SkillNavora
Upskill for free · Practice & test yourself

Top 100 interview questions

For AI Product Management, Data Science & AI, and GenAI & Agents, with model answers and the lesson that teaches each topic.

The 100 most important interview questions for each career path, with model answers and a link to the lesson that teaches the topic. Click a question to see the answer.

01What is your favourite product and why?
“Google Maps for commuting. My job is ‘get to work on time with least stress.’ It works because (1) live traffic turns uncertainty into an ETA I trust, (2) it re-routes proactively, (3) it combines modes. A weakness for me: on crowded metro lines it doesn't tell me how full trains are, so I can't choose a less crowded one. I'd test showing crowding levels from anonymised, aggregated data, measuring route-choice changes and satisfaction, with privacy as a guardrail.”
Learn it: Product sense →
02What makes a product great?
“A great product reliably helps a specific user make progress on an important job, with as little effort and risk as possible, in a way the business can sustain. Signals: users return without being pushed (retention curves flatten), they recommend it, and the economics improve with scale. Example: a payments flow that just works at checkout — invisible reliability is a feature.”
Learn it: Product sense →
03Critique the onboarding of an app you used recently.
“A budgeting app asked for bank linking, notifications and goals before showing anything useful. Heuristic: visibility and user control — I had no idea what value I'd get. Biggest fix: show a sample dashboard immediately and ask for bank linking in context when I tap ‘see my spending’. Metric: activation (first budget created within 3 days); guardrails: permission grant rate, privacy complaints.”
Learn it: Product sense →
04Which product do you think is badly designed, and how would you fix it?
“Many parking-payment apps: you must register, add a card and find a zone code while standing in the street. Constraint: municipalities require identification and payment records. Fix: a QR on the meter that opens a lightweight web flow with wallet payment, with optional account creation later. Trade-off: less data capture; mitigate with optional emailed receipts.”
Learn it: Product sense →
05How would you improve YouTube Shorts?
“I'd focus on new viewers whose first Shorts feel irrelevant — they leave before personalisation kicks in. Ideas: a lightweight topic picker after the first few swipes; a ‘not interested’ that visibly teaches the feed; separation of short vs long results in search. I'd A/B the topic picker; primary: day-7 return rate of new viewers; guardrails: session-satisfaction surveys and time-spent spikes among younger users.”
Learn it: Product sense →
06Should Netflix let users stop a title from leaving the catalogue?
“The need is real — losing a show mid-watch is frustrating — but licences expire contractually, so a ‘keep’ button is usually infeasible. Better: warn earlier (‘leaving in 14 days’), a ‘finish before it leaves’ row, and use the count of mid-watch members as a signal in licence negotiations.”
Learn it: Product sense →
07What product would you build for elderly users to manage medicines?
“Start with a caregiver-set schedule and a very simple ‘taken’ confirmation (large button, voice option); missed-dose alerts to a caregiver after a grace period; pharmacy refill reminders. Avoid medical advice; partner with pharmacists for dose data. Metrics: doses confirmed on time; caregiver alerts resolved. Guardrails: false alarms, privacy consent.”
Learn it: Product sense →
08What is a product you think will fail, and why?
“I'd question standalone AI gadgets that duplicate phone capabilities: the job (quick answers) is already served by phones; the device adds cost and charging burden; switching costs favour phones. It could succeed only if it serves a job phones do badly — hands-free contexts or accessibility. I'd watch day-30 retention and daily use versus phone usage.”
Learn it: Product sense →
09Design a search engine for children under 14.
“Focus on 8–12-year-olds doing homework. Needs: readable results, trustworthy sources, no unsuitable content, parents' reassurance. Design: curated educational sources first, reading-level summaries, image search with strict filtering, ‘ask a grown-up’ prompts for sensitive topics, a parent dashboard with privacy-preserving summaries. Filtering is imperfect — don't promise complete safety; add a report button; no personalised ads. Metrics: task success (found an answer); parent trust score; guardrail: unsafe-content reports.”
Learn it: Product sense →
10Help a professional regain control of an overflowing inbox after a holiday.
“Job: find what's urgent and relevant without reading everything. Design: a ‘Catch-up’ view grouped by people I work with, deadlines and mentions; AI thread summaries with links to originals; bulk actions with preview and undo. Risks: misclassifying an important email, wrongly auto-archiving, exposing confidential data to AI — keep human review for bulk actions, test accuracy, respect permissions. Metric: time to clear important items; guardrail: ‘missed important email’ reports.”
Learn it: Product sense →
11Design a feature to reduce food-delivery checkout abandonment.
“First confirm where users drop — say, the payment step. If research shows fee surprise, show an estimated total in the cart with ‘est.’ labels, refresh on address change, and an honest ‘Estimate unavailable’ fallback if the fee service fails. A/B test with checkout completion as primary; complaints and payment errors as guardrails.” *(See Modules 5 and 7.)*
Learn it: Product sense →
12Design a dating feature for a large social network.
“Segment by intention (serious vs casual). Core: opt-in profile separate from the main profile, verified identity, mutual-interest reveal, events-based matching. Safety: consent, harassment reporting, location fuzzing, age verification, no exposure of private conversations. Monetisation: freemium features that don't compromise safety (no paying to message strangers). Metric: mutual matches that lead to conversations; guardrails: reports per 1,000 users, blocks.”
Learn it: Product sense →
13Design an ATM for the visually impaired.
“Audio guidance via a headphone jack, a tactile keypad with raised markers, phone-based pre-staging (set the withdrawal in the app, then tap card/phone), screen blanking for privacy, consistent layouts and generous timeouts. Test with visually impaired users; measure task success and time; guardrail: fraud and shoulder-surfing incidents.”
Learn it: Product sense →
14Design a parking app for a busy city.
“Users: daily commuters vs occasional visitors. Jobs: find a spot quickly, pay easily, avoid fines. Features: live availability (sensors/partner garages), reserve and pay, extend remotely, reminders before expiry. Metrics: time to park, successful reservations, fine reduction (self-reported). Risks: inaccurate availability → show confidence levels.”
Learn it: Product sense →
15Design a product to help small restaurants reduce food waste.
“Job: prepare and order the right quantities. Start with a simple demand forecast from past sales + weather + local events, giving suggested prep quantities; later add a marketplace for end-of-day surplus. Low integration first (POS export/CSV). Metrics: waste % by weight, margin; guardrail: stock-outs.”
Learn it: Product sense →
16Design a feature to increase channel-creation quality in a team-chat tool.
“Suggest similar existing channels when someone types a new channel name; offer naming templates and pre-filled purposes; improve directory search. Primary metric: duplicate-channel rate; secondary: channel-message share. Randomise by workspace to avoid contamination.” *(See Case 7 and Project 3.)*
Learn it: Product sense →
17Design an emergency feature for a ride-hailing app.
“Users: rider, driver, emergency contacts, safety operations. Features: an SOS button that shares live location with emergency services/contacts per local rules, trip sharing, unusual-stop detection with a check-in prompt, and a safety centre. Design for false positives without delaying urgent calls; offline/poor-network fallback (SMS). Metrics: time to response, check-in resolution; guardrails: false alarms, privacy.”
Learn it: Product sense →
18Should a food-delivery company launch grocery delivery?
“It's product development in an existing market (Ansoff). Assets: customer base, courier network, app. Need: fast top-ups for urban households. Economics differ: lower AOV and margin, inventory and dark stores. I'd pilot in two dense neighbourhoods with a limited assortment, measure repeat rate, contribution per order and cannibalisation of food orders, and scale only if contribution turns positive within ~6 months.”
Learn it: Product strategy →
19How should a travel marketplace respond to a sudden collapse in travel (e.g., a pandemic)?
“Focus on demand that still exists — nearby getaways and long stays for remote workers; adapt search to flexible destinations and monthly stays; strengthen trust with cleaning standards and flexible cancellation; support hosts with clear refund rules. Measure booking recovery by segment, host retention and refund disputes.”
Learn it: Product strategy →
20How would you enter a new country with a fintech product?
“Scan regulation, payments infrastructure and trust (PESTLE); choose a beachhead segment; partner with a licensed local bank first rather than seeking a licence; localise language, KYC documents and payment methods; pilot in one city. Key risks: regulatory timelines and new fraud patterns. Metrics: activation, fraud rate, CAC payback.”
Learn it: Product strategy →
21A competitor launched a cheaper version of our product. What do you do?
“Don't react instantly. Check churn and win/loss by segment, interview switchers, and see whether price or features drive it. Options: hold price and sharpen value messaging; launch a lighter tier for price-sensitive users; bundle; raise switching value ethically. Choose based on margin impact and which segment matters strategically.”
Learn it: Product strategy →
22Should a music-streaming service expand into podcasts or audiobooks?
“Frame it as expanding from ‘music’ to ‘audio companionship’. Benefits: more listening time, differentiation, ad inventory. Costs: content and rights, discovery complexity. Strategy: invest in discovery and creator tools rather than only exclusives. Measure incremental listening time and retention, not catalogue size.”
Learn it: Product strategy →
23Build, buy or partner for identity verification?
“Buy or partner: identity verification is regulated and commodity-like, and speed matters. Keep control of the onboarding UX and data handling, check vendor compliance and uptime, build an abstraction layer to switch vendors, and revisit building only if scale economics or differentiation justify it.”
Learn it: Product strategy →
24What is the moat for a ride-hailing company?
“Local network effects (density cuts wait times), brand trust and safety record, data for pricing and ETAs, and driver relationships. Weaknesses: riders and drivers multi-home, and regulation. So the moat is local and must be continuously defended with reliability and driver economics.”
Learn it: Product strategy →
25How would you set a 3-year vision and strategy for a B2B SaaS product?
“Vision: the customer future (e.g., ‘every team closes the loop on decisions’). Strategy: win a beachhead, then adjacent segments; three product pillars; integrations as a platform; GTM from PLG to sales-assist; pricing evolution; an NSM with guardrails; explicit ‘won't do’ list. Validate yearly with retention, NRR and win/loss.”
Learn it: Product strategy →
26Your launch is slipping by three weeks. What do you do?
“Find out why — scope growth, a dependency or quality issues. Options: cut scope to keep the date, add people (rarely helps late), or move the date. Choose by cost of delay and quality risk. Tell stakeholders immediately with impact and the new plan; update the decision log.”
Learn it: Product strategy →
27How do you write good user stories and acceptance criteria?
“Stories describe user value: ‘As a first-time buyer, I want to see my estimated total in the cart so I can decide before paying.’ Criteria cover the happy path, edge cases, errors, accessibility and analytics events, in Given/When/Then form. I review them with engineering and QA before sprint planning.” *(Module 6.)*
Learn it: Product strategy →
28Engineering says a feature will take 3 months; leadership wants it in 1.
“Understand what drives the estimate; find the smallest valuable slice; defer non-essential risk behind a flag; present options (1-month slice, 3-month full) with trade-offs; never commit dates on engineers' behalf. If needed, escalate jointly with an options paper.”
Learn it: Product strategy →
29A production incident is causing payment failures. What is your role as PM?
“Engineering leads the fix. I quantify user and business impact, coordinate communications (support, status page, leadership), decide customer remediation (refunds, messaging), and ensure a blameless post-incident review whose prevention items get prioritised.” *(Extra topic 13: don't declare it over just because the dashboard looks better briefly.)*
Learn it: Product strategy →
30How do you decide when a feature is ready to launch?
“Acceptance criteria met; QA/UAT, performance, accessibility, security and privacy reviews done; tracking validated; support ready; rollout plan with flag and rollback owner; success and guardrail thresholds defined.”
Learn it: Product strategy →
31How do you manage dependencies across teams?
“Map dependencies with owners, dates and risks; agree interface contracts; track them on a shared board; sequence work to unblock the critical path; hold short syncs; escalate early with options when blocked.”
Learn it: Product strategy →
32What is the difference between Definition of Ready and Definition of Done?
“Ready = criteria a team may use before starting an item (clear, estimated, testable). Done = shared quality criteria for completed work (tested, reviewed, documented, deployed per team rules).” *(Lecture 22.)*
Learn it: Product strategy →
33After launch, adoption is low. What do you do?
“Check tracking; check exposure (did people even see it?); segment by user type; interview users who saw it but didn't use it; compare with the original hypothesis. Then improve onboarding or placement, iterate on value, or sunset. Low use alone isn't enough to remove a feature — persistent low value is.”
Learn it: Product strategy →
34What metrics would you track for a food-delivery app?
“North Star: completed, problem-free orders per active customer per month. Inputs: activation, order frequency, cart conversion, payment success, on-time delivery. Economics: AOV, contribution per order, CAC payback. Guardrails: refunds, complaints, cancellations.” *(Module 7.)*
Learn it: Analytics →
35Define activation for a team-collaboration product.
“An event reflecting team value that predicts retention, e.g., ‘a workspace with ≥ 3 members exchanging messages in a channel within 7 days.’ Validate by comparing retention of activated vs not, then run an experiment to see whether helping teams reach it improves retention.”
Learn it: Analytics →
36An A/B test shows +3% conversion with p = 0.04. Ship?
“Check: was this the pre-registered primary metric? Adequate power and full weeks? Any sample ratio mismatch? Guardrails flat? Does the low end of the confidence interval still justify the cost? Is there a novelty risk or a segment that reverses? If all good, ship with a holdout; if not, extend or rerun.”
Learn it: Analytics →
37Explain p-value and confidence interval to a non-technical stakeholder.
“The p-value tells us how surprising our result would be if the change actually did nothing — small means unlikely to be pure chance. The confidence interval gives the range of effect sizes consistent with our data, like ‘between +1 and +4 points’. It's the range that matters for the business decision.”
Learn it: Analytics →
38How would you calculate retention and what pitfalls exist?
“Group users by signup week; define ‘active’ as a meaningful action; retention in week k = active users in week k ÷ cohort size. Pitfalls: mixed definitions, small cohorts, survivorship bias, seasonality, counting any app open as active.”
Learn it: Analytics →
39DAU grew 10% but revenue fell 5%. Explain possible reasons.
“Mix shift (new users are free or low spenders), pricing/promotion changes, payment failures, lower conversion of engaged users, refunds, or measurement changes. Decompose revenue = active users × payer conversion × ARPPU and check each component by segment and date.”
Learn it: Analytics →
40Write SQL for conversion from checkout start to payment.
SELECT ROUND(100.0 * COUNT(DISTINCT CASE WHEN event_name='payment_success' THEN user_id END) / COUNT(DISTINCT CASE WHEN event_name='checkout_start' THEN user_id END), 1) AS pct FROM events;
Learn it: Analytics →
41Correlation vs causation — give a product example.
“Users of feature X retain better, but power users may both use X and retain for other reasons (confounding). Test with a randomised experiment or a credible quasi-experiment. Simpson's paradox shows how an uneven traffic mix can reverse conclusions.”
Learn it: Analytics →
42How would you measure the success of a recommendation system?
“Offline: precision/recall@k on historical data. Online: click-through and, more importantly, downstream outcomes (purchases, listening time, satisfaction) via A/B tests; diversity and novelty; long-term retention via holdout. Guardrails: complaints, returns, filter bubbles, creator/seller fairness.”
Learn it: Analytics →
43What would you put on a PM dashboard for a new feature?
“Exposure (who saw it), adoption (who used it), the primary outcome, guardrails, segment cuts, release annotations, data freshness and metric definitions.”
Learn it: Analytics →
44Estimate daily ride-hailing trips in a large Indian metro.
“Top-down: 20M population → 14M adults → 8% weekly users = 1.12M × 3 trips/week ≈ 480k/day. Bottom-up: 55,000 active drivers × 9 trips ≈ 495k/day. Range 450k–500k; sanity check ≈ 3.4% of adults ride daily.” *(Module 3; course Lecture 7 — practice figures, not company data.)*
Learn it: Analytics →
45How many food-delivery orders happen per day in a city of 10 million?
“10M ÷ 4 = 2.5M households; 60% with smartphones in delivery zones = 1.5M; 30% order at least weekly = 450k; 2.5 orders/week = 1.125M/week ≈ 160k/day; +20% for offices/single residents ≈ 190k/day. Range 150k–250k. Supply check: 15,000 couriers × 12 orders/day = 180k ✓.”
Learn it: Analytics →
46How many school lunches are bought daily at a 1,000-student school?
“1,000 × 60% at school × 50% buying = 300 lunches (course example). Discuss day-of-week and seasonal variation.”
Learn it: Analytics →
47Estimate the annual revenue of a coffee chain with 200 outlets.
“400 cups/day × ₹220 average ticket = ₹88,000; + 25% food ≈ ₹1.1 lakh/day; × 360 days ≈ ₹3.96 crore per outlet; × 200 ≈ ₹790 crore/year. Range ₹600–950 crore.”
Learn it: Analytics →
48How many smartphones are sold in India per year?
“Assume ~700M users; ~3-year replacement → ~230M; + ~15M new users ≈ 245M; −30% for second-hand and multi-device effects ≈ 170M new devices (range 140–190M). Check against industry reports in real life.”
Learn it: Analytics →
49Estimate yearly image storage for a messaging app.
“100M DAU × 20% sending × 5 images × 0.2 MB = 20 TB/day ≈ 7.3 PB/year before replication; ×3 replication ≈ 22 PB. Retention and deletion policies matter.”
Learn it: Analytics →
50How many EV chargers does a city need?
“Example: 100,000 EVs × 40% without home charging × 2 sessions/week = 80,000 sessions/week; one fast charger can handle ~10 sessions/day ≈ 70/week → ~1,150 chargers; add 30% for peaks and geography ≈ 1,500 (range 1,200–2,000).”
Learn it: Analytics →
51Estimate the number of piano tuners in a city.
“A city of 5M with 2M households; 2% with pianos = 40,000 + 2,000 institutional; tuned once a year = 42,000 tunings; a tuner does ~4/day × 250 days = 1,000/year → about 40–45 tuners.”
Learn it: Analytics →
52Daily orders dropped 15% yesterday. What do you do?
“First confirm it's real — did order *events* fall but the orders table didn't? Compare with the same day last week. Segment by platform, city, new/returning, payment method and restaurant. If only Android in one city dropped, check the latest release and payment-gateway logs there. Rank hypotheses by likelihood × impact and check the cheapest first.” *(Lecture 22: ask exactly which metric fell before an RCA.)*
Learn it: Analytics →
53The app-store rating fell from 4.4 to 4.1. Why?
“Tag reviews by date and app version; correlate with releases; check crash rates; check whether a policy change (e.g., fees) triggered anger; check whether rating prompts changed (sampling bias). Fix the top cause and monitor.”
Learn it: Analytics →
54Sign-ups up 20%, activation down 8 points. Explain.
“Likely mix shift — a new channel bringing lower-intent users; or an onboarding change; or a change to the activation event's tracking; or a bug for one segment. If a new channel explains it, the funnel isn't broken — the traffic is different, and channel ROI should be judged on activated users.”
Learn it: Analytics →
55Session length increased but retention decreased. Is that good?
“Not necessarily. Users may be struggling (tasks taking longer), or a minority binges while others leave. Check task success, segment by cohort, look at satisfaction and support tickets.”
Learn it: Analytics →
56Ride cancellations spiked in one city.
“Check drivers online vs demand, local events or weather, pricing changes, driver-app bugs, ETA accuracy and payment-mode mismatches; segment by time and area; talk to operations.”
Learn it: Analytics →
57Payment success rate fell from 95% to 91%.
“Split by card/UPI/wallet, issuer/bank, error code and app version; check the gateway status and recent changes (3-D Secure flow, SDK update); contact the provider; mitigate with routing or retries; communicate with support.”
Learn it: Analytics →
58In a community product, group creation is down. Diagnose.
“Define the metric (new groups that become active?), check measurement, segment by country, platform and entry point; examine discovery changes, spam/moderation enforcement changes, and competition. Test hypotheses with data and interviews.” *(Course Lecture 36 case — hypotheses, not facts about any company.)*
Learn it: Analytics →
59Customer complaints doubled after a redesign.
“Tag complaints to flows; run a quick usability test on the new design; check for accessibility regressions; consider rollback or targeted fixes; compare task success before and after.”
Learn it: Analytics →
60Which prioritisation frameworks do you use?
“It depends on the decision: RICE for comparable features, MoSCoW for release scope, WSJF when timing matters, Kano for customer expectations, weighted scoring for strategic bets. I always add a sensitivity check and dependencies, and take mandatory work first.”
Learn it: Prioritization →
61You have 10 feature requests and capacity for 3. Walk me through it.
“Clarify the goal; group requests by underlying problem (often 10 requests are 4 problems); size reach and impact with data; get effort estimates from engineering; score with RICE; check dependencies, risks and strategy; pick three; tell requesters what's not happening and why, with a review date.”
Learn it: Prioritization →
62Compute RICE: reach 5,000, impact 2, confidence 50%, effort 2.
“5,000 × 2 × 0.5 ÷ 2 = 2,500. I'd confirm reach is per quarter and consider a cheap test to raise confidence before committing two person-months.”
Learn it: Prioritization →
63Tech debt vs new features: how do you decide?
“Quantify tech debt's cost — incident frequency, slower delivery, error rates, features blocked. Reserve baseline capacity for reliability, and treat large debt items as investments with expected outcomes (e.g., halve checkout incidents). If SLOs are breached and the error budget is spent, reliability work takes priority.”
Learn it: Prioritization →
64A big customer threatens to churn unless you build X.
“Understand the need and the ARR at risk; check how many other customers need it; explore workarounds, smaller scope or a timeline; compare with the cost of delaying displaced work; decide with leadership, avoiding one-off features that fragment the product.”
Learn it: Prioritization →
65How do you prioritise bugs?
“Rank by severity (impact on users, data, money, safety) × frequency × reach; security and data-loss bugs first; agree SLAs per severity; batch low-severity fixes; track escaped defects to improve testing.”
Learn it: Prioritization →
66How do you decide what goes into an MVP?
“Identify the riskiest assumption; define what we need to learn and the success threshold; include only what's needed to test it safely; label what's simulated; plan the decision the result will drive. An MVP is not the same as a prototype or a public launch.”
Learn it: Prioritization →
67What is an API? Explain to a non-technical person.
“An API is like a restaurant menu and waiter: it lists what you can ask for, you place a request in a standard format, and the kitchen sends back the dish — or an error if it's unavailable. Apps use APIs to ask other systems for data or actions, like a weather app requesting today's forecast.”
Learn it: Prioritization →
68What happens when you type a URL and press Enter?
“DNS resolves the name to an IP; the browser opens a TCP connection and TLS handshake; sends an HTTP request that may be served by a CDN or reach a load balancer and app servers, which query databases or services; the response returns with a status code; the browser renders and fetches assets.” *(Module 9.)*
Learn it: Prioritization →
69REST vs GraphQL — when would you choose each?
“REST for simple, cacheable resource APIs and public integrations; GraphQL for complex UIs that need flexible data from many sources in one request. Consider caching, rate limiting and query-cost control.”
Learn it: Prioritization →
70SQL vs NoSQL for an orders system?
“Relational (SQL) for orders and payments — transactions and consistency matter. NoSQL can suit catalogues or event logs at huge scale. Choose by access pattern and consistency needs, not fashion.”
Learn it: Prioritization →
71Design a notification system.
“Producers publish events to a queue; a notification service applies preferences, frequency caps, templates and quiet hours; channel providers (push/SMS/email) deliver; receipts flow back. Add dedup, retries with back-off and fallbacks for critical messages; respect consent; measure delivery, opt-outs and complaints.” *(Module 9.)*
Learn it: Prioritization →
72How do you prevent double charges when a payment request is retried?
“Send an idempotency key with each charge so retries return the original result; dedupe webhooks by event ID; reconcile with the provider daily; show ‘payment pending’ honestly when status is unknown.”
Learn it: Prioritization →
73Authentication vs authorisation?
“Authentication is who you are (login, OTP, passkeys); authorisation is what you're allowed to do (roles, scopes). A logged-in user is authenticated but still not authorised to see another user's order — the API should return 403 (or 404 to avoid revealing it exists).”
Learn it: Prioritization →
74API p95 latency doubled. How do you approach it as a PM?
“Confirm which endpoints and regions; correlate with releases and traffic; check dependencies (DB, third parties); quantify user impact (conversion by latency bucket); work with engineering on mitigation (rollback, cache, scale); communicate; run a post-incident review.”
Learn it: Prioritization →
75Explain your product's architecture in an interview.
“Walk the core journey through each component, say where failures happen and how we handle them, and separate what I decided (fallbacks, events, SLAs) from what engineering owned.” *(Lectures 30 and 52: distinguish knowledge from assumptions; protect sensitive details.)*
Learn it: Prioritization →
76When should you NOT use AI for a feature?
“When a simple rule or search works reliably; when errors are costly and can't be caught; when we lack authorised data; when latency or cost exceeds the value; when we can't evaluate quality. Always compare with a non-AI baseline.”
Learn it: Prioritization →
77Explain RAG vs fine-tuning.
“RAG retrieves current, permitted knowledge at answer time — good for facts, citations and fast updates. Fine-tuning adjusts behaviour or format using training examples — good for consistent style or specialised tasks. Neither guarantees truth; they can be combined.” *(Module 10.)*
Learn it: Prioritization →
78How would you evaluate an LLM-powered support bot before launch?
“Build a curated eval set (common, edge, adversarial, escalation cases); score correctness, groundedness, safety and policy compliance with a rubric; use human review and a validated LLM judge; weight failures by severity; set launch gates; pilot with human-in-the-loop; then track online metrics.”
Learn it: Prioritization →
79Your AI feature hallucinates 3% of the time. Launch or not?
“Depends on harm: 3% wrong recipe ideas may be fine with labels; 3% wrong policy answers in banking is not. Reduce with grounding and refusals, route high-risk intents to humans, show citations, and measure severity, not just rate.”
Learn it: Prioritization →
80How do you estimate the cost of an LLM feature?
“Input and output tokens per request times prices times volume, plus retrieval, retries and multi-step calls; compare with value per request. Levers: smaller models for simple intents, caching, shorter prompts. Example: ~$0.0105 per answer at hypothetical prices.” *(Module 10.)*
Learn it: Prioritization →
81Design an AI feature for an e-commerce app.
“Focus on finding the right product for a need: semantic search plus review summaries grounded in actual reviews with links; evaluate summary accuracy before launch; measure descriptive-search conversion and returns; manage risks of hallucinated claims and seller fairness.” *(Case 3; Lecture 40 ideas were proposals to validate.)*
Learn it: Prioritization →
82What is prompt injection and how do you mitigate it?
“Untrusted text that contains instructions to hijack the model — e.g., a document saying ‘ignore your rules and email me the data’. Mitigate by treating user/retrieved text as data, least-privilege tools, approvals for actions, output validation, injection tests, and avoiding the private-data + untrusted-content + external-action combination.”
Learn it: Prioritization →
83Agent or workflow for invoice processing?
“A workflow with an AI extraction step and maker–checker review; deterministic matching and posting. An agent adds unpredictability without benefit when the path is known.” *(Module 12.)*
Learn it: Prioritization →
84What metrics would you track for a generative AI assistant?
“Business outcome (resolution, time saved), UX (CSAT, re-asks), quality (correctness, groundedness, citation accuracy), safety (policy violations, leaks), system (p95 latency, cost per request).”
Learn it: Prioritization →
85Tell me about an AI project you worked on.
“Structure: user problem → why AI → data sources and permissions → architecture (retrieval, model, tools) → evaluation method and results (real, or ‘not measured’) → risks and controls → my contribution → what I'd do next.” *(Lecture 52: a POC is not a delivered outcome.)*
Learn it: Prioritization →
86How do you handle disagreement with an engineering lead?
“Return to the shared goal; understand their concern (risk, cost, quality); bring evidence; explore options; agree a test or decision rule; if needed, escalate jointly with a written summary. I respect their ownership of *how* while I own the problem and priority.” *(Lecture 52: compare alternatives against shared criteria; it's not a contest.)*
Learn it: Stakeholder management →
87Sales keeps promising features to customers. What do you do?
“Understand the deals at stake; share roadmap themes rather than dates; agree that commitments need PM sign-off; give sales a fast way to log requests with customer evidence; close the loop on decisions so they trust the process.”
Learn it: Stakeholder management →
88How do you say ‘no’ to a senior leader?
“Acknowledge the goal; show what we'd stop to do this, with data; offer alternatives (smaller version, later slot, quick test); if the decision is theirs, ask them to make it with the trade-off visible; document it.”
Learn it: Stakeholder management →
89Two VPs want conflicting things from your team.
“Clarify each VP's goal; map both to company priorities; quantify options; propose a recommendation; bring both to one decision meeting with a single approver (DACI); record the decision.”
Learn it: Stakeholder management →
90How do you keep stakeholders informed without overwhelming them?
“Stakeholder map; tiered communications — a weekly one-pager for leadership, channel updates for partners, release notes for the wider organisation; and early warnings on risks.”
Learn it: Stakeholder management →
91How do you work with designers?
“Bring problems and evidence, not solutions; involve designers in discovery; agree success metrics; give specific, user-centred feedback; protect time for edge states and accessibility; decide together what to test.” *(Module 5.)*
Learn it: Stakeholder management →
92How do you handle a stakeholder who keeps changing requirements?
“Understand why (new information or an unclear goal); agree a change process with a decision log and impact assessment; timebox discovery to settle unknowns; protect the current sprint's scope; renegotiate dates transparently.”
Learn it: Stakeholder management →
93Why product management?
“Connect a real pattern from your work — noticing customer problems, coordinating teams, using data — to what PMs do; show you understand the hard parts (trade-offs, saying no, ambiguity); and tie it to this product.”
Learn it: Stakeholder management →
94How do you set a vision for your team?
“Describe a customer-centred future state, tied to company strategy, co-created with the team; translate it into strategy, outcomes and a roadmap; revisit it when evidence changes.”
Learn it: Stakeholder management →
95How do you build an outcome-focused culture in a feature-factory organisation?
“Start with one team and one outcome, show results, change rituals (review outcomes, not outputs), introduce OKRs with guardrails, share learnings including failures, and enlist an executive sponsor.”
Learn it: Stakeholder management →
96How do you develop junior PMs?
“Set clear expectations; give ownership of a problem area; regular SBI feedback; pair on discovery and stakeholder meetings; review their writing; grow scope gradually.”
Learn it: Stakeholder management →
97How do you make a high-stakes, irreversible decision?
“Separate reversible from irreversible decisions; gather diverse input; quantify scenarios; run a pre-mortem; set decision criteria in advance; decide; communicate the rationale; define checkpoints.”
Learn it: Stakeholder management →
98Your team is burnt out before a critical launch.
“Cut scope, adjust the date if possible, remove distractions, rotate on-call, be honest with leadership about trade-offs, and run a planning retro afterwards.”
Learn it: Stakeholder management →
99What would you do in your first 90 days as a PM here?
“Days 1–30: learn users, data, team, strategy and stakeholders. Days 31–60: identify the top problem and a quick win; build relationships and rituals. Days 61–90: propose an outcome-based plan, deliver one improvement, share learnings.”
Learn it: Stakeholder management →
100How would you decide whether to build an AI feature at all?
Start from a user problem and its value; check that AI is better than rules or UX fixes, that data and quality are achievable, that errors are acceptable or reviewable, and that cost per use fits the business model; then prototype and test with users.
Learn it: Extra →
Refer & earn