Lesson 1 of 8 · 55 min

How to run any product case (senior bar)

10-dimension senior rubric, type-spotting map for every case shape, and a Slack notifications weak vs strong micro-case.

Why framework recitation fails senior loops

Senior product cases grade structured judgment under constraints, not whether you can recite CIRCLES. The candidate who treats every prompt the same way fails both classic and AI-native loops. This lesson gives you the 10-dimension hire rubric, a type-spotting map for the seven case shapes in the rest of this track, and a micro worked case (Slack notifications) so you can hear the difference between a junior and senior 10-minute answer. Framework mechanics live in the companion track AI Product Sense & PRDs — here we use light structure only.

Two loops sit under the same title 'product case.' Mainstream PM loops (Meta, Google, Stripe, Airbnb) still run product sense plus execution. AI-native employers add safety judgment, eval thinking, and frontier-product judgment. What unites them: the interviewer is listening for an executive making trade-offs, not a checklist filling itself in. Structure is the floor in 2026 (AI can generate a clean outline in seconds). Differentiation is a named user, a cut you defend, metrics that falsify you, and risks you monitor.

The 10-dimension senior rubric

Score yourself and others on these ten dimensions. Hire / lean-hire candidates are strong on most; no-hire answers collapse on framing, segmentation, or trade-offs. Memorize the strong column — that is the bar you are practicing toward.

code

1SENIOR CASE RUBRIC (10 dimensions)23  Dimension          Strong (senior)                         Weak (junior)4  ------------------ ---------------------------------------- ------------------------------5  Problem framing    User + job + constraints in ~2 min;     Jumps to features6                     clarifying Qs that reframe scope7  Segmentation       1–2 segments chosen; non-goals named    'All users' / no prioritization8  Trade-offs         Explicit: lose X to gain Y              Feature list, no sacrifice9  Prioritization     Ruthless cut; sequencing defended       Equal weight to every idea10  Metrics            North star + drivers + guardrails       Vague 'engagement / growth'11  XFN / stakeholders Named partners + incentives             'I'd work with eng'12  Technical fluency  Right altitude (latency, eval, data)    Buzzwords only13  Risk / failure     Kill criteria, monitoring, fallback     Absent14  Communication      Waypoints, pauses, structure            Stream of consciousness15  Learning signal    Specific failure + lesson applied       Cliché or none1617  Mid bar = restates, lists segments without pick, mentions trade-offs in passing.18  Self-score after every mock: pick your two weakest dimensions and rewrite only those.

Two dimensions fail most mid-level candidates under time pressure: segmentation with non-goals and metrics that can falsify the plan. If you only have energy to upgrade two habits this track, upgrade those. Technical fluency matters more on AI cases (L6) and marketplace/T&S (L8); do not force model jargon into a classic metric case.

Universal flow (light structure, not gospel)

Use this spine when the type is still unclear, or when the interviewer gives you open air. Do not announce framework names. Narrate decisions. CIRCLES as a recited checklist is a cargo-cult signal on senior loops — coaches and hiring managers flag it. If you already know CIRCLES from the companion track, treat it as optional scaffolding you discard the moment the case type is clear.

code

1UNIVERSAL FLOW  (~40–50 min round)23  1. CLARIFY (60–90s)   reframe-grade questions only; then state assumptions4  2. THESIS (30s)       one line: user · job · constraint · what success means5  3. SEGMENTS (3–5m)    3 candidates → pick 1–2 → name non-goals6  4. OPTIONS (5–8m)     2–4 bets, not 12 features; compare on impact/risk7  5. MVP + NON-GOALS   what ships, what waits, what you refuse8  6. METRICS + FALSIFY north star · 2 drivers · 1 guardrail · kill if...9  7. RISKS             failure mode · detection · mitigation / fallback1011  Timing rule: if you are still clarifying at minute 5, you are stalling.12  Make a smart assumption, name the risk, move.

Clarify vs stall. A reframe question changes the problem: 'Is this growth, monetization, or retention of existing users?' 'Is success measured by weekly active teams or by notification open rate?' A stall question buys time without narrowing: 'Can you tell me more about Slack?' Cap yourself at 2–3 clarifying questions, then state 2–3 assumptions out loud.

Type-spotting map (first 60 seconds)

The rest of this track is one archetype per lesson. Your first job in any live case is to spot the type so you load the right structure — not the same universal flow every time. Say the type quietly to yourself, then open with the matching first move.

code

1TYPE-SPOTTING MAP  (prompt cue → first move → this track)23  Cue in the prompt                         Type              First move4  ----------------------------------------- ----------------  -------------------------5  'Design a product / from scratch'         0→1 (L2)          JTBD + ruthless MVP cut6  'Grow / 2x / improve X metric'            Improve (L3)      North star + driver tree7  'X dropped / declined — what happened?'   Root-cause (L4)   Funnel decomp + hyp tree8  'Favorite product / critique / fix'       Critique (L5)     Tradeoffs product chose9  'Add AI / copilot / agent / LLM'          AI feature (L6)   Eval set FIRST10  'Should we enter / launch / compete?'     Strategy (L7)     Beachhead + kill criteria11  'Two-sided / supply / demand / liquidity' Marketplace (L8)  Cold side + unit econ12  'Abuse / spam / fraud / safety'           Trust&safety (L8) Threat model + layers1314  Hybrid prompts exist ('engagement flat — design an AI fix'). Name both types,15  then lead with diagnosis before solution (L3/L4 before L6).

When the prompt is hybrid, say the split out loud: 'This is primarily a metric-improve case with an optional AI solution path — I'll diagnose the driver tree before proposing any model-backed feature.' That sentence alone separates senior candidates from people who bolt AI onto the first idea.

Worked micro-case: Improve Slack notifications

Prompt: 'You're the PM for Slack notifications. Engagement with notifications is down and users complain of noise. What do you do?' Time box: ~10 minutes for a micro answer (full round would go deeper on experiments). We contrast weak vs strong so you can hear the rubric dimensions fail and pass.

Weak answer (what mid candidates actually say)

code

1WEAK (~10 min) — checklist mode23  'I'd talk to users and look at data. Notifications matter for engagement.4   I'd improve relevance with AI, add digests, mute options, better defaults,5   smarter batching, channel-level controls, priority inbox, and maybe6   machine learning ranking. Success = higher open rates and less complaint7   volume. I'd work with design and eng to ship iteratively.'89  Rubric kills:10  · Framing: restates, no constraint thesis11  · Segmentation: 'users' as one blob12  · Prioritization: 8 ideas, no cut13  · Metrics: open rate without guardrails (you can raise opens by spamming)14  · Risk: none (over-muting kills activation of new teams)15  · Trade-offs: 'AI relevance' without cost of false suppressions

The weak answer is not wrong on facts — digests and ranking are real levers. It fails because it never makes a product decision. Interviewers grade decisions under constraints. A laundry list is free; a defended cut is expensive (and therefore scarce).

Strong answer (senior bar, same 10 minutes)

code

1STRONG (~10 min) — judgment mode23  CLARIFY (45s): Is the goal reduce opt-out / complaint, or recover DAU from4  notification-driven opens? I'll assume free + paid workspace product;5  enterprise DLP/compliance is out of scope unless you pull me there.67  THESIS: Notification value = signal density for people who must respond8  in-flow at work. Noise is a trust tax; we optimize 'useful interrupts per9  user-day' not raw sends.1011  SEGMENTS (pick 1):12    A) Power responders (on-call-ish, many @mentions) — high need, high noise13    B) Peripheral members (lurk, few channels) — low need, easy to over-ping14    C) Admins (policy) — different job entirely15  Focus v1: A, because they drive team retention and feel pain daily.16  Non-goal: redesigning all mobile push chrome; full AI rewrite of ranking.1718  OPTIONS (3 bets, ranked):19    1) Default-off for low-urgency event classes + explicit 'why you got this'20    2) Per-channel response-SLA presets (mentions vs everything vs nothing)21    3) Smart batching for non-mention traffic (hour windows)22  Pick 1+2 for v1. Defer 3 until we measure residual noise after defaults.2324  MVP: ship safer defaults + channel presets for power users; instrument25  before/after complaint rate, mute rate, and 'time-to-first-response' on26  @mentions (guardrail: must not rise).2728  METRICS:29    North star: useful-interrupt rate (opens that lead to a reply or reaction30                within 15m on mention-class events)31    Drivers:    mention precision; default sensitivity by event class32    Guardrails: @mention response latency; new-member activation via notifs33    Kill if:    mention response latency +15% after 2 weeks → rollback defaults3435  RISKS: over-suppressing → silent teams; enterprise admin override conflicts.36  XFN: data (event taxonomy), clients (defaults UX), trust (abuse of @channel).

Walk the strong answer against the rubric: framing names the job and constraint; segmentation picks power responders and non-goals; trade-offs defer ranking; metrics include a kill criterion; risks name over-suppression. That is what 'structured judgment' sounds like in ten minutes.

Interview ways (method)

Practice these out loud until the first 90 seconds of any case is automatic. Record yourself on the Slack micro-prompt and self-score the 10 dimensions.

01"How do you structure a product case?" → Clarify only to reframe, one-line thesis, pick 1–2 segments with non-goals, 2–4 options, MVP + metrics that can falsify you, risks with kill criteria — structure as decisions, not framework names.
02"What's a good clarifying question?" → One that changes the decision space (growth vs retention, which segment, what success metric) — not "tell me more about the product."
03"How do you show seniority?" → Name what you refuse, the trade-off you accept, and the metric that would kill the plan.
04"What if I don't know the domain?" → State assumptions, pick a segment you can reason about, invite correction; never stall for five minutes of background.
05"CIRCLES or not?" → Light structure is fine; reciting CIRCLES as gospel is a senior red flag. Adapt to case type in the first 60s.
06"How do you practice?" → Timed cases + self-score on the 10 dimensions + rewrite only the two weakest sections — not endless outline polishing.

Homework before L2: pick any IGotAnOffer or Exponent product-sense prompt, run 12 minutes, score yourself on the rubric, rewrite the weakest two dimensions only. That practice loop is how the rest of this track compounds.

Full spoken senior answer (~12 minutes)

Read this aloud at interview pace. It is the Slack notifications micro-case expanded to full senior length — the same judgment as the short strong answer, with room for pushback, sequencing, and XFN. If you can deliver this quality without reading, you clear the method bar.

code

1SPOKEN SENIOR ANSWER — Improve Slack notifications (~12 min)23  [0:00–1:00 CLARIFY + THESIS]4  "I'll treat this as an improve-metric / product-sense hybrid: engagement with5  notifications is down and noise complaints are up. Two different goals hide6  here — reduce opt-out and complaints, versus recover DAU from notification-7  driven opens. I'll assume free + paid workspaces, consumer and mid-market;8  enterprise DLP and compliance admin policy is a non-goal unless you pull me9  there. Thesis: notification value is signal density for people who must10  respond in-flow at work. Noise is a trust tax. I optimize useful interrupts11  per user-day, not raw sends. If useful interrupts fall while complaints fall,12  we failed — we just muted the product."1314  [1:00–3:00 SEGMENTS + NON-GOALS]15  "Three segments. A: power responders — on-call-ish, many @mentions, high need16  high noise. B: peripheral members — few channels, easy to over-ping. C: admins17  who set policy. V1 focuses on A because they drive team retention and feel18  pain daily; B benefits from safer defaults as a side effect. Non-goals: full19  ML rewrite of ranking, redesigning all mobile push chrome, enterprise DLP20  notification policy. If Enterprise is the real ask, I restart with admin21  buyer and audit logs — different product."2223  [3:00–6:00 OPTIONS + CUT]24  "Three bets only. One: default-off for low-urgency event classes plus explicit25  why-you-got-this on each interrupt. Two: per-channel response-SLA presets —26  mentions only, everything, or nothing — so power users encode team norms.27  Three: smart batching for non-mention traffic in hour windows. I pick one and28  two for v1. Batching waits until we measure residual noise after defaults;29  batching on a broken taxonomy just delays spam. I am deliberately not doing30  a priority inbox clone of email — Slack's job is in-flow work chat, not a31  second Gmail. Trade-off I accept: some low-urgency channel activity becomes32  invisible unless users open the app. Trade-off I refuse: suppressing @mentions33  to juice complaint metrics."3435  [6:00–8:30 MVP + METRICS]36  "MVP: safer defaults by event class, channel presets for power users, and37  instrumentation before/after. North star: useful-interrupt rate — opens that38  lead to a reply or reaction within fifteen minutes on mention-class events.39  Drivers: mention precision, default sensitivity by event class. Guardrails:40  @mention response latency must not rise; new-member activation via notifs41  must not collapse. Kill if mention response latency is up fifteen percent42  after two weeks — rollback defaults. Secondary watch: mute rate and support43  tickets tagged notification. I will not celebrate higher open rate alone;44  opens are gameable with more spam."4546  [8:30–10:30 RISKS + XFN + SEQUENCE]47  "Risks: over-suppression creates silent teams; admins override defaults and48  reintroduce noise; power users game @channel. Mitigations: holdout experiment49  by workspace, admin analytics showing useful-interrupt not just volume, and50  rate limits on mass-mention abuse. XFN: data for event taxonomy, clients for51  defaults UX, trust for @channel abuse, sales/CS for enterprise exceptions.52  Sequence: week one taxonomy + instrumentation; weeks two to four defaults53  experiment on ten percent of workspaces; then channel presets; batching only54  if residual noise remains."5556  [10:30–12:00 CLOSE]57  "Summary: optimize useful interrupts for power responders; safer defaults and58  channel presets; defer ranking rewrite; kill on mention latency regression.59  What I refuse: volume metrics as success, and enterprise DLP scope creep.60  Happy to go deeper on experiment design or admin surfaces if useful."

Practice: record yourself delivering the close without notes. If you cannot name the north star, guardrail, non-goal, and kill criterion in thirty seconds, the structure is not yet muscle memory.

Follow-up pressure (method)

code

1PRESSURE Q → SENIOR REPLY23  Q: "Why not start with AI ranking? Everyone else does."4  A: Ranking on a messy event taxonomy amplifies noise. Fix defaults and5     labels first; ranking is a v2 unlock when residual noise is measured.67  Q: "Cut your plan by fifty percent — what ships?"8  A: Default-off for low-urgency classes only, plus instrumentation. Presets9     wait. Still protect mention latency as kill criterion.1011  Q: "Open rate went up in your test. Ship it?"12  A: Not if useful-interrupt and mention latency moved the wrong way. Open13     rate alone is not success.1415  Q: "Enterprise customer wants everything on by default."16  A: That's a different buyer. Offer admin override as a paid policy surface;17     don't reverse consumer defaults for one deal without a package.1819  Q: "You only asked two clarifying questions — is that enough?"20  A: Yes if they reframed goal and scope. I stated assumptions; correct me21     if Enterprise or DAU recovery is the real objective.2223  Q: "How is this different from CIRCLES?"24  A: Same bones — clarify, user, prioritize, measure — spoken as decisions,25     not chapter titles. Type-spotting chose improve-metric structure.2627  Q: "What's your learning signal from a past failure?"28  A: Cite a real mute/spam incident if you have one: e.g. optimized sends,29     raised opt-out — now I always pair engagement with a trust guardrail.

The 8 types of PM interview questionsLewis C. Lin

Watch with the type-spotting map open. Pause after each example prompt and force yourself to name the archetype in five seconds before the video answers.

articleThe definitive guide to mastering product sense interviewsBen Erez (Lenny's Newsletter)articleThe 8 types of PM interview questionsIGotAnOffer articleProduct Manager case study interview guideHacking the Case Interview articleMeta product sense interview (questions, process, prep)IGotAnOffer

Checkpoint

In a product case, you have already asked two clarifying questions. The interviewer says "whatever you think is reasonable." Best next move?

AAsk three more detailed questions about company strategy, org structure, and historical metrics before proposing anythingBState 2–3 explicit assumptions (user, success metric, scope constraint), name what would reverse them, and open with a one-line thesisCJump straight into a 10-item feature list so you show creativity under time pressure

Checkpoint

Which clarifying question is reframe-grade (changes the decision space) rather than a stall?

A"Can you tell me more about how the product works today?"B"Is the primary goal recovering retention of existing power users, or acquiring new users via this surface?"C"What tech stack does the team use?"

Checkpoint

You hear: "Conversion dropped 15% week-over-week in segment X — what do you do?" Correct type-spot and first move?

A0→1 design — immediately brainstorm features that could raise conversionBRoot-cause — funnel decomposition, ranked hypotheses, falsify before prescribing fixesCStrategy entry — discuss whether to exit the market

Checkpoint

In the Slack notifications micro-case, which metric pair best matches the senior bar?

ANorth star = raw notification sends; guardrail = app store ratingBNorth star = useful-interrupt rate (e.g. mention opens that lead to reply/reaction); guardrail = @mention response latency must not worsenCNorth star = open rate only — if opens rise, the product is working

Checkpoint

An interviewer pushes: "Name a non-goal for your v1." Weakest answer?

A"Full ML ranking rewrite of the entire notification graph — too much surface area before we fix defaults and taxonomy."B"Enterprise DLP and compliance notification policy — different buyer and risk model; out of scope unless you want B2B admin."C"I wouldn't do anything low impact — I'd only do high-impact work."

How ready are you to open any product case with type-spotting + the senior rubric?

New to itGetting thereConfident

Method locked

Senior bar = structured judgment under constraints (10-dimension rubric).
Universal flow is light scaffolding — never CIRCLES-as-gospel.
Type-spotting in 60s loads the right first move for L2–L8.
Slack micro-case: weak laundry list vs strong thesis + cut + falsifying metrics.

Next: Case type 0→1 — design a product from scratch with ruthless MVP cuts.

Sources

Free to read · better with Enzo

Learn it with Enzo

Save your progress, answer the checkpoints, and let Enzo quiz you on what you just read.

Continue with Enzo Sign up free