segment. 7. Do not ship yet. The test was sized for 21 days and 18,082 users; at four days it has about 3,500 users, so early significance is likely to be noise from peeking. Run to the planned duration and check guardrails and sample ratio. 8. Segment errors by topic; route refund questions to people or ground answers in the current policy text with citations; add the topic to the evaluation set; set a threshold for re-enabling; monitor re-contact rate. 9. Use a story from section 7: situation, your decision, the data that changed it, the result with a number, and what you now do differently. 10. Plain explanation: the app sends the search to a server, which queries a large database and ranks results; delays come from the query, ranking and network. Options: indexes, caching popular searches, showing first results while the rest load. Trade-off: freshness versus speed and cost. ## Revision history | Version | Date | Author | Change summary | |---|---|---|---| | 0.1 | YYYY-MM-DD | [Your name] | First pass through sections 1 to 9 | | 0.2 | YYYY-MM-DD | [Your name] | *Example:* Added own product teardowns and mock scores | ## Standard and licence note Based on public research on product metrics and online experiments listed above, and on common practice in product manager interviews. The text, examples, numbers and practice prompts are original; frameworks by other authors, such as Lewis C. Lin's CIRCLES Method, are named only and not reproduced, no interview question bank or book is copied or paraphrased, and Northfield Outdoor Co. and every other product here are fictional. The template is independent and not affiliated with or endorsed by any company, business school or publisher. Check current rules on AI and data protection before you rely on them. Template licence: MIT.