Every class, every recap, one link

Strong answers typically pick from: 9/18 said they would pay ₹150/day right now → tells you stated willingness to pay; does not prove actual payment behaviour. 2/18 actually paid during the pilot → tells you real, weak but non-zero commitment; does not prove it converts consistently or at scale. 8/18 could not get equipment in time for a lab → tells you the borrowing workaround has real friction; does not prove people will pay to remove that friction rather than keep borrowing. 4/18 actually joined the pilot despite 7 agreeing → the intent-to-action drop-off is itself evidence of weak commitment.

Verdict (4 marks): the strongest answers land on "partly" — real friction exists (13/18 borrowed last month, 8 missed a lab), but the SAID→DID drop-off (73 said probably-rent → 9 said they would pay → 7 agreed → 4 joined → 2 paid) is steep, so the hypothesis is only weakly supported so far. Full marks are available for a yes or a no if the evidence citation is genuinely strong and internally consistent — the conclusion is never penalised, only an unsupported one.

Marking: 2 marks per evidence piece (tells + does-not-prove), 4 for the verdict. Reward the shape of the reasoning, not the chosen word.

E1 WhatsApp pilot: ICE = 7 × 8 × 8 = 448. RICE = (120 × 7 × 0.8) ÷ 1 = 672. E2 App MVP + ads: ICE = 8 × 5 × 5 = 200. RICE = (2,500 × 8 × 0.5) ÷ 3 = 10,000 ÷ 3 ≈ 3,333.

ICE ranks the WhatsApp pilot first (448 vs 200); RICE ranks the app first (3,333 vs 672). They disagree because RICE adds the two terms ICE ignores: Reach rewards the app’s scale 20× over, and Effort divides — the pilot’s cheapness helps it far less under RICE than its high Ease helps it under ICE. Accept either pick with reasoning; early-stage teams reasonably favour ICE for a first cheap test, and that reasoning earns full marks.

Marking: 1 mark per correct score (4), 1 per correct ranking under each method (2). Method marks apply — right formula, slipped arithmetic loses 1, not all 6.

TAM = 300,000 — every engineering undergrad in the state. SAM = 300,000 × 18% = 54,000 — the portion behaviourally in Segment B. SOM = 54,000 × 10% = 5,400 — the share reachable on campuses the founders can actually operate on this year.

Assumptions that could make SOM misleading (any two): the 18% Segment-B estimate came from a small interview sample (n=18) and may not generalise; "10% of campuses" assumes those campuses behave like the surveyed one; SOM as calculated is population reached, not customers won — no conversion assumption exists yet.

Marking: 2 per correct figure with its one-line meaning (6), 1 per misleading assumption (2). Correct method with an arithmetic slip loses 1 mark, not the question.

Model answer (Narrow): ICP — second-year students in equipment-heavy lab courses on the two pilot campuses. Strongest evidence — 13/18 actually borrowed equipment last month and 2 paid in the pilot (behaviour, not opinion). Biggest contradiction — 73 said "probably rent" but only 2 paid. Riskiest assumption — kit owners will list (Segment C, the supply side, was never interviewed). Next experiment — recruit 20 kit owners and 40 renters on one campus for three weeks; success = ≥25% of renters complete at least one paid rental AND ≥8 owners list; failure = under 10% of renters pay anything. Won’t build — the rental app MVP and automated payments.

Marking: 1 mark per item present and evidence-linked, whichever path is chosen. A Stop decision with all seven items filled in coherently is full marks. The experiment line must carry numbers — adjective-only thresholds ("if it goes well") lose that mark.

Visitor→Signup = 3,000 ÷ 12,000 = 25%. Signup→First Session = 1,050 ÷ 3,000 = 35%. First Session→Trial = 210 ÷ 1,050 = 20%. Trial→Paying = 126 ÷ 210 = 60%. Overall Visitor→Paying = 126 ÷ 12,000 = 1.05%.

Largest % drop-off: First Session→Trial (80% lost at that step). But the biggest numerical loss is not automatically the biggest opportunity: Trial→Paying already converts at 60%, unusually strong, and cold top-of-funnel always leaks — repairing a high-intent stage can beat buying more traffic, while a small stage can be the wrong place to spend effort. Accept either conclusion if reasoned from the numbers.

Marking: 1 per correct conversion (show the division), 1 for naming First Session→Trial, 2 for a numbers-grounded argument either way.

CAC = ₹60,000 ÷ 50 = ₹1,200. LTV = ₹349 × 4 = ₹1,396. LTV:CAC = 1,396 ÷ 1,200 ≈ 1.16 : 1.

Interpretation: below the commonly cited 3:1 healthy benchmark — each paid customer only just pays back their acquisition cost. But at six weeks and 126 total customers the sample is very early, so the ratio is directional, not a verdict. Do not fail "concerning" vs "acceptable for this stage" — grade the reasoning.

Marking: right formula with a slipped number loses 1 mark per line, not the line.

Any two of: the sample is very small and early (126 customers, 6 weeks), so both CAC and lifetime are unstable estimates; organic/referral customers (76 of 126) already outnumber paid ones and may carry a completely different, cheaper CAC; the 4-month average paid lifetime is a 6-week snapshot and likely to move as the product matures; CAC from one channel at ₹60,000 of spend does not predict CAC at ₹6,00,000 — channel costs change non-linearly with scale.

Marking: 3 marks per reason that is specific to THESE numbers. A reason generic enough to apply to any startup ("it is too early to tell") earns at most 1.

No single correct decision. Model answer (fix before scaling): hold paid spend and fund funnel repair, because (1) 1.16:1 is below 3:1 — scaling paid marketing buys customers who barely pay back ₹1,200 each; (2) the funnel loses 80% at First Session→Trial — the 840 students who finished a session but never trialled are the cheapest recoverable cohort; (3) Trial→Paying converts at 60%, so any trial lift flows strongly into paying customers. A fourth acceptable point: 76 of 126 customers came free via organic/referral, so a referral incentive may beat a ₹1,200 CAC.

A reasoned "spend now" answer also scores — e.g., month-2 retention of 75% suggests lifetime will improve, so buying data while fixing onboarding is defensible. Marking: 2 for a clear decision, 1 per number-grounded reason (max 3). Reasons must cite figures, not vibes.

There is no single correct set of scores — grade the justification, not the number. A reasonable strong answer: Evidence strength 3/5 (18 real completed sales, but 60 students reached over three weekends), Market reachability 4/5 (one contained campus, both sides reachable through hostel groups), Founder-team fit 4/5 (repair-stall background is a real, relevant asset for condition checks), Ease to test further 4/5 (a WhatsApp-based pilot has already proven possible).

Weighted total: 0.30 × 3 + 0.25 × 4 + 0.20 × 4 + 0.25 × 4 = 0.90 + 1.00 + 0.80 + 1.00 = 3.70/5. Full marks are available for any individually justified set of scores with correct weighted arithmetic — write the products, then the total.

Marking: score with no evidence-linked justification earns at most 1 per criterion; the 2 arithmetic marks reward correct method on the student’s own numbers.

Any two of: the three-weekend pilot may reflect pent-up one-time supply rather than durable liquidity — once the backlog sells, listings could dry up; 18 sales from 60 students reached could be novelty, not repeatable demand; only 9 of 34 listings came from repeat sellers, so supply durability is thin; the 11 buyers who "would pay" ₹99 for verification is stated intent, not paid behaviour; moving transactions from hostel groups onto a platform transfers the scam risk to the venture — one bad sale could kill trust.

Marking: reward flaws that re-examine the student’s own Q3A scoring logic (e.g., noticing they scored evidence 3/5 while supply-side durability was never tested). Restating a weakness already spelled out in the case text earns at most 1.

Model answer: run the verified-tag pilot in four more hostels for four weeks. Success = at least 20% of listed cycles sell within 14 days AND at least 30% of buyers pay ₹99 for the verified tag. Failure = under 8% sell-through, or fewer than 10% of buyers paying for the tag. Both thresholds test the two riskiest assumptions at once: durable liquidity and paid verification.

Marking: 3 for a concrete experiment, 3 for a numeric success threshold, 3 for a numeric failure threshold. Adjective-only thresholds ("if it goes well") lose the numeric marks even when the experiment idea is reasonable.

Model memo: Why now — every semester ends with final-year sell-offs, and hostel groups are chaotic (16/36 called them scammy), so the workaround is visibly failing. Strongest evidence — 18 completed sales in three weekends with zero marketing. Biggest unknown — whether supply repeats (only 9/34 listings were repeat sellers) and whether buyers will actually pay ₹99, not just say it. Next 30-day bet — the four-hostel verified-tag pilot from Q3C. Kill signal — under 8% sell-through or under 10% tag payment after four weeks → kill or narrow to a listings board. Need next — permission from two more hostels and one person trained to run condition checks.

Marking: roughly 1.5-2 marks per line; specificity beats generic phrasing. "Biggest unknown: whether demand holds once novelty wears off" scores higher than "we don’t know everything yet". The kill line must be a number, not a mood.