Unit Economics

Phase 4 · Week 37 · ~20 hours

Where it sits. Second module of the strategy-and-moats phase. Last week taught you to name the machine: the six revenue archetypes, the ₹1/$1 waterfall, the profit engine hiding in the segment note. Now the lens shrinks to the atom, the smallest repeatable unit the whole company is built from. One store. One subscriber, one order, one policy, one billed hour. A company P&L is an average, and averages are where the truth goes to hide: a chain can report rising profits while every new store it opens is worse than the last; a subscription business can report record revenue while every cohort it acquires churns faster than the one before. Unit economics is the discipline of asking the marginal question, what happens when one more unit is added?, because that, not the blended average, is what growth will actually deliver. You already own the tools: contribution margin and DOL (last week), the annuity and perpetuity machinery (time value of money, where an LTV is just an annuity wearing a hoodie), ROIC as the master test (returns on capital), and the deep suspicion of adjusted numbers (Phase 2). Here they compress into the analyst's fourth teardown question: do the unit economics work, and are they getting better or worse? Next week widens back out to industry structure; the moat weeks after that ask what protects good unit economics from competition; Phase 5 speaks this atom in seventeen sector dialects (NIM per rupee of assets, RevPAR per room, RASK−CASK per seat-km, VNB per policy).

Suggested hour budget. Core content: 8h · worked examples: 2h · practice set (including the three gate rebuilds): 5h · applied mini-project: 3h · flashcards + mastery check: 2h.

App link. Four drills in the Industry & Business Model branch grow out of this week: the unit-picker (business description in, correct unit + justification out), the cohort-triangle reader (spot maturation vs vintage decay vs calendar stripes against the clock), the LTV/CAC honesty auditor (a pitch-deck metric in, the list of inflations out), and the rebuild engine (raw operational data in, full unit P&L out, graded line by line). The module gate (master map: "Unit-economics rebuilds ×3 vs key") runs on the rebuild engine.

Learning objectives

By the end you can:

1. Choose the correct unit of analysis for any business using the three tests (it repeats; it carries a full revenue-and-cost loop; it matches how growth capital is actually spent), whether that unit is a store, a subscriber, an order, a policy, a seat-hour or a loan, and run the two-layer split (flow unit for contribution, stock unit for capital and CAC) where one unit is not enough.
2. Compute contribution margin properly as price minus all costs that vary with the unit, classified by behavior rather than by P&L caption, then build the CM1/CM2/CM3 ladder for an e-commerce business from raw data, reading each company's own ladder definition from its disclosures before comparing anything.
3. Compute CAC three ways (paid, blended, fully-loaded), decompose it by channel, and explain why blended CAC is a mix statement that flatters a company exactly when it is scaling paid acquisition.
4. Work the churn algebra: logo vs revenue churn, monthly↔annual conversion (compounding, not ×12), churn-derived expected lifetime (1/churn) with its geometric assumption stated, and GRR vs NRR as the gross/net retention pair (previewing the full NRR treatment in Phase 5).
5. Compute LTV honestly: margin-based, churn-derived, and discounted (LTV = m·r/(1 + d − r)). Quantify how much the undiscounted, revenue-based version overstates, and name every lever in the LTV/CAC abuse catalogue.
6. Compute CAC payback in months as the cash-reality metric, place it against archetype thresholds, and connect payback × growth rate to cash burn and financing dependence.
7. Build and read a cohort retention triangle: rows as maturation curves, columns as vintage quality, diagonals as calendar shocks; then demonstrate with numbers how growing blended totals can mask decaying cohorts.
8. Build two complete unit P&Ls end-to-end, one store (DMart-style, through capex, maturation, and payback) and one subscriber (SaaS/OTT-style, through CAC, LTV, and payback). Compute break-even units and utilization thresholds, and run the reconciliation test (units × unit P&L + corporate ≈ reported P&L) to catch allocation games.

Prerequisites & connections

Builds on. M4.01 for the archetype taxonomy (each archetype implies its natural unit) and the waterfall from revenue to FCF, which the unit P&L shrinks down to a single unit. M4.01 also gave you contribution margin and DOL, which get formalised per unit here, plus the habit of keeping GMV apart from net revenue. M2.01–M2.03 for margin structure, DuPont and ROIC. A store payback calculation is ROIC computed one store at a time. M2.04–M2.05 for cash conversion and working capital, because a unit has a balance sheet as well as a P&L. M3.01 for TVM. The discounted LTV formula is the growing-perpetuity/annuity algebra applied to a customer, so if LTV = m·r/(1+d−r) looks alien, rederive it from the geometric series before you go on. M3.05 for reverse DCF, where "what does the price imply?" becomes, for money-losing consumer tech, "what unit economics does the price imply?" M1.06 for revenue recognition. Whether a discount is contra-revenue or a marketing expense (consideration payable to a customer, Ind AS 115/ASC 606) decides where it goes in the CM ladder, and companies exploit this.

Feeds forward. In M4.03, industry structure decides whether good unit economics survive contact with competition. In M4.04–M4.06, a moat is, mechanically, the thing that stops unit economics from mean-reverting, and you can't grade what you can't compute. In Phase 5, every sector playbook is this plus sector physics: SaaS (NRR, magic number), retail (GMROI, sales/sq ft), banks (NIM − credit cost per rupee of assets), insurance (VNB per policy, combined ratio), hotels/airlines (occupancy and load-factor breakevens, the utilisation maths of the break-even section below). In Phase 8, teardown question 4 ("the unit economics: smallest repeatable unit, contribution, CAC/LTV, payback") is this material run in ten minutes. Competency C3 is certified partly on the skills built here.

In short. Find the smallest thing the company does over and over. Work out what one more of it really earns after every cost that comes with it. Acquisition and discounting count too. Check that claim against cohorts instead of averages, and reconcile the unit story back to the reported P&L. Only then decide whether growth is a promise or a threat.

Core teaching content

4.7 LTV/CAC: the ratio and its abuse

LTV/CAC answers a single question. For every rupee spent manufacturing a customer, how many rupees of present-value margin come back? The folk threshold says ≥ 3× is healthy, and there's real logic behind it. Contribution in LTV still has to cover fixed costs (tech, G&A). That money is locked up for years, and even after discounting you carry forecast risk. On top of that, the estimate itself is soft. At 1×, acquisition is a charity. A true 3× funds a durable business.

The ratio is only as good as its two inputs, though. Both can be gamed in the same direction. Run this abuse catalogue as a checklist on every deck and DRHP:

1. Revenue LTV skips margin and inflates the figure 3–5×.
2. Undiscounted lifetime skips d and inflates 1.5–2.5× for long-lived customers, as the LTV arithmetic above showed.
3. Blended-early churn, or young-cohort churn extrapolated, inflates the lifetime, in a direction you can't know without the triangle.
4. Expansion assumed forever, as in "NRR is 130%, so LTV is infinite". At NRR > 100% the naive series literally diverges. The fix is a fade to ≤ 100%, plus discounting.
5. Blended CAC while scaling paid shrinks the denominator with free customers.
6. Denominator games count "users" in place of transacting customers.
7. Fully-loaded costs omitted from CAC: sales payroll, SBC, onboarding, first-order discounts.
8. Survivor cohorts compute LTV on customers still active ("our year-3 customers spend 2× more"). The right base is the whole acquired cohort, including the 60% who left.

Pull three of these levers and a stated "LTV/CAC of 6×" is routinely a true 1.5×. So recompute it yourself from the cohort exhibits, or refuse to cite it. Even computed properly, the ratio can't see when the money comes back. That's payback's job.

4.11 The unit P&L, build two: a subscriber (SaaS/OTT-style)

The subscriber build runs the same logic on a unit you can't touch. Here's the template first, then a real-scale anchor:

1. Revenue per subscriber. ARPU/month. Watch the base, which should be net of GST and net of app-store commission. A ₹299 plan sold through an app store at 15–30% commission nets ₹254 or less.
2. − Variable serving cost per subscriber. Streaming delivery/CDN or hosting, payment collection, per-user support. It's usually small. Content and R&D are fixed with respect to the next subscriber, and that's the archetype's entire operating-leverage story from last week, so they stay below the line.
3. = Contribution per subscriber per month.
4. Divide by churn for lifetime. Multiply and discount for LTV. Set that against CAC for LTV/CAC and payback.
5. Then the two sanity checks. Verify churn and expansion against the cohort triangle, and reconcile subscribers × contribution − fixed costs to the reported P&L.

Real-scale anchor: Netflix. The figures are illustrative, modelled on ~FY24 scale, so pull the actual 10-K and rebuild them. Take ~300m members and revenue of ~$39bn, which gives a blended ARPU of ~$11.6/month. Cost of revenues (content amortisation + delivery) runs ~54% of revenue, but almost all of it is fixed content. The cost that really varies per member (delivery, payments, support) is maybe $1–1.5/month.

Treat content fairly both ways. For operating leverage, contribution per incremental member ≈ $10+/month, and that's how margins expand mechanically as members grow. For steady-state unit economics, load each member with their share of content, ≈ $6.3/month, and you're left with ~$5.3/month of fully-loaded contribution. Marketing runs ~$2.9bn against gross adds of ~100m+. Those gross adds are net adds plus replaced churn, and at ~2%/month churn on 300m members, churn alone recycles ~70m/year. That puts CAC ≈ $25–30, so payback ≈ 5–6 months on fully-loaded contribution. Lifetime at ~2%/month churn ≈ 50 months. For the discounted LTV, use annual m ≈ $64, r ≈ 0.785 and d = 10%: 64 × 0.785 / (1.10 − 0.785) ≈ $160, so LTV/CAC ≈ 5–6×.

Five numbers, and the whole business is in them. Churn that low for the industry makes a modest CAC extraordinarily productive. An Indian OTT running the same arithmetic at ₹120 blended ARPU, 5–6%/month churn, and app-store tolls finds out why the model that compounds in Los Gatos bleeds in Mumbai. Its unit template is identical, but its inputs describe a different business.

4.13 When unit economics mislead: the games catalogue

Unit economics is self-reported analytics, and none of it is audited GAAP. No standard defines "contribution," "CAC," or "cohort." Companies use that freedom. Here are the five game families. Each comes with its tell and its antidote:

1. Allocation games (moving costs below the line). This is the commonest. Reclassify variable costs as fixed "corporate" and contribution fattens. Delivery subsidies get parked in "other expenses", per-order support gets centralised into "technology", and returns processing becomes "one-time".

The Ind AS 116 special. Since 2019, rent is no longer an operating expense. It became depreciation + interest, as you saw with lease accounting, so a leased retailer's "store EBITDA" mechanically excludes its largest store-level cost. Go back to SwiftMart's leased store. True post-rent store profitability is 6.9, while reported "store EBITDA" is 11.4. That's 65% inflation with zero change in economics.

The famous specimens are WeWork's "community-adjusted EBITDA", a building's P&L before the costs of running buildings, and Uber's early "core platform contribution margin". That was profitable-city arithmetic before the R&D, G&A, and driver incentives that made the platform run.

Antidote: the reconciliation test. Force units × unit P&L + disclosed corporate ≈ reported EBITDA. If the residual is >10–15% of the total, or labelled vaguely, the unit story is curated. And always restate store/branch economics with rent in, whatever the lease standard did to the P&L.

2. Blended masking (mixing the good into the bad). You saw it with cohorts in the triangle, and the same trick runs on every axis. Mature stores mask dilutive new ones, which is the store mix effect from earlier. That's legitimate if disclosed and a game when it isn't. Top-10 profitable cities mask the bleeding in expansion cities. Organic masks paid CAC. Enterprise NRR masks SMB churn. Antidote: demand the split along the axis of growth, or rebuild it yourself. Growth happens at the margin, and the margin is what blending hides.

3. Growth-spend classification ("that cost is an investment"). The pitch goes "marketing is CAC: it builds an asset, so exclude it from profitability". That's true only if the cohorts repay it. Payback and the triangle tell you whether they do. If month-12 retention is 20% and falling by vintage, marketing is the price of renting this quarter's revenue, and calling it an investment changes nothing. There are related moves. First-order discounts get booked as marketing instead of contra-revenue, which inflates net revenue and CM1, as the ladder showed. "Launch costs" and "pre-opening expenses" get excluded from store economics forever, while the company never stops launching. Brand campaigns get classified as fixed while sales clearly track them. Antidote: reclassify by behaviour yourself, and test the "investment" claim against the cohort triangle. An investment has a measurable, retained return. An expense recurs.

4. Denominator and survivor games. CAC per "user", where it should be per transactor. Retention among "active" users, where active = opened the app. LTV measured on surviving customers only. Cohort charts that start at month 3, labelled "early-period noise excluded", which means the cliff has been cropped out. Dark-store economics shown for "mature stores only" while 60% of stores are immature. Antidote: one denominator, the one you chose when you picked the unit, applied everywhere. And whole cohorts, from month 0.

5. The missing balance sheet. Per-unit P&Ls get shown without the unit's capital. You'll see store payback without the security deposits and capex, subscriber LTV without the content balance sheet that serves the subscribers, and "asset-light" cloud kitchens whose kitchens are carried in someone else's numbers through leases. Antidote: every unit P&L you build gets a unit balance sheet, the way SwiftMart's did, until it's a reflex. And the number you carry forward is unit ROIC. Unit margin won't do, because, as you learned with returns on capital, a margin without capital is half a fraction.

The meta-rule. Every game inflates the unit story at the moment the company story can't yet stand on GAAP numbers. So calibrate your trust by distance from GAAP. Unit economics reconciled to audited statements deserve real weight. DMart's stores and Chipotle's restaurant-level margins with rent in are examples. Unit economics that exist only in a deck, with a large unexplained bridge to the filed loss, are marketing. The only version you ever fully trust is the one you rebuild yourself, from raw filed data. And so the gate is three rebuilds.

Common mistakes & how experts think differently

Mistake 1: computing LTV on revenue. It's the commonest error in the wild, and it inflates LTV by 1/margin (3–5×). Expert reflex: the phrase "lifetime value" triggers the question "of contribution, discounted at what?" before any number gets accepted.

Mistake 2: lifetime = 1/churn, from blended churn, on a young base. Blended churn over-weights fragile new cohorts, which understates mature lifetime. And young companies have no mature cohorts at all, so any lifetime claim they make is unfounded. Expert reflex: no triangle, no lifetime. Churn comes from the flat part of old cohorts' curves, or it doesn't get used.

Mistake 3: skipping the discount rate. "It's just a metric, not a valuation," goes the excuse. But the 3× LTV/CAC threshold was calibrated on discounted LTV, and feeding it undiscounted numbers halves the standard without anyone noticing. Expert reflex: m·r/(1+d−r), always, with the naive number recomputed alongside so you can measure how much the deck's version flattered.

Mistake 4: treating contribution margin as the destination. Being CM3-positive doesn't make you profitable. Fixed costs and capital still exist, which is what the break-even N* and game 5, the missing balance sheet, are about. The reverse holds too. CM2-negative at the order level isn't automatically fatal if density mechanically fixes it, and P4 makes you decide that very question. Expert reflex: every contribution number gets placed straight away against the fixed-cost altitude it must climb and the capital it must service.

Mistake 5: comparing CM ladders across companies without reading definitions. One company's CM2 is another's CM3. One computes on net revenue, another on GOV. Expert reflex: pull the definitions footnote first and restate both companies onto one ladder. It's last week's lesson about bases, applied one level down.

Mistake 6: believing blended CAC while paid mix scales. Expert reflex: ask what the marginal customer cost, because the average one tells you little. And model organic as capacity-constrained, since it won't grow in proportion to spend.

Mistake 7: reading store economics off the P&L without the maturity mix. Margins "compressing" during a store-opening sprint, or "expanding" when growth stops, is arithmetic. It tells you nothing about performance. Expert reflex: split units by vintage before you judge the trend. That's the store version of cohort analysis.

Mistake 8: accepting unit metrics that don't reconcile. The deck says stores earn 12%, while the company files 4% and hand-waves the bridge. Expert reflex: run units × unit + corporate ≈ reported on every unit-economics claim, and treat the residual as the credibility score.

The deeper difference is posture. Amateurs consume unit economics. They take the metrics management hands them and admire or doubt them whole. Experts produce them. They rebuild from raw operational and filed data with their own definitions. Then they reconcile to GAAP, cut by cohort, stress-test.

The company's slide is the hypothesis, and your rebuild is the test. Use the practice set ahead to turn the rebuild into a 20-minute habit, so it stops being a special project.

Worked examples

Worked example 1: Auditing a pitch-deck LTV/CAC, lever by lever

"FitKit" (synthetic) sells a ₹1,000/month protein-snacks subscription box. Its Series-B deck claims "LTV ₹16,200 (45% gross margin × ₹1,000 ARPU × 36-month life, churn only 2.8%/month); CAC ₹2,800; LTV/CAC = 5.8×." Audit it with the abuse catalogue, one lever at a time.

Step 1: fix the margin (abuse #1, partially; the deck at least used a margin). The 45% is product-only gross margin. Cost every variable line per month: revenue 1,000 − product 500 − shipping 110 − payment gateway (2%) 20 − support & returns 50 = contribution of ₹320/month (32%). At the deck's own lifetime that's 320 × 36 = ₹11,520, so LTV has already fallen from 16,200 to 11,520.

Step 2: fix the lifetime (abuses #3 and #6). The 2.8% churn is measured on "active users", where active = opened the app. Read transacting churn from the flat part of cohorts older than a year and you get 4.5%/month, so lifetime = 1/0.045 = 22.2 months, against the deck's 36. Restated, that's 320 × 22.2 = ₹7,104, which takes LTV from 11,520 to 7,104.

Step 3: discount (abuse #2). Monthly survival r = 0.955. Use d = 14%/year for a young company's cohort forecasts, which is a monthly d ≈ 1.1%. LTV = m·r/(1 + d − r) = 320 × 0.955 / (1.011 − 0.955) = 305.6 / 0.056 = ₹5,457. Discounting alone took off another 23%, moving LTV from 7,104 to 5,457.

Step 4: fix the CAC (abuses #5 and #7). The deck's ₹2,800 = total marketing ₹5.6 cr ÷ 20,000 new subscribers. But 6,000 of those came through referral and organic. Paid spend of ₹5.04 cr bought 14,000 subscribers, a paid CAC of ₹3,600. Add the first-box 50% discount (₹400) and the welcome kit (₹200) that the deck expensed elsewhere, and fully-loaded marginal CAC ≈ ₹4,200.

Step 5: the honest ratio, and the cash check. LTV/CAC = 5,457 / 4,200 = 1.3×. Against the deck's 5.8×, that's a 4.5× overstatement, and there's no single lie anywhere in it. Four flattering choices just multiplied. Payback = 4,200 / 320 = 13.1 months, against the consumer-subscription bar of 6–12. Every new cohort is a year-plus loan to the customer, funded by the Series B.

Verdict. At ~1.3× discounted LTV/CAC and a 13-month payback with churn this high, growth spends value. Fix retention before scaling spend, since the lifetime lever moved the number most. Notice the shape of the audit: margin, lifetime, discounting, CAC. You'll run it in that order for the rest of your career.

Worked example 2: A café chain's store economics: the cascade from deck to filed number

"BrewNation" is synthetic, modelled on Indian café/QSR scale, and leases all its sites. Cross-check it against Devyani, Sapphire Foods, or Westlife disclosures. Its deck says "Store-level EBITDA margin 31%." Its filed consolidated EBITDA margin, on a pre-Ind AS 116 basis, is 1.6%. Both are true, so build the cascade that connects them.

Step 1: one mature café, rent where it belongs (₹ lakh/year):

Step 2: the Ind AS 116 version (game 1, allocation). Since 2019 rent leaves opex and becomes ROU depreciation + lease interest. So the reported "store EBITDA" = 43.2 + 31.2 = ₹74.4 lakh = 31.0%, a 72% optical inflation with zero change in cash. That's the deck's 31%. Always restate with rent in before you judge a leased retailer's unit.

Step 3: the maturity mix. The chain runs 36 mature cafés (₹240 lakh revenue @ 18.0%) + 24 young ones (avg ₹150 lakh @ ~4.0% = ₹6 lakh each). Chain revenue = 8,640 + 3,600 = ₹12,240 lakh. Store EBITDA = 1,555.2 + 144 = ₹1,699.2 lakh, a blended store margin of 13.9% against 18.0% for a mature café. Those 4.1 points are pure vintage mix, the same effect SwiftMart showed.

Step 4: corporate, and the reconciliation. Corporate fixed costs are ₹1,500 lakh, so company EBITDA (rent-in basis) = 1,699.2 − 1,500 = ₹199.2 lakh = 1.6%. Laid end to end, the cascade runs 31.0% (deck) → 18.0% (rent in) → 13.9% (all vintages) → 1.6% (corporate in). That's 29 points of margin evaporating between the slide and the filing, and not one line along the way is dishonest. Break-even store count N* = 1,500 / 43.2 = 35 mature-equivalent stores. Right now the chain runs 1,699.2/43.2 ≈ 39 mature-equivalents, barely past corporate breakeven. That's what the 1.6% is telling you.

Step 5: the unit's balance sheet and payback (game 5, the missing balance sheet). Per store, fit-out + equipment is ₹110 lakh, deposits ₹15 and working capital ₹5, for a total of ₹130 lakh. On a mature basis, EBIT = 43.2 − 13.75 (fit-out ÷ 8 yr) = 29.45. Post-tax + D&A ≈ ₹35.8 of cash a year, which pays back in ~3.6 years. But stores don't open mature. On the actual ramp (Y1 revenue 144 @ ~2%, Y2 204 @ ~11%, mature Y3), cumulative pre-tax store cash runs 2.9, 25.3, 68.5, 111.7, 154.9. So the ₹130 lakh is repaid part-way through year 5. Interpolate instead of counting whole years: 4 + (130 − 111.7)/43.2 = 4.4 years pre-tax.

Post-tax, the ₹13.75 lakh of fit-out depreciation shields tax at 25% (₹3.44 lakh) in every year the store earns an operating profit. It shields nothing in the year-1 loss. The cumulative post-tax run is 2.9, 23.1, 59.0, 94.8, 130.7, crossing at 4 + (130 − 94.8)/35.8 = about 5.0 years.

So here's a real business with workable stores. Its own deck prices it as a 31%-margin machine and its filings as a 1.6% one. Your job is to hold the whole cascade in one head, and to check next year whether the mature-store line moved, since it's the only structural number in it.

Worked example 3: A revenue cohort triangle, end-to-end: NRR, observed payback, honest LTV

"CloudDesk" (synthetic SME SaaS) acquired a January-2024 cohort of 200 customers at ₹60,000 fully-loaded CAC each, a cohort CAC of ₹120 lakh. Starting MRR is ₹4,000 per customer and gross margin is 80%. Here's the cohort's revenue row, in ₹ lakh of MRR, with month 1 as the first billed month:

At month 12, the logo count is 168 of 200 (84%).

Step 1: read the row. MRR dips at the start because churn bites early, with the leavers going first. Then it bottoms at M3 and grows past its starting point, as expansion among the survivors outruns the leak. Per-survivor MRR at M12 = 8,64,000/168 = ₹5,143, against ₹4,000 at entry. That's +29% expansion among the stayers, while 16% of logos left.

Step 2: NRR, straight from the exhibit (as promised with the revenue triangle). Cohort NRR (12-month) = M12 revenue ÷ M1 revenue = 8.64 / 8.00 = 108%. You don't have to trust anyone's definitions or believe anyone's deck. It's the ratio of two cells. Logo retention of 84% alongside revenue retention of 108% is the classic signature of healthy SaaS. Small customers leave, and the survivors expand.

Step 3: observed payback. Cumulative contribution = 80% × cumulative MRR. The row sums to ₹98.19 lakh, so cumulative contribution at M12 = ₹78.55 lakh against CAC of ₹120 lakh. That's 65% repaid at month 12. Contribution is running at ~₹7 lakh/month and rising ~1%/month, so the remaining ₹41.45 lakh clears in ~6 months. Observed payback ≈ month 18. Cross-check with the formula: 60,000 ÷ (4,000 × 0.8) = 18.75 months, which agrees. For SaaS that's workable but not elite, inside the 12–24 band.

Step 4: LTV, with expansion credited only as far as the data supports. Per original customer, year-1 contribution ≈ ₹38,400 (78.55 lakh ÷ 200, rounded to the clean run-rate). Credit the observed 8%/year net revenue growth for three years. Then fade to 96%/year revenue retention, so there's no net expansion forever, and discount at 12%:

• Years 1–3: PV = 38,400/1.12 + 41,472/1.2544 + 44,790/1.4049 = 34,286 + 33,061 + 31,880 = ₹99,227
• Year 4 onward (r = 0.96): value at end-year 3 = 44,790 × 0.96/(1.12 − 0.96) = ₹2,68,740 → PV = 268,740/1.4049 = ₹1,91,283
• LTV ≈ ₹2.9 lakh, so LTV/CAC = 2,90,510 / 60,000 ≈ 4.8×

Step 5: what the naive version would have said. Feed NRR = 108% into the perpetuity as if it ran forever: LTV = 38,400 × 1.08/(1.12 − 1.08) = ₹10.37 lakh, or 17.3×. That's 3.6× the faded answer. Had NRR been 112%+, the denominator would hit zero and the "formula" would output infinity, which is abuse #4 in the flesh. The faded 4.8× is the real finding. A good business didn't need the inflation, and that's a screening rule in itself. When a deck reaches for the perpetual-expansion version, ask what the faded version shows. If they won't compute it, you just did.

Worked example 4: Building a subscriber unit honestly, end to end, and the float that funds it

"Cloudline" is synthetic, modelled on a US vertical-SaaS business, so pull the actual 10-K/S-1 and rebuild it. Think of it as the constructive twin of Worked example 1. There we tore a deck's number down. Here we build one up from raw data, in the order you learned it, from CAC through payback, and then ask the question the ratio can't answer. When does the cash come back?

Step 1: the unit and its contribution. The unit is the account. ARPU is $6,000/year and gross/contribution margin is 80%, so contribution m = $4,800/year = $400/month. Serving cost, meaning hosting, payments and support, is the other 20%. Core R&D is fixed and stays below the line.

Step 2: fully-loaded CAC. The year's acquisition spend was performance marketing + inside-sales payroll including SBC + onboarding = $7.2m, and it signed 1,200 new accounts. So fully-loaded CAC = 7,200,000 ÷ 1,200 = $6,000. A "CAC" that counted only the ad line would be fiction, because for sales-led SaaS, payroll usually dwarfs the ad budget.

Step 3: payback. CAC ÷ monthly contribution = 6,000 ÷ 400 = 15 months. That puts it inside SaaS's 12–24 "workable" band, short of the sub-12 "great" one.

Step 4: discounted LTV. Mature-cohort annual logo churn is 12%, so r = 0.88. Discount at d = 11%, because a young book's retention forecast is riskier than the firm's WACC floor. LTV = m·r/(1 + d − r) = 4,800 × 0.88 ÷ (1.11 − 0.88) = 4,224 ÷ 0.23 = $18,365. The naive m/churn = 4,800 ÷ 0.12 = $40,000, a 118% overstatement. That's the survival-and-discounting correction a deck would skip without telling you.

Step 5: the ratio, then the question it cannot answer. LTV/CAC = 18,365 ÷ 6,000 = 3.06×, which clears the folk 3× bar with nothing inflated. Now the cash reality, using the float identity from payback. Cloudline spends $600,000/month acquiring at a 15-month payback, so it permanently floats 600,000 × 15 ÷ 2 ≈ $4.5m of not-yet-repaid CAC. Picture its twin, identical at 3.06× LTV/CAC but with stickier, longer-lived enterprise customers bought at a 30-month payback. The twin floats 600,000 × 30 ÷ 2 = $9.0m, double the financing dependence for the same wealth-creation ratio.

So use the pair, always. 3.06× says the customer is worth building. 15 months says Cloudline can build it on its own cash flow far more safely than the twin can. In 2022, that distinction decided which of two "3× businesses" lived.

Worked example 5: A D2C contribution ladder, and the repeat rate hiding inside CM3

"AyurGlow" is synthetic, modelled on the scale and shape of a listed Indian D2C beauty brand like Honasa/Mamaearth, so pull the real DRHP and rebuild with their ladder. You'll take one delivered order, then do the decomposition the CM ladder section warned you never to skip.

Step 1: the CM ladder.

A CM3 of 17% looks like a healthy D2C order. Don't admire it yet.

Step 2: split the marketing line (the new/repeat trap, live). Over 100 orders, the ₹300/order blends two utterly different rupees. Suppose 40 are first orders from newly acquired customers and 60 are repeat. Total marketing = ₹30,000. New-customer CAC = ₹650 each (40 × 650 = ₹26,000). The remaining ₹4,000 is repeat-order vouchers, which comes to ₹66.67 per repeat order.

• First-order CM3 = 504.4 − 650 = −₹145.6. Acquisition loses money on the first order, as it should, since you're buying a stream.
• Repeat-order CM3 = 504.4 − 66.67 = +₹437.7. A repeat order is enormously profitable.
• Weighted: 0.4 × (−145.6) + 0.6 × 437.7 = ₹204.4, which reconciles to the blended CM3 with nothing left over.

Step 3: the number the whole order economics rides on. Set blended CM3 = 0 and solve for the repeat share: f × 437.7 + (1 − f) × (−145.6) = 0 → f = 25.0%. Above a 25% repeat share the order book is contribution-positive. Below it, AyurGlow rents revenue at a loss. At today's 60% it's comfortable, but the entire "17% CM3" thesis is a bet that repeat behaviour holds, and only the cohort triangle can prove that. Acquisition itself pays back fast. CAC ₹650 ÷ CM2 ₹504.4 = 1.29 orders, so it's recovered inside order #2, if order #2 happens.

The lesson, one level below the ladder itself. A blended CM3 is a weighted average of a loss on acquisition and a fat margin on repeats. Quoting it without the new/repeat split is the contribution-margin twin of quoting blended CAC while scaling paid. Decompose before you believe.

Practice set

Work in order: P1–P3 guided, P4–P10 independent, P11–P12 timed. P4, P7, and P10 are the gate. They're the three unit-economics rebuilds the master map asks you to complete against the key. The app's rebuild engine grades them line by line, and on paper you grade yourself against the solutions, every line. All companies are synthetic unless noted, in ₹ and at 25% tax unless stated. Give each problem a real attempt before you read any Solution.

P1 (guided: choose the unit). For each business, name the unit(s) of analysis using the three tests, and apply the two-layer rule where it's needed. Then state the unit's headline KPI. (a) An online apparel retailer. (b) A hospital chain. (c) A microfinance lender. (d) A cloud-kitchen operator running 8 brands from 40 kitchens. (e) A gym chain selling monthly memberships. (f) An IT-services firm.

Solution. (a) The order (flow) for contribution, because the full revenue-and-cost loop (product margin, shipping, returns) attaches per order. Add the customer (stock) for CAC and repeat behaviour. Track CM2 per order + orders/customer/year. (b) The occupied bed-day (flow), since ARPOB and variable clinical costs attach per bed-day. Add the hospital/bed (stock), since growth capital is spent per bed. Track occupancy against its breakeven, and EBITDA per bed. (c) The loan, or ₹1 of AUM, since spread, credit cost, and opex attach per rupee lent. Track NIM − credit cost − opex per rupee of AUM. (d) The order (flow) + the kitchen (capacity stock). It's the kitchen that carries the rent and the idle capacity, and a per-order view alone hides it, the same trap as the dark store. Watch orders/kitchen/day against kitchen breakeven. (e) The member (stock, the revenue and churn unit) + the club (capacity stock, the capex and fixed-cost unit). Track members per club against club breakeven, and member churn. (f) The billed hour, with the hirable engineer as the stock unit. Watch utilisation against breakeven. Notice the pattern. Every capacity business needs the flow unit and the box the flow runs through. Every acquisition business needs the flow unit and the customer who repeats it.

P2 (guided: churn algebra). A SaaS company has monthly logo churn of 2.2%. (a) Find annual logo churn. (b) Find expected customer lifetime in months and in years, computed both from the monthly and from the annual rate, and explain why they differ. (c) Separately, revenue is ₹100 cr at the start of the year. Customers who left took ₹9 cr, downgrades cost ₹3 cr, expansion added ₹15 cr and new customers added ₹22 cr. Compute GRR, NRR, and closing revenue. (d) There were 1,000 customers at the start, and 150 left. Compare logo churn with revenue churn, and interpret the difference.

Solution. (a) Annual = 1 − (1 − 0.022)¹² = 1 − 0.978¹² = 1 − 0.766 = 23.4%. It's never 2.2 × 12 = 26.4%. (b) From monthly, 1/0.022 = 45.5 months ≈ 3.8 years. From annual, 1/0.234 = 4.3 years. They differ because annualising compounds the leak into one year-end step and throws away the timing within the year. The finer period gives the better estimate, so use the granularity in which churn is actually measured. (c) GRR = (100 − 9 − 3)/100 = 88%. NRR = (100 − 9 − 3 + 15)/100 = 103%. Closing revenue = 103 + 22 = ₹125 cr. (d) Logo churn = 15% and revenue churn = (9 + 3)/100 = 12%. Revenue churn < logo churn, so the leavers were small and the leak is at the shallow end, which is benign. The reverse pattern, revenue churn > logo churn, is the early-warning sign that the whales are leaving.

P3 (guided: LTV three ways). An OTT service has a sticker price of ₹149/month. 60% of subscribers pay through app stores at a 15% commission. Variable serving cost (delivery, payments, support) is ₹25/month, monthly churn is 4%, and the discount rate is 12%/year. The company's deck claims "LTV ₹3,725 = ARPU × 25-month lifetime." Compute (a) net ARPU and contribution per month, (b) undiscounted margin-based LTV, (c) discounted LTV, and (d) the deck's total overstatement.

Solution. (a) Net ARPU = 149 × (0.4 + 0.6 × 0.85) = 149 × 0.91 = ₹135.6 ≈ ₹135, because the app-store toll comes off the top. Contribution = 135 − 25 = ₹110/month. (b) Lifetime = 1/0.04 = 25 months, so undiscounted LTV = 110 × 25 = ₹2,750. (c) Monthly, r = 0.96 and d ≈ 0.95%/month (1.12^(1/12) − 1). LTV = 110 × 0.96/(1.0095 − 0.96) = 105.6/0.0495 ≈ ₹2,134. (d) 3,725/2,134 = 1.75× overstated. Break it down. Moving from revenue to contribution cuts 3,725 to 2,750 (−26%), and discounting cuts 2,750 to 2,134 (−22%). Two "small" flatteries compound into a 75% one. And look at the deck's base. ₹149 × 25 uses the sticker price, so it counts the app-store commission the company never receives. That's three levers, none outrageous on its own, and together they describe a different business.

P4 (independent · REBUILD 1 of 3: the dark store). "ZipNow" is a quick-commerce business, and this is one of its dark stores today, all raw data. AOV is ₹420 and product margin 18%. It charges a delivery fee of ₹15/order and earns ad/monetisation income of ₹10/order. Current volume is 300 orders/day (26 days/month). Riders cost ₹25,000/month each, and a rider can do 30 orders/day, but coverage requires a minimum of 16 riders on shift whatever the volume. Packaging is ₹8/order, the payment gateway takes 1.6% of AOV, support costs ₹6/order, and coupons currently run at ₹30/order. Store fixed costs (rent, power, manager, pickers) are ₹6.0 lakh/month, and store capacity is ~1,200 orders/day. Management claims "CM2 is negative today only because of density; at 900 orders/day the store prints money." Rebuild the following. (a) Per-order CM2 today. (b) The store's monthly P&L today. (c) Per-order CM2 at 900 orders/day holding coupons at ₹30, the mechanical case, and the store P&L. (d) The same if coupons fall to ₹15, the behavioural case. (e) Break-even orders/day in both cases. (f) A verdict on management's claim.

Solution. (a) Gross per order = 420 × 18% + 15 + 10 = 75.6 + 25 = ₹100.6 ≈ ₹100. Riders today = max(16, 300/30) = 16, so the floor binds. That's 16 × 25,000 = ₹4.0 lakh ÷ 7,800 orders = ₹51.3/order. Variable = 51.3 + 8 + 6.7 (1.6% × 420) + 6 + 30 = ₹102, so CM2 ≈ −₹2/order. (b) Store P&L = 7,800 × (−2) − 6,00,000 = −₹6.16 lakh/month. (c) At 900/day, riders = 900/30 = 30. That's ₹7.5 lakh ÷ 23,400 = ₹32/order, so variable = 82.7 and CM2 = +₹17.3. Store P&L = 23,400 × 17.3 − 6,00,000 = 4,04,820 − 6,00,000 = −₹1.95 lakh/month, so it's still losing. (d) With coupons at ₹15, CM2 = +₹32.3. Store P&L = 23,400 × 32.3 − 6,00,000 = +₹1.56 lakh/month. (e) Breakeven = 6,00,000 ÷ CM2 ÷ 26. In the mechanical case that's 6,00,000/17.3 = 34,682/month = 1,334/day, above the store's 1,200 capacity. In the behavioural case it's 6,00,000/32.3 = 18,576/month = 714/day, which is reachable. (f) Verdict: the claim is half true, and the true half isn't the mechanical one. Density really does fix the rider line, taking it from 51 to 32, and that part is arithmetic. But density alone can't carry the store to breakeven within its own capacity. Profitability hinges on cutting coupons from ₹30 to ₹15 while keeping 900 orders/day. That's a bet on habit formation, with competitors voting. So separate what scale does mechanically from what management hopes customers will tolerate. Fund the first. Underwrite the second only with cohort evidence. Do repeat customers already order at low coupon levels?

P5 (independent: CAC three ways). Over one quarter, a consumer app spends ₹18 cr on performance ads, which are credited with 150,000 new transacting customers. It spends ₹6 cr on a brand campaign. Referral credits of ₹2 cr bring 40,000 referred customers, and organic/SEO brings 110,000. Every new customer gets a first-order discount of ₹150. Contribution per active customer is ₹95/month. Compute (a) paid CAC, (b) blended CAC, (c) fully-loaded blended CAC, (d) fully-loaded marginal paid CAC, (e) payback months on (c) and (d), and (f) state which CAC answers which question.

Solution. (a) Paid = 18 cr/150,000 = ₹1,200. (b) Blended = (18 + 6 + 2) cr/300,000 = ₹867. (c) Fully-loaded blended = (26 + 4.5) cr/300,000 = ₹1,017. (d) Marginal paid, fully loaded = (18 cr + 150,000 × ₹150)/150,000 = (18 + 2.25) cr/150,000 = ₹1,350. That adds its share of discounts and leaves out brand spend, because brand spend can't be attributed per customer. Flag that as the judgement call it is. (e) Payback is 1,017/95 = 10.7 months blended and 1,350/95 = 14.2 months marginal. (f) Blended fully-loaded CAC answers the cash question, what this quarter's growth costs the treasury per customer added. Marginal paid fully-loaded CAC answers the growth decision, what the next tranche of spend will buy, and it's the only CAC that matters for "should we accelerate". Quoting (b) while deciding with the economics of (d) is the blended-CAC trap in one problem. The next 150,000 customers cost ₹1,350+ each while the blend says ₹867, and they'll cost more still if the paid channel's auction steepens.

P6 (independent: read a triangle with three stories in it). These are monthly cohorts, and each cell shows the % of the cohort still transacting at that age in months:

One month in this calendar contained a price increase. (a) Which calendar month was it, and what's the evidence? (b) Corrected for that event, is there vintage deterioration? (c) Give the habit floor, and the mature monthly churn you'd feed into an LTV. (d) Name the single mistake an analyst would make reading only column M1.

Solution. (a) April. Follow the diagonal. Jan's M3 (April) = 38 breaks its own curve's flattening, then recovers to 39. Feb's M2 (April) = 41 dips below its trend, then steadies. Mar's M1 (April) = 51 is the worst M1 in the table, yet Apr's and May's M1 (May, June) bounce back to 53. A stripe that cuts across cohorts at different ages in the same calendar month, followed by partial recovery, is an event. Decay doesn't behave like that. (b) Corrected, the M1s read ≈ 58, 57, ~56, ~55, ~54, a mild drift of ~1 point per cohort. That's real but slow vintage softening, consistent with cohort sizes growing from 10k to 16k, since the marginal customer is slightly weaker. (c) Rows flatten near ~39–40%, which is the habit floor. The flat-part decay (40, 39, 39) implies mature churn ≈ 1–3%/month, and that figure is what feeds the LTV, whatever the blended rate says. (d) Read M1 alone and the drop from 58 to 51 at March looks like collapse, though it's the price hike landing on that cohort's first month. The 53 after it looks like recovery to a worse plateau. If you don't separate the diagonal from the column, you'll re-rate the machine for what was a one-time event. You'd be selling the structural story on cyclical evidence, and reading the triangle in three directions exists to prevent that confusion.

P7 (independent · REBUILD 2 of 3: the SaaS subscriber). "LedgerIQ" is an SME accounting SaaS. Its FY25 raw data: opening customers 8,000, closing 11,000, gross new customers signed 4,200. ARR is ₹36,000 per customer (flat pricing) and gross margin is 78%. On spend, performance marketing is ₹7.2 cr and inside-sales payroll ₹8.4 cr, which includes ₹1.2 cr of SBC. The onboarding team costs ₹1.8 cr and channel-partner commissions ₹1.5 cr. R&D + G&A (fixed) come to ₹32 cr. Within existing customers, GRR is 91% and expansion adds +14 points, for NRR of 105%. The discount rate is 13%. Rebuild the subscriber unit end to end. (a) Logo churn. (b) Fully-loaded CAC. (c) Contribution per customer and CAC payback. (d) Discounted LTV two ways, a logo-survival base case (r = 0.85) and a revenue-retention case (r = 0.91), plus LTV/CAC for each. (e) Break-even customer count N*, and the company's run-rate EBITDA ex-growth-spend. (f) A three-line verdict.

Solution. (a) Churned logos = 8,000 + 4,200 − 11,000 = 1,200, which is 15% of opening (annual). (b) CAC = (7.2 + 8.4 + 1.8 + 1.5) cr ÷ 4,200 = 18.9 cr/4,200 = ₹45,000. SBC stays in, because it's sales compensation. Onboarding stays in too, since it exists only because a customer was acquired. (c) Contribution = 36,000 × 0.78 = ₹28,080/year = ₹2,340/month, so payback = 45,000/2,340 = 19.2 months. That's workable for B2B SaaS, though not elite. (d) Base case: LTV = 28,080 × 0.85/(1.13 − 0.85) = 23,868/0.28 ≈ ₹85,200, so LTV/CAC ≈ 1.9×. In the revenue-retention case, crediting the observed net expansion with r = 0.91, it's 28,080 × 0.91/(1.13 − 0.91) = 25,553/0.22 ≈ ₹1,16,200, or ≈ 2.6×. The defensible range is 1.9–2.6×, below the 3× bar. (e) N* = 32 cr ÷ 28,080 = 11,396 customers. At 11,000 closing, run-rate contribution = 11,000 × 28,080 = ₹30.9 cr against ₹32 cr fixed, so run-rate EBITDA ex-S&M ≈ −₹1.1 cr. FY EBITDA including the ₹18.9 cr of S&M ≈ −₹24 cr, on avg-base contribution of ~₹26.7 cr. (f) Verdict: the machine nearly covers its fixed costs, and each new customer repays in ~19 months at ~2× lifetime return. That's fundable but thin. The thesis lives or dies on NRR holding ≥105% and CAC not inflating as the cheap SME channels saturate. Watch those two lines, quarterly.

P8 (independent: utilisation breakeven). "Sanjeevani Care" runs a 250-bed hospital. Fixed costs are ₹130 cr/year (staff, leases, equipment depreciation). ARPOB is ₹42,000 per occupied bed-day, and variable cost per occupied bed-day (drugs, consumables, case-fee clinicians) is ₹19,000. (a) Break-even occupancy. (b) EBITDA at 72% and at 65% occupancy. (c) The value of one occupancy point, and the operating margin of safety at 72%. (d) Why does a hospital's earnings sensitivity look more like a hotel's than a supermarket's?

Solution. (a) Contribution per occupied bed-day = 42,000 − 19,000 = ₹23,000. Available bed-days = 250 × 365 = 91,250, so fixed cost per available bed-day = 130 cr/91,250 = ₹14,247. Break-even occupancy = 14,247/23,000 = 61.9%. (b) At 72%, 91,250 × 0.72 × 23,000 = ₹151.1 cr − 130 = ₹21.1 cr. At 65%, it's 136.4 − 130 = ₹6.4 cr. Seven points of occupancy ≈ 3.3× the EBITDA. (c) One point = 912.5 bed-days × 23,000 ≈ ₹2.1 cr, and nearly all of it is profit. Margin of safety = (72 − 61.9)/72 ≈ 14%. (d) A hospital sells perishable capacity. An empty bed-night expires worthless while its fixed costs keep running, so profit is the thin wedge between two large numbers, occupancy × contribution against fixed cost. That's the hotel/airline shape from the utilisation section. A supermarket's costs travel with its sales, since COGS of ~85%+ of revenue is variable, so its EBIT breathes gently. Revenue growth can be the same while the earnings physics are violently different, so Phase 5 reads hospitals on occupancy and ARPOB before anything else.

P9 (independent: the allocation game, caught). "StyleLoft" is a fashion retailer, all leased. Its deck says "180 stores; store-level EBITDA margin 19%; corporate overhead only ₹90 cr." The filings show revenue of ₹2,700 cr and consolidated EBITDA (post-Ind AS 116) of ₹378 cr, with ROU depreciation attributable to stores of ₹130 cr and lease interest of ₹60 cr. (a) Run the reconciliation test on the deck's claim. (b) Restate store economics with rent in. (c) List the two most likely explanations for the residual, and the disclosure you'd demand.

Solution. (a) The deck implies store EBITDA = 2,700 × 19% = ₹513 cr. Take off corporate's ₹90 cr and the implied consolidated figure is ₹423 cr, against ₹378 cr reported. That's a residual of ₹45 cr ≈ 12% of reported EBITDA, above the 10–15% suspicion line, so the unit story doesn't reconcile. (b) Proxy rent = 130 + 60 = ₹190 cr, which is the Ind AS 116 pieces put back together. Rent-in store EBITDA = 513 − 190 = ₹323 cr, or 12.0% of revenue against the 19% claimed. One-third of the "store margin" was accounting presentation. (c) There are two likely residuals. (i) "Store-level" leaves out pre-opening costs, warehousing, or last-mile that belong to stores. (ii) The 19% is mature stores only, while a fifth of the fleet is immature, which is blended masking. Demand the definition footnote for "store-level EBITDA," the store-vintage split, and a bridge from store EBITDA to consolidated. A company confident in its stores publishes all three. DMart effectively does, and that's the tell.

P10 (independent · REBUILD 3 of 3: the store chain, from annual-report-style data). "AnnaMart" is a grocer. Its FY25 raw data shows 120 stores at year-end, up from 100 at the start. Revenue is ₹9,240 cr, of which the 85 mature stores (open > 3 years) did ₹8,500 cr and the 35 younger ones ₹740 cr. Gross margin is 14.8% across the board. Store-level opex is ₹485 cr for mature stores and ₹90 cr for young ones. Corporate costs are ₹185 cr, and reported company EBITDA is ₹607 cr. A new store costs ₹42 cr (owned property + fit-out) plus ₹6 cr of working capital. On the ramp, Y1 revenue is ₹45 cr @ 2% store EBITDA margin, Y2 ₹70 cr @ 5%, Y3 ₹88 cr @ 7.5%, and mature ₹100 cr. Rebuild the following. (a) The mature-store unit P&L and margin. (b) The young-store margin, and the maturity-mix bridge to the blended store margin. (c) The reconciliation test against reported EBITDA. (d) New-store payback (pre-tax, and state the convention) and mature-store pre-tax ROCE. (e) A two-line verdict against the SwiftMart template.

Solution. (a) Mature revenue/store = 8,500/85 = ₹100 cr, so gross profit = 14.8 cr. Store opex = 485/85 = 5.71 cr, leaving store EBITDA of ₹9.1 cr = 9.1%. (b) Young gross = 740 × 14.8% = 109.5. Minus 90, that's ₹19.5 cr over 35 stores, a 2.6% margin (₹0.56 cr/store). Blended store EBITDA = (773.5 + 19.5) = ₹793 cr on 9,240 = 8.6%. That's half a point of drag purely from vintage mix, the same effect as SwiftMart's, and it's mild here because young stores are only 8% of revenue. (c) Reconciliation: 793 − 185 = ₹608 cr against reported ₹607, a residual of 0.1%. The unit story is real. Remember how this feels, because it's what a clean bridge looks like. (d) Mature-store EBITDA is ₹9.1 cr and D&A ≈ ₹1.7 cr, so EBIT is ₹7.4 cr. Capital = 42 + 6 = ₹48 cr, so pre-tax ROCE ≈ 15.4% (post-tax ≈ 11.6%). For payback on the ramp, use pre-tax store EBITDA as the cash proxy, and say so. Cumulative cash runs 0.9, 4.4, 11.0, 20.1, 29.2, 38.3, 47.4, 56.5, crossing ₹48 cr early in year eight. Interpolating gives 7 + (48 − 47.4)/9.1 = about 7.1 years pre-tax, and about 8.3 years post-tax once the ₹1.7 cr of depreciation starts shielding tax from year two. That's the owned-property trade you saw with SwiftMart, slow payback bought for permanent cost immunity. (e) Verdict: a real DMart-shaped machine, since the units reconcile to the filed P&L within rounding, but a slightly weaker one, at 9.1% mature margins and ~15% pre-tax store ROCE against the template's 9.5%/19.6%. The next question is whether that gap comes from site quality, scale, or execution, and that's territory for industry structure and moats in the weeks ahead.

P11 (timed · 12 minutes, the honesty audit sprint). A quick-commerce DRHP's "unit economics" page makes six claims. Name the game, from the LTV/CAC abuse catalogue or the five game families, and state the direction of the distortion. (i) "LTV of ₹18,000 = ARPU ₹500 × 36 months". (ii) "CAC of ₹350 (marketing spend ÷ new app registrations)". (iii) "our year-3 customers spend 2.1× their year-1 levels", on a chart that begins at month 3. (iv) "contribution margin 18%, excluding last-mile delivery, which we treat as a strategic investment in density". (v) "our mature dark stores are store-EBITDA positive". (vi) "monthly churn of just 2%", where churn = did not open the app for 90 days.

Solution. (i) Revenue LTV + undiscounted + unverified lifetime (abuses #1, #2, #3). It overstates 3–5×+, so demand the contribution-based, discounted, mature-cohort version. (ii) Denominator game (abuse #6/game 4). Registrations ≠ transacting customers, so CAC is understated, likely 2–4×. (iii) Survivor bias + cropped cliff (abuse #8/game 4). The 2.1× is computed only on those still alive at year 3, and the month-0–3 collapse is hidden, so it overstates cohort value. (iv) Allocation game (game 1). Last-mile is a per-order variable cost by any behavioural test, so "18%" could turn negative once restated. The word "strategic" is doing the accounting. (v) Blended masking, inverted, which is cherry-picking (game 2). The mature subset is shown and the bleeding majority isn't, so ask for the full vintage split and the share of stores that are "mature". (vi) Denominator game on churn. A 90-day-inactivity definition mechanically lowers churn. Transacting churn will be far higher, so the definition inflates lifetime and LTV. Score 1 point per correctly named game + ½ for direction. ≥5/6 is gate-level fluency.

P12 (timed · 10 minutes, mixed sprint). (a) Monthly churn is 4%. What's annual churn? (b) CAC is ₹2,400 and contribution ₹150/month. Find payback, and give a verdict against the SaaS bands. (c) Corporate fixed costs are ₹66 cr and contribution ₹5,500/customer/year. Find break-even customers. (d) A hotel's fixed cost per available room-night is ₹2,600 and contribution per occupied room-night ₹3,700. Find break-even occupancy. (e) An airline's yield is ₹5.20/RPK and CASK ₹4.42/ASK. Find the break-even load factor, and give one caveat on the formula. (f) With m = ₹2,000/year, r = 0.75, d = 12% and CAC ₹2,500, find the discounted LTV/CAC.

Solution. (a) 1 − 0.96¹² = 38.7%. It isn't 48%. (b) 2,400/150 = 16 months, inside the 12–24 workable band and outside the <12 elite band. That's fine if retention is proven and dangerous if capital is scarce. (c) 66,00,00,000/5,500 = 120,000 customers. (d) 2,600/3,700 = 70.3%. (e) BELF = CASK/yield = 4.42/5.20 = 85.0%, right in the low-cost-carrier band. One caveat applies. This convention treats all CASK as load-independent. If some costs vary per passenger, use BELF = fixed CASK/(yield − variable cost per RPK), which lowers the denominator and raises the true breakeven. (f) LTV = 2,000 × 0.75/(1.12 − 0.75) = 1,500/0.37 = ₹4,054, so LTV/CAC = 1.6×. Acquisition roughly returns its cost with a thin buffer, and nobody should scale spend on this until m or r moves.

P13 (guided: build the CM ladder, and catch a misclassified discount). One delivered order at a D2C brand has net revenue of ₹2,500 and product COGS of ₹1,500. Shipping is ₹140, packaging ₹40, the payment gateway 1.8% of net revenue, and per-order support ₹25. Variable marketing per order is ₹380. Separately, the brand books a ₹120/order "first-order welcome voucher" (redeemable against the bill) inside fixed brand marketing, below CM3. (a) Build the CM1/CM2/CM3 ladder as reported. (b) Reclassify the voucher by its behaviour under Ind AS 115/ASC 606 and recompute the true CM3. (c) What did the misclassification inflate, and by how much?

Solution. (a) CM1 = 2,500 − 1,500 = ₹1,000 (40.0%). Fulfilment = 140 + 40 + 45 (1.8% × 2,500) + 25 = ₹250, so CM2 = 1,000 − 250 = ₹750 (30.0%). CM3 = 750 − 380 = ₹370 (14.8%). (b) A voucher redeemable against the bill is consideration payable to a customer, so it's contra-revenue, and marketing is the wrong home for it. True net revenue = 2,500 − 120 = ₹2,380, CM1 = 880 and CM2 = 630, which makes true CM3 = ₹250 (10.5% of true net revenue). (c) Parking a variable discount in fixed costs left CM1, CM2, and CM3 each ₹120 too high. The reported 14.8% CM3 is really 10.5%. Same cash, better-looking ladder. That's game 3 from the games section, and you catch it by classifying every cost by behaviour before you believe any rung.

P14 (guided: LTV/CAC and payback, with a verdict). A US B2B SaaS has ARPU of $1,200/year and a contribution margin of 75%. Mature annual churn is 15%, the discount rate is 12%, and fully-loaded CAC is $1,500. Compute (a) monthly contribution and CAC payback, (b) discounted LTV, (c) the naive m/churn LTV and the overstatement, and (d) the discounted LTV/CAC, with a one-line verdict against the folk bar.

Solution. (a) m = 1,200 × 0.75 = $900/year = $75/month. Payback = 1,500 ÷ 75 = 20 months, inside the 12–24 workable band but toward the slow end. (b) r = 0.85, and LTV = 900 × 0.85 ÷ (1.12 − 0.85) = 765 ÷ 0.27 = $2,833. (c) Naive = 900 ÷ 0.15 = $6,000, a 2.12× overstatement from skipping survival-weighting and discounting. (d) LTV/CAC = 2,833 ÷ 1,500 = 1.89×, below the 3× bar. Acquisition roughly returns its cost with a thin buffer, and the 20-month payback makes every cohort a near-two-year loan. Lift retention or contribution before scaling spend. Accelerating into a sub-3× ratio gets it backwards.

P15 (independent: the unrecovered-CAC float). A consumer-subscription company spends ₹0.9 cr/month on acquisition at an 8-month CAC payback. (a) Estimate the steady-state stock of not-yet-repaid CAC it floats. (b) It doubles monthly acquisition spend (payback unchanged). (c) Instead, from the original base, payback stretches to 24 months (spend unchanged). (d) Give one line each on what the float is and why it matters, plus one on the 2022 lesson.

Solution. (a) Float ≈ monthly acquisition spend × payback ÷ 2 = 0.9 × 8 ÷ 2 = ₹3.6 cr. Cohorts are at every stage of repayment, so on average each one carries half its window. (b) Doubling spend gives 1.8 × 8 ÷ 2 = ₹7.2 cr, so the float doubles. (c) Stretching payback gives 0.9 × 24 ÷ 2 = ₹10.8 cr, and the float triples. (d) The float is the working capital a company permanently lends its own customers, between paying to acquire them and being repaid. It matters because it has to be financed, so fast growth at long payback is a standing call on capital markets. 2022 taught the lesson. Two companies with identical LTV/CAC but 8- vs 30-month paybacks met very different fates the year cheap capital vanished. Payback priced survival, and the ratio didn't.

P16 (independent: the billed-hour utilisation breakeven). At an IT-services firm, a fully-loaded engineer costs ₹15 lakh/year and other direct costs are ₹9 lakh/year. There are ~2,000 billable hours available, at a realised rate of ₹2,400/hour. Treat both cost lines as fixed with respect to hours billed. Compute (a) engineer-level margin at 85% and at 75% utilisation, (b) the value of one utilisation point, (c) break-even utilisation and the operating margin of safety at 85%, and (d) explain why "hiring ahead of demand" is a margin warning.

Solution. (a) Fixed cost = 15 + 9 = ₹24 lakh. At 85%, revenue = 1,700 × 2,400 = ₹40.8 lakh, so margin is ₹16.8 lakh (41.2%). At 75%, revenue = 1,500 × 2,400 = ₹36.0 lakh, so margin is ₹12.0 lakh (33.3%). (b) One point = 20 hours × ₹2,400 = ₹48,000, and nearly all of it is margin, because the cost base doesn't move. (c) Break-even utilisation = 24,00,000 ÷ (2,000 × 2,400) = 50%. Margin of safety = (85 − 50)/85 = 41.2%. That comfortable gap makes IT services a structurally good business, set against an airline's wafer-thin one. (d) Hiring ahead of demand adds available hours (the denominator) before billed hours (the numerator), so utilisation, and with it margin, falls first. Headcount growth is a margin headwind until the bench fills, "a warning wearing a growth costume."

P17 (independent: blended CAC across two years). Take a consumer app. In Year 1, performance ads costing ₹8.0 cr acquire 100,000 paid customers, referral spend of ₹0.6 cr acquires 30,000, and organic brings 70,000 free. In Year 2 the company targets 300,000 new customers. Organic stays at 70,000, and referral stretches to 40,000 at the same ₹200 each. Paid has to cover the rest, and the marginal paid customer now costs ₹1,050. Compute (a) Year-1 paid and blended CAC, (b) Year-2 blended CAC, (c) organic's share of adds in each year, and (d) explain in a single sentence why blended CAC rose, naming the assumption a flat-blended-CAC guide would be smuggling in.

Solution. (a) Paid CAC = 8.0 cr ÷ 100,000 = ₹800. Blended = 8.6 cr ÷ 200,000 = ₹430. (b) Paid must deliver 300,000 − 70,000 − 40,000 = 190,000 at ₹1,050 = ₹19.95 cr. Referral is 40,000 × ₹200 = ₹0.8 cr. Blended = (19.95 + 0.8) cr ÷ 300,000 = ₹692. (c) Organic share = 70,000/200,000 = 35% in Y1, falling to 70,000/300,000 = 23.3% in Y2. (d) Blended CAC rose ~61%, from 430 to 692, because the mix shifted toward paid while the marginal paid customer costs more than the average, and no channel got worse. Organic can't scale on command. So any guide promising faster growth at flat blended CAC is assuming organic scales linearly with paid, and that's history's most expensive assumption.

P18 (timed · 12 minutes, the growth-acceleration decision). A D2C brand's board wants to triple performance-marketing spend. For the current cohort, fully-loaded paid CAC is ₹900 and contribution ₹250/customer/month. Mature monthly churn is 5%, and the discount rate is 15%/year (monthly d ≈ 1.171%). The marketing team estimates that the marginal cohort the extra spend would buy costs ₹1,600 to acquire and churns at 7%/month, because its channels are weaker. The brand spends ₹1.0 cr/month on acquisition today, and 3× would be ₹3.0 cr/month. For the current and the marginal cohort, compute (a) discounted LTV and LTV/CAC, (b) CAC payback, (c) the unrecovered-CAC float before and after the 3× decision, and (d) the call, in three lines, with the single condition that would change it.

Solution. (a) Current: r = 0.95, LTV = 250 × 0.95 ÷ (1.01171 − 0.95) = 237.5 ÷ 0.06171 = ₹3,848, so LTV/CAC = 4.28×. Marginal: r = 0.93, LTV = 232.5 ÷ 0.08171 = ₹2,845, so LTV/CAC = 1.78×. (b) Payback = 900 ÷ 250 = 3.6 months for the current cohort, against 1,600 ÷ 250 = 6.4 months for the marginal one. (c) Float = spend × payback ÷ 2. Now it's 1.0 × 3.6 ÷ 2 = ₹1.8 cr. After 3× at the marginal payback it's 3.0 × 6.4 ÷ 2 = ₹9.6 cr, a 5.3× jump in financed working capital. (d) The decision buys the marginal cohort (1.78×, 6.4-month payback), and the blended average (4.28×) is beside the point. At a sub-2× ratio the extra spend barely creates value, while the float balloons from ₹1.8 cr to ₹9.6 cr. So don't blanket-triple spend. Fund it only if cohort evidence shows the marginal channels' churn improving toward the base cohort's. Otherwise fix retention or AOV first, because the marginal customer prices growth, whatever the average one says. One decision, and it draws on everything from CAC through payback.

Applied mini-project

The Rebuild: one atom, two markets. Rebuild the unit economics of one Indian and one US company from primary filings, run the reconciliation test on each, and deliver a verdict. Think of it as the field version of the gate. P4/P7/P10 proved you can rebuild from clean data. This proves you can rebuild from the messy, partial disclosures real companies actually give you.

Choose one from each list. Pairs with the same unit type make the comparison sharpest:

• India: Eternal/Zomato (the order; quarterly shareholder letters disclose GOV, take rate, and contribution definitions), Swiggy (the order + the dark store; the RHP's dark-store disclosures), Nykaa (the order; DRHP and investor decks carry AOV, repeat, and CM-ladder exhibits), Avenue Supermarts (the store), Jubilant FoodWorks (the store), Dr Lal PathLabs (the patient/test).
• US: Chipotle (the restaurant; it discloses restaurant-level operating margin with rent in, the gold standard of honest unit disclosure), Netflix (the member; 10-K plus regional membership/ARM tables), Spotify (the premium subscriber), Costco (the warehouse + the member: two-layer), Planet Fitness (the club + the member).

Pull (all free): India: DRHP/RHP (the unit-economics and cohort chapters), the last four quarterly investor decks or shareholder letters, the annual report (revenue disaggregation, segment note, lease note for Ind AS 116 rent reassembly), screener.in for five-year context. US: 10-K (Item 7 MD&A unit disclosures, revenue note, lease note), the last two 10-Qs, investor-day deck, stockanalysis.com/TIKR for history. Both: two earnings-call transcripts (managements answer unit-economics questions in Q&A that never appear in decks).

Build, per company:

1. The unit, defended. Name the unit using the three tests, and apply the two-layer split if the business needs one. Note where your unit differs from the one management presents, and why theirs is a disclosure decision.
2. The unit P&L, with every line sourced to a page or slide. Go from revenue per unit, through full variable cost by behaviour, to contribution (CM1/CM2/CM3 where applicable), and then to the unit's share of its capacity layer's fixed costs. If Ind AS 116/ASC 842 moved rent, restate it into the store-level numbers, as game 1 demands.
3. The creation cost. For customer units that's CAC, fully loaded, with your inclusion list stated. For store or capacity units it's capex + working capital. Then work out payback on the disclosed or estimated ramp/retention.
4. The cohort or vintage exhibit. Use whatever the company discloses: retention triangles, repeat-rate curves, store-vintage tables, same-store sales by age. Read it in all three directions. If nothing is disclosed, build the maturity-mix decomposition from store counts and openings, and say what the silence itself tells you.
5. The reconciliation test. Units × unit P&L + corporate ≈ reported EBITDA. Compute the residual as a % of reported, and classify it as timing, definitions, or curation.
6. The verdict memo, one page, conclusion first. Are the unit economics working? Are they improving or decaying at the margin? Name the two assumptions the whole story hangs on, and what you'd check next quarter. End with the sentence that admits what you could NOT verify from public data.

Rubric (0 = missing/wrong · 1 = partial · 2 = complete and correct; 16 points; pass ≥ 13):

File both rebuilds in the company knowledge base you set up during orientation. Give the project ~3 hours. A Phase 8 teardown allots ~10 minutes to this question. The compression comes from doing this slowly, correctly, twice.

Case Lab: For a guided, auto-graded dry run before the live filings, run case CL-4.1 in the app's Cases tab for this module: it drops you into a synthetic company's raw disclosures and grades your unit choice, unit P&L, cohort/vintage read, and reconciliation line by line against the key: the same rubric above, on rails, so you arrive at the two real companies already fluent in the sequence.

Reading & resources

• Bill Gurley, "The Dangerous Seduction of the Lifetime Value (LTV) Formula" (Above the Crowd blog, September 2012). The canonical warning label for LTV and the LTV/CAC ratio, written by the VC who watched the formula misprice a decade of startups. His ten caveats map almost one-to-one onto the abuse catalogue you've just worked through. [Free] [Intermediate]
• David Skok, "SaaS Metrics 2.0" (forEntrepreneurs.com). The standard operator's reference for CAC, LTV, months-to-recover-CAC, and the unit-economics dashboard; where most of the SaaS thresholds you will hear in the wild come from. [Free] [Intermediate]
• Michael Mauboussin & Dan Callahan, "The Economics of Customer Businesses: Calculating Customer-Based Corporate Valuation" (Morgan Stanley Counterpoint Global, December 2021). The bridge from here to valuation: it builds company value bottom-up from CAC, retention, and unit cash flows; the professional-grade version of what you did in Worked example 3. [Free] [Advanced]
• Aswath Damodaran, "Going to Pieces: Valuing Users, Subscribers and Customers" (SSRN working paper, 2018; companion posts on Musings on Markets). User-based valuation with full DCF machinery: existing vs new units, corporate cost drag, and why unit value × units ≠ equity value without the reconciliation you learned here. [Free] [Advanced]
• Daniel McCarthy & Peter Fader, "How to Value a Company by Analyzing Its Customers" (Harvard Business Review, Jan–Feb 2020). Customer-based corporate valuation for a general reader. Then find McCarthy's public teardowns (Blue Apron, 2017, taken apart from its own S-1 cohort disclosures, a masterclass in reading a cohort triangle adversarially). [Paid: the HBR piece; the teardowns circulate free] [Intermediate]
• Jeff Jordan et al., "16 Startup Metrics" and "16 More Startup Metrics" (a16z blog, 2015–16). The definitions list for deck-land: GMV vs revenue, blended vs paid CAC, the churn caveats; useful precisely because founders quote these terms at you. [Free] [Beginner]
• Bessemer Venture Partners, "State of the Cloud" (annual) and the "Efficiency" sections of "Scaling to $100 Million". Live benchmark distributions for CAC payback, NRR, and gross margin by stage, and the calibration source for the payback bands above (check the current-year edition). [Free] [Intermediate]
• Alistair Croll & Benjamin Yoskovitz, Lean Analytics (O'Reilly, 2013). Chapters on the e-commerce, SaaS, and marketplace model archetypes: the "one metric that matters" discipline and per-model unit definitions; skim as a pattern reference. [Paid] [Beginner]
• Filings practice set, the raw material for all of it, free: Eternal/Zomato quarterly shareholder letters (GOV, take-rate, and contribution definitions to interrogate) · Zomato DRHP (2021) and Nykaa DRHP (2021) (cohort and repeat-rate exhibits; read them in the triangle's three directions) · Swiggy RHP (2024) (dark-store counts and maturity language, the P4 problem in the wild) · Honasa DRHP (2023) (CM-ladder definitions) · Avenue Supermarts AR + earnings calls (store adds, sq ft, per-store revenue; rebuild the SwiftMart store with real numbers) · Chipotle 10-K/10-Q (restaurant-level operating margin, rent in) · Netflix 10-K (regional members and ARM; content amortisation note) · Spotify AR (premium vs ad-supported gross margin split) · Uber 10-K + 2019 S-1 (the "Core Platform Contribution Margin" specimen; read what it excluded) · WeWork 2018–19 offering documents ("community-adjusted EBITDA", the allocation-game hall-of-famer, preserved in amber). [Free] [Intermediate]
• Business Breakdowns (Colossus) episodes on Chipotle, Wingstop, and Costco. Operators and analysts walking through store-level economics aloud; and the Invest Like the Best interview with Daniel McCarthy (2021) on customer-based corporate valuation. [Free] [Beginner]
• screener.in / Tijori (India) · stockanalysis.com / TIKR (US). Five-year pulls for the mini-project. [Free] [Beginner]
• Going deeper: Peter Fader & Bruce Hardie, the "buy-till-you-die" (BTYD) papers, plus the open-source lifetimes / BTYD packages. The probabilistic step beyond 1/churn. Instead of assuming geometric decay, you fit an actual purchase-and-retention curve (BG/NBD, Pareto/NBD) to cohort data. That's the churn-and-LTV caveat ("real curves flatten; take churn from mature cohorts") industrialised into a model you can run on a spreadsheet of orders. [Free] [Advanced]
• Going deeper: Peter Fader & Sarah Toms, The Customer Centricity Playbook (Wharton School Press, 2018), read alongside Daniel McCarthy's academic papers on customer-based corporate valuation. The rigorous foundation under the LTV build and Worked example 3: moving from a single point-estimate LTV to a fitted, cohort-level customer-value model that rolls up into an equity value, the bridge Phase 3's valuation work will want. [Paid: the book; the papers circulate free on SSRN] [Advanced]

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand — that is how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds; it never replaces. (R1/R10)

AI-Augment this skill

{ "skill": "Auditing a growth deck's stated LTV by testing which retention-curve SHAPE its number implicitly assumes", "use": "Feed an LLM the cohort retention chart or table from a pitch deck / investor update and have it transcribe the month-by-month percentages into a data table, so you spend your time on the curve-fitting and verification rather than re-typing numbers off a screenshot.", "tools": ["Claude or ChatGPT to transcribe a cohort chart image or PDF table into structured data", "a notebook (plain Python, or pandas for a large panel) to fit constant-hazard vs declining-hazard survival curves", "the company's own S-1 / investor-update cohort disclosures where available"], "prompt": "Here is a cohort retention table transcribed from [company]'s deck: month, % of cohort still active. Fit a constant monthly churn rate to (a) the first three months and (b) the most recent three months separately, compute the honest discounted LTV under each, and flag how large the gap is between them.", "verify": "Recompute the deck's own headline LTV by hand using the module's LTV = m.r/(1+d-r) formula with the SAME two churn windows the AI used, and confirm which window (if either) the deck's stated number actually matches -- decks routinely quote the flattest, most flattering window without saying so.", "diy": "The mastery check hands you a raw cohort triangle and asks for the honest discounted LTV by hand, with no curve-fitting tool in the room. You must be able to read which part of the curve is the flat, mature part and compute from there -- the workflow above only removes the transcription and cross-checks the choice." }

Modern Data Analysis

By hand first. The cohort-triangle habit you built by hand hasn't changed. Read the flat part of cohorts older than a year, never the first few months. That's what the mastery check gates.

Today's workflow. The reading list above already names the professional next step, Fader & Hardie's buy-till-you-die literature, which fits an actual retention curve instead of assuming geometric decay. The block below is the smallest version of that idea. It builds a monthly survival curve with a declining hazard rate, the shape real cohorts show, because the highest-risk customers leave first and the survivors left behind are progressively stickier. Then it fits the constant-hazard LTV = m·r/(1+d−r) formula to two different windows of that one curve, and compares both to the true, curve-integrated LTV.

Tools & sources (IN + US). India: cohort exhibits in DRHPs (Zomato, Nykaa, Swiggy, already the raw material above) and quarterly shareholder letters. US: S-1 cohort tables on EDGAR; the open-source lifetimes package (BG/NBD, Pareto/NBD models) for fitting an actual retention curve instead of eyeballing a chart. Bessemer's "State of the Cloud" for payback/NRR benchmark bands to sanity-check any fitted curve against.

# recomputed in-session -- R3. A synthetic 24-month cohort with a genuinely DECLINING hazard
# rate (a power-law survival curve S(t) = (1+t/k)^-p -- the shape a mixture of constant-hazard
# customers with heterogeneous risk produces at the POOLED level; see Fader & Hardie). Then two
# constant-hazard ("exponential") LTV fits on two different windows of the SAME curve.
k, p = 5.0, 1.6
def S(t): return (1 + t/k)**(-p)

months = range(1, 25)
monthly_r = {t: S(t)/S(t-1) for t in months}          # implied monthly retention at age t

m, d_annual = 500, 0.14                                 # monthly contribution; annual discount rate
d_month = (1+d_annual)**(1/12) - 1

true_ltv = sum(m*S(t)/(1+d_month)**t for t in range(1, 241))   # non-parametric, full curve

r_early = sum(monthly_r[t] for t in (1,2,3))/3           # naive fit: months 1-3
ltv_early = m*r_early/(1+d_month-r_early)

r_late = sum(monthly_r[t] for t in (10,11,12))/3          # naive fit: months 10-12 (the "flat part")
ltv_late = m*r_late/(1+d_month-r_late)
# true_ltv ~ 2,812 ; ltv_early ~ 1,676 (-40%) ; ltv_late ~ 4,123 (+47%) ; late/early ~ 2.46x

Verify. Run this against the Cloudline figures first, as the anchor. m=4,800, r=0.88 (annual), d=0.11 reproduces LTV = 4,800×0.88/(1.11−0.88) = ₹18,365 and naive LTV = 4,800/0.12 = ₹40,000, a 117.8% overstatement that matches the earlier "118%" to rounding. The synthetic declining-hazard curve above then isolates a second, independent error that the Cloudline comparison doesn't show. Even the discounted formula overstates or understates, depending on which window of a non-constant curve supplies its single churn rate.

Quantitative lens

1. Fitting the wrong survival shape changes LTV by a large, computable factor. The Python block's curve has month-1 retention of 74.7%, climbing to 94.5% by month 24. That's a realistic concave, flattening shape, while the LTV formula assumes straight-line decay. Fit a constant monthly churn rate, the formula's only input, from three different places on that one curve and you get three different answers:

The early-window fit and the late-window fit differ by 2.46× on identical underlying data. That's a bigger swing than the naive-against-discounted comparison in the LTV section. Both parametric fits also miss the true, curve-integrated number, in opposite directions. Read churn too early, the classic new-company mistake the triangle warns you about, and you understate LTV. Read it too late and extrapolate a flat rate forever, and you overstate it, because the true hazard keeps declining past month 12. The instruction to "take churn from the flat part of cohorts older than a year" is the right practical fix because it's closer to the second row than the first. It's still a fit, though. Only the third row gives you the truth.

2. A closed form for payback as a function of gross margin and churn. The payback you learned is CAC ÷ monthly contribution. It's a single customer's clock, and it says nothing about churn. A cohort's payback, the month when cumulative expected contribution across all its survivors first repays CAC, has an exact closed form. With monthly retention r and monthly contribution per surviving customer m = ARPU × gross margin, cumulative contribution through month T is the geometric sum m·r·(1−r^T)/(1−r). Set that equal to CAC and solve:

T = ln[1 − CAC·(1−r)/(m·r)] / ln(r)

Apply it to Cloudline's own numbers (CAC = ₹6,000, m = ₹400/month, monthly r = 0.88^(1/12) = 0.9894). Naive payback is 6,000/400 = 15.0 months, the figure from the worked example, and the churn-adjusted closed form gives 16.4 months. Churn stretches payback because some of the cohort has already left before month 15, so the survivors' contribution has to cover the departed's unrecovered share too. The gap widens fast as gross margin falls or churn rises. A low-margin, high-churn business (30% margin, 8%/month churn) can show a naive payback of 20 months while the churn-adjusted number is infinite. CAC exceeds the cohort's entire lifetime-contribution ceiling m·r/(1−r), so no amount of time repays it. The naive formula can't see this failure mode, because it never looks at retention at all.

3. Blended CAC hides the marginal CAC governing the next rupee of spend. Take a two-channel model. Pair a fixed, cheap pool (₹500 CAC, capacity 6,000 customers/quarter) with a paid channel whose marginal cost rises with volume (marginal CAC = ₹700 + ₹0.18 × paid customers). That makes abuse #5, blended CAC while scaling paid, exact. Set it against an LTV of ₹4,500 and the standard LTV/3 ≈ ₹1,500 ceiling. Marginal CAC crosses ₹1,500 at ~4,400 paid customers/quarter, while blended CAC, still diluted by the cheap pool, doesn't cross ₹1,500 until ~13,700. In the ~9,300-customer gap between those two points, the arithmetic above puts the excess marginal spend, over what an LTV/3-respecting budget would have allowed, at roughly ₹78 lakh for that one quarter alone. A finance team watching only the blended number keeps signing off on the marketing budget for the entire gap.

Limits. That declining-hazard curve above is a constructed illustration. It isn't a fitted model of any real company's cohorts. Fader & Hardie document the qualitative lesson, that pooled hazard rates decline with cohort age and a single constant-churn number is shape-dependent. But the specific 2.46× factor is a property of this one curve's parameters, and you shouldn't treat it as a universal constant. Refit it on your own company's actual cohort ages before you quote a factor. The churn-adjusted payback formula assumes retention is constant across the payback window, which is the very assumption Lens 1 shows is usually wrong. So treat its output as a better approximation than the naive formula, short of an exact answer. When the stakes justify it, the trustworthy version is a full numeric integration over the true survival curve, as in the true-LTV code above. Similarly, the marginal-CAC model assumes a known, smooth cost-per-customer function for the paid channel. Real auction dynamics are noisier. What a finance team can actually observe as "marginal cost" is usually the last channel's or cohort's blended cost, itself a coarse proxy for the true derivative. It's good enough to catch the abuse in Lens 3, but it isn't precise enough to set a budget to the rupee.

Where this goes next: galaxy cross-links

• G1 · QM1.01 (probability-distributions-estimation): the estimation theory behind fitting a hazard/survival model to cohort data instead of eyeballing a curve.
• G2 · DA1.05 (statistics-in-code): where the survival-curve fitting above becomes a real, checked statistical workflow rather than a teaching toy.
• G2 · DA1.02 (pandas-data-wrangling): turns a single cohort triangle into a multi-cohort panel across years of filings.

Teach it back & journal

Feynman prompt. Your 15-year-old cousin runs a cookie-subscription box from her kitchen. It's ₹500/month, growing fast on Instagram, and "almost profitable, once we're bigger." Teach her all of it using only her own stall. One box. What does it really leave behind after flour, butter, packaging, the delivery boy, and the "first box half price" she keeps forgetting to count? (That's contribution, by behaviour. And yes, the discount counts, because it wouldn't exist without the box.) One customer. Take the ₹300 of boosted posts that found her. How many months of box-profit until that ₹300 is repaid, and what is she living on until then? (CAC, payback, and the float. Her pocket money is the Series A.) The January friends. Of the ten who subscribed in January, how many still order in June, and are the March joiners holding up as well as January's did? (The triangle: maturation, vintage, and the exam-month dip that hit everyone at once.) Then comes the killer question. Her Instagram total says "orders up 25% every month" while each month's new joiners quit faster than the last. Is her business growing or melting, and which single table settles it? If she says "melting, so show me each month's group separately," she's understood what most pitch decks are built to prevent you from seeing. Finish with the one-page explainer in her language: find the smallest thing you sell again and again; count everything one more of it really costs, including what it cost to find the buyer; check the old groups, not the total; and only call it a business if one more box makes money on its own.

Journal reflection. Pick one subscription you personally pay for: an OTT, a gym, a meal plan, a SaaS tool at work. Working from public prices and your own behaviour, estimate its unit economics on you as fairly as you can. What's your monthly contribution to them, what did they likely pay to acquire you, and have they earned it back yet? Then work out which cohort you belong to. Are you a habit-floor resident, or a churn statistic waiting to happen? Then two harder entries. (1) Which of the five games do you think you would most likely have accepted in a deck a month ago (allocation, blending, growth-spend classification, denominator, or the missing balance sheet), and what specifically will you now demand before believing that flavour of claim? (2) Record how long your three gate rebuilds (P4, P7, P10) actually took, start to finish. The Phase 8 target is twenty minutes for one rebuild. Write down what you'll drill to close the gap (churn algebra, the CM ladder, reconciliation) and file the number where you'll see it again in Phase 5.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.