Business Model Anatomy

Phase 4 · Week 36 · ~20 hours

Where it sits. A grocer with a 4.6% net margin can be a great business, and a software company with an 80% gross margin can be a mediocre one. Two companies with identical revenue can deserve multiples 10× apart, and the biggest revenue line is often not where the profit lives. Nothing in the first three phases explains any of that. This is the first module of the strategy-and-moats phase, and the direct answer to Question 1 of the five-questions spine: what does it do and how does it make money? You arrive armed. Phase 1 taught you to read the statements, Phase 2 to distrust and dissect them (margins, returns on capital, cash conversion, red flags), Phase 3 to convert cash flows into value and to reverse-engineer what a price implies. What you do not yet have is a machine-level understanding of the businesses behind those numbers. Here you build the anatomy chart: the value chain from input to cash, the six revenue-model archetypes and their economics, the P&L waterfall each archetype produces on its way to free cash flow, operating leverage, revenue quality, customer concentration, and the flywheel logic (including "scale economies shared") that lets a few businesses convert size into an ever-wider advantage. The next module will zoom into the smallest repeatable unit (contribution, CAC, LTV, cohorts); the one after widens to industry structure; the three after that ask what protects the economics you learn to read here. By Phase 8, all of it compresses into the first three questions of the 12-question teardown, answered in minutes, from filings, unaided.

Suggested hour budget. Core content: 8h · worked examples: 2h · practice set: 4h · applied mini-project (the Business Model X-Ray): 4h · flashcards + mastery check: 2h.

App link. The Industry & Business Model branch is seeded here: the model-ID drill deck (classify real revenue engines from segment-note excerpts against the clock; the gate sits at ≥85%), the waterfall builder (drag cost blocks to reconstruct how ₹1 of revenue becomes FCF for a mystery archetype), "find the profit engine" cases (segment tables in, verdict out), and the flywheel verifier (five years of KPIs in, "compounding or just growing?" out).

Learning objectives

By the end of the week you can:

1. Trace any company's value chain end-to-end (inputs → transformation → distribution → customer → cash), marking where margin pools, who carries inventory and receivables, and when cash actually arrives relative to costs.
2. Classify any revenue engine into the six archetypes (transactional, recurring/subscription, marketplace/take-rate, advertising, licensing/royalty/franchise, usage-based), state each archetype's characteristic economics (gross-margin shape, marginal cost of the next unit, capital intensity, working-capital sign, the KPI that matters, the characteristic leak), and decompose hybrid stacks (Apple, Amazon, Eternal) from segment disclosures.
3. Walk ₹1/$1 of revenue down the P&L waterfall to free cash flow for each archetype, name where the money leaks (COGS, TAC, S&M, SBC, tax, working capital, capex, minorities), and explain why FCF conversion differs so widely across models.
4. Keep GMV, gross bookings, and systemwide sales strictly separate from net revenue: apply the principal-vs-agent control test (Ind AS 115 / IFRS 15 / ASC 606), compute take rates on a consistent base, and refuse cross-model revenue comparisons until both are restated on the same basis.
5. Compute and interpret operating leverage: contribution margin, degree of operating leverage DOL = Contribution / EBIT, breakeven revenue, margin of safety. Push the same revenue shock through a high-fixed-cost and a low-fixed-cost structure, in both directions.
6. Grade revenue quality on the recurring ↔ one-off ladder, find the evidence in filings (contract liabilities/deferred revenue, RPO, order books, renewal and repeat rates), and refuse to capitalize windfall earnings at recurring-revenue multiples.
7. Read customer-concentration disclosures (the ≥10% single-customer rule under ASC 280 / Ind AS 108 / IFRS 8, risk factors, receivables notes) and translate concentration into consequences for pricing power, margins, terms of trade, and cliff risk.
8. Find the real profit engine: build the revenue-share vs profit-share table from any segment note, identify which line actually deserves the valuation, and verify a claimed flywheel / scale-economies-shared story in the numbers rather than the narrative.

Prerequisites & connections

Builds on. M1.02 to M1.06 cover the P&L and revenue recognition, plus the segment note you learned to find in M1.10. Revenue recognition there meant Ind AS 115/ASC 606's five steps, taught for recording. Here you reuse them for classifying. M2.01 to M2.02 gave you margin structure and the DuPont identity. What follows explains why margins and turns trade off across models, which is the grocer-vs-software question DuPont raised but couldn't answer. M2.03 set ROIC/ROCE as the master test. Every archetype here gets judged by the return-on-capital shape it produces, against the course's illustrative hurdles of ~12% INR for India and ~8 to 9% USD for the US (as of mid-2026, check against your M3.03 builds). M2.04's CFO/EBITDA and FCF/PAT cash-conversion checks, plus the maintenance-vs-growth capex split, get used here whole. M2.05 supplies the cash conversion cycle, where negative working capital comes back as a design feature of certain models. Then come M3.04 to M3.07 and the value drivers. Growth is worth nothing without a spread, so a business model is in the end a machine for producing (or failing to produce) a durable ROIC−WACC spread. The reverse-DCF habit from M3.05 reappears as "what is the market assuming this model can do?"

Feeds forward. M4.02 takes today's contribution-margin preview and makes it rigorous (unit economics, CAC/LTV, cohorts, payback). M4.03 zooms out from the company's own chain to the industry's chain and profit pool. M4.04 to M4.05 ask what defends the economics, because you can't judge a moat until you can say what it protects. Phase 5 industrializes all of it into seventeen sector playbooks, each a named archetype plus sector physics and KPIs. Phase 8's teardown questions 1 to 3 are these skills under a stopwatch. They ask what the company does and where the real profit engine is, who the customers are (concentration, recurring?), and how ₹1 flows to FCF. Competency C3 in the master map ("understand a business model cold") is certified on the skills built here.

The short version. Revenue lines aren't equal. Each of the six ways of charging for value carries its own gross-margin shape, marginal cost, capital hunger, cash timing and failure mode. So your first job is to name the machine and walk one rupee through it to free cash flow. Then find the engine room. It's often not the biggest room on the ship.

Core teaching content

4.1 What a business model actually is

Strip away the consultant fog and a business model is the answer to five mechanical questions:

1. Who pays? That's the customer, who isn't always the user. You use Google Search free, while advertisers pay.
2. For what? This is the unit of value: a basket of groceries, a month of access, a completed match, an impression, the right to a brand, a consumed gigabyte.
3. How is the charge structured? Per transaction, per period, per match, per attention-unit, per unit of usage, as a royalty on someone else's sales.
4. What does the next unit cost to serve? This is the marginal cost, the most underrated number in analysis.
5. When does cash move, in which direction, relative to cost? Cash might come upfront or after 90 days. It might even arrive before you pay your own suppliers.

Everything else in this phase (unit economics, industry structure, moats) hangs off these answers. Notice what's missing from the list: the product, the technology, the founder's story. Two companies can sell the identical product with opposite models. A film studio sells a movie once to a distributor, while Netflix rents it by the month forever. One company can also run several models on one product. Apple sells you the iPhone, which is transactional. It takes ~15 to 30% of App Store spend as a marketplace and collects iCloud monthly as a subscription. Google also pays it roughly $20bn a year to be the default search engine, which is licensing of an audience (illustrative, per the DOJ trial record).

The habit to install this week. When you meet any company, before you compute a single ratio, write one line per revenue stream in this grammar: "[Who] pays [what basis] for [what unit]. The next unit costs roughly [X]. Cash arrives [when]." What follows teaches you what each answer implies.

4.2 The value chain: from input to cash

Before the taxonomy, walk the physical journey once. Every model is a decision about where in this chain to stand, and how to charge for standing there:

inputs → transformation → distribution → customer → CASH
(suppliers)  (make/assemble/   (reach: stores,   (purchase     (collection:
             curate/match)     feet-on-street,    decision)     upfront, on
                               servers, spectrum)               delivery, on
                                                                credit, never)

Three questions locate a company on this map:

• Where does the margin pool? In branded consumer goods the brand owner keeps most of it, with HUL's gross margin at ~50%+. Its distributors run on 4 to 6% margins and its kirana retailers on 8 to 12%. In PC hardware, for two decades most of the pool sat with Intel and Microsoft. The assemblers kept little of it. Never assume the consumer-facing name earns the money.
• Who carries the working capital and risk? Inventory risk, receivable risk, warranty risk, demand risk: a company can hold each one itself or push it up or down the chain. DMart holds inventory but sells for cash, so it has no receivables. An EPC contractor holds neither inventory nor the asset. It still carries brutal receivables and execution risk.
• When does cash arrive relative to cost? This is the cash conversion cycle you already know, reborn as strategy. A subscription business collects twelve months upfront and provides the service later, so its customers finance the company. An insurance company is paid decades before it pays claims, and for an insurer the float is the business model. Amazon historically collected from your card days after shipping but paid suppliers on 60 to 90 day terms. That negative cash cycle meant growth generated cash instead of consuming it. Reverse the sign and growth becomes a cash furnace, even when the P&L shows profit. You saw that with the EPC and jewellery-retail examples in Phase 2.

Follow the rupee, DMart edition. Take a ₹100 grocery basket at DMart (Avenue Supermarts). The figures are illustrative, modelled on ~FY25 scale, with revenue of ~₹59,000 cr and ~415 owned stores. Pull the actual filing to check them. Here's how the basket splits:

Five rupees on a hundred looks miserable next to a software company. Then you remember the DuPont lesson. DMart turns its capital fast, with inventory turns of ~11 to 12× and sales per store among the highest in Indian retail. It pays suppliers within roughly a week in exchange for cash discounts, and it passes those discounts into prices. You'll see that loop again in the flywheel. Thin margin × high velocity × tight capital = high-teens ROCE. The thin margin is the model, chosen on purpose. So judge every model on spread and reinvestment, never on margin alone.

4.3 The six archetypes: the taxonomy on one page

Classify by how the charge is structured. The industry label won't help you. "Tech company" isn't a model at all: Alphabet is advertising, Snowflake is usage, Uber is take-rate, Apple is a stack.

Two habits for using the table. First, most real companies are stacks. So classify each stream on its own, instead of hanging one label on the whole company. You'll take stacks apart properly later in the week. Second, the columns are causal. The charge structure determines the margin shape, which determines the operating leverage. That in turn decides how hard the same demand shock hits profit. You're learning physics. Zoology would only teach you the names.

4.10 Hybrid stacks: real companies run several engines

Almost nothing pure survives at scale, because managements bolt new engines onto old audiences. So don't ask what model a company is. Ask what the stack is, and what each layer contributes to revenue, gross profit, and growth.

The decomposition protocol works on any filing:

1. Revenue disaggregation note. Ind AS 115 / ASC 606 requires it, and it gives revenue by stream or type. Your archetype labels go here.
2. Segment note (Ind AS 108 / ASC 280). This gives profit by segment, the raw material for finding the profit engine. Streams and segments rarely line up perfectly, so build the bridge between them.
3. MD&A / shareholder letter KPIs: GMV, ARR, membership, take rates. This non-GAAP layer is what makes the GAAP layers comparable.
4. Produce the stack table: stream | archetype | revenue share | gross-profit share (estimate if undisclosed) | growth rate | the KPI. Build that one table and the company is anatomically yours.

Three worked stacks, at the same illustrative scales as before:

• Apple: products ~$295bn at ~37% gross margin (transactional) + services ~$96bn at ~74%. Services is a stack within the stack, with an App Store take-rate of ~15 to 30%, iCloud subscriptions, AppleCare and ~$20bn of licensing receipts from Google. Services ≈ 25% of revenue but ~39% of gross profit. The multiple the market pays re-rated as that mix shifted.
• Amazon: online stores ~$247bn (1P transactional, gross) + 3P seller services ~$156bn (take-rate, net) + AWS ~$108bn (usage) + ads ~$56bn (advertising) + subscriptions ~$44bn (Prime). That's five archetypes in one P&L. The segment note shows AWS at ~17% of revenue producing ~58% of operating income.
• Eternal (Zomato): food-delivery take-rate + Blinkit take-rate + advertising (restaurant ads inside both) + subscription (the loyalty programme) + Hyperpure. Hyperpure is 1P B2B distribution booked at gross revenue on a thin margin. Notice how it flatters the consolidated revenue line, and restate before you compute "take rate on total revenue".

The stack view immediately upgrades three judgments. The first is growth quality. Which engine is growing, the 37%-margin one or the 74%-margin one? The second is margin trajectory, where mix shift is destiny. Apple's blended margin rises with zero operational improvement. The third is valuation. Different engines deserve different multiples, which is Phase 3's SOTP logic, and a sum-of-the-parts is how you price a stack.

4.11 How ₹1 of revenue becomes free cash flow: the waterfall by archetype

Now the centrepiece. Take the same ₹100 of revenue in each model and walk it to FCF. The marketplace is also shown per ₹100 of GMV, and you must never confuse the two bases. The numbers are synthetic but calibrated to the real exemplars above, with tax at 25% throughout so the columns compare.

Now read the table the way an analyst would:

• Each model leaks in its own characteristic places. The grocer leaks at COGS and growth capex. SaaS leaks at S&M and SBC. The marketplace leaks at serving costs and incentives, and the ad platform at TAC and now capex. The franchisor barely leaks at all. Its revenue line is tiny relative to system sales because the leak happened before the revenue, at the franchisee.
• Two FCF adjustments keep the columns fair. (1) Maintenance vs growth capex. The grocer's 0.6 is what's left after choosing to grow, and its steady-state FCF is ~4.6. Punishing chosen growth as if it were decay is a rookie error. So is waving all capex away as "growth". (2) SBC. The SaaS column's 17.5 of "FCF" includes 12 paid in shares. Treat SBC as the cash cost it economically is, which retests FCF at ≈ 5.5, or model the dilution explicitly. Never neither. Snowflake is the live case.
• Deferred revenue and float are real money. SaaS collects ahead of recognition (+3), and the marketplace holds seller money for days (+0.3 per 100 of GMV). Models that collect before they serve self-finance their growth. Models that serve before collecting rent working capital from banks. EPC firms and exporters are the usual examples. Same P&L, opposite cash lives. It's the reason you learned to compute CFO/EBITDA before trusting anyone.
• FCF conversion norms. These are steady-state rules of thumb to calibrate against, and they aren't laws. Franchisors and licensors convert ~90 to 100% of NOPAT. Mature SaaS converts ~80 to 100% once you've subtracted SBC. Ad platforms historically ran ~90%+ and now depend on capex. Marketplaces convert highly once mature. Transactional retail converts 60 to 90% on a maintenance basis, and little while it's rolling out stores. Infra-usage businesses (cloud, telecom) run chronically below 100% during buildouts. When a company's conversion is far from its archetype's norm, you've found either a red flag of the kind Phase 2 taught you to spot or a story worth understanding.

4.17 Unit economics: the preview

The next module covers this fully, so carry just three sentences in as luggage. Every model here breaks down into a smallest repeatable unit: a store, an order, a subscriber, a member, a policy, a seat. The truth about the model is found at that grain. It's in the contribution per unit (unit revenue − unit variable cost), the cost to acquire the unit (CAC), and how long the unit stays. Retention and churn measure that, and they lead to LTV and the payback period. Consolidated P&Ls are just units × maturity mix. That's how a reported loss can hide excellent unit economics still in their CAC-heavy youth, like Eternal's city cohorts. And a reported profit can hide units that stopped paying for themselves. In the next module, you rebuild companies from the unit up.

Common mistakes & how experts think differently

1. Reading revenue size as importance. The rookie summarises the mix ("46% catering"). The expert builds revenue share against profit share and finds the 31% line earning 76% of profit, as in the IRCTC example. The biggest line is context. The most profitable line is the thesis.
2. Comparing revenue across accounting presentations. P/S screens that rank a 1P retailer "cheaper" than a 3P marketplace are comparing a gross number to a net number. Experts restate to GMV or gross profit before any cross-model comparison. They also check the principal-vs-agent note before trusting a revenue trend.
3. Margin worship. "Only 4.6% net margin" (DMart) and "80% gross margin!" (a churning SaaS) are both category errors. Margin is one factor of ROIC, and on its own it can't give you a verdict. Experts run DuPont instinctively (margin × velocity × leverage) and judge the spread, never the margin.
4. Paying recurring multiples for windfall earnings. Think of peak-cycle contracts, exclusivity windows, pandemic surges. Classify them correctly as rung-5 and they're one-time cash. Classify them lazily as "growth" and they're the top of your drawdown. Experts normalise first and multiply second.
5. Believing "subscription" means "safe". A consumer streaming sub with 4%/month churn is a leaky bucket wearing an annuity's badge. A depository's issuer charges churn ~never. Experts price the churn and the switching reality and ignore the billing frequency.
6. Adding back the real costs. "Adjusted EBITDA before SBC and incentives" counts dilution and customer bribes as profit. Experts treat SBC as compensation (cash-equivalent) and ask whether growth survives when the incentives stop.
7. Admiring operating leverage only on the way up. Margin expansion in a boom is arithmetic. Experts compute the DOL, run the shock both ways, and stack it against financial leverage before calling a HighFix business "high quality".
8. Ignoring the 10% footnote. Concentration hides in the last paragraph of the segment note while the MD&A celebrates "marquee client relationships". Experts read the footnote first and the adjectives never. When one buyer is >25%, they underwrite the customer's health as well.
9. Accepting flywheel slides as evidence. Every prospectus has arrows in a circle. Experts demand the four-number verification: policy-flat gross margin, falling opex %, rising ROIC and throughput, retention improving without bought traffic. If those numbers don't show up, what you're looking at is a marketing budget.
10. Classifying by industry instead of charge structure. "Retailer" DMart and "retailer" Nykaa (inventory + marketplace + ads) are different animals, and "tech" contains at least four archetypes. Experts name the engine or engines from the revenue note. Then they let the archetype dictate which KPIs even matter.

Worked examples

Worked example 1: One restaurant, three business models

One pizza outlet rings up ₹2.4 cr of annual sales, and that same consumer demand supports three different claims on the value chain. Compute each party's economics. The numbers are synthetic and calibrated to Indian QSR disclosures. Tax is ignored, and the view is store-level.

(a) The franchisee-operator (runs the store, owns the P&L):

On invested capital of ~₹150 lakh (fit-out + deposits + WC), pre-tax ROIC = 32.3/150 ≈ 21.5%. That's a good business, but a capital-hungry, execution-heavy one, with rung-3 revenue re-earned daily.

(b) The franchisor (owns the brand): revenue = 240 × 3.3% = ₹7.9 lakh from this store. Servicing costs (audits, allocated menu R&D) run ~15 to 20% of that. So the franchisor makes ~₹6.5 lakh EBIT, ~82% margin, on ~zero incremental capital. Per store it's pocket change. Across 2,000 identical stores it's ₹130 cr of nearly capital-free EBIT, compounding at the franchisee's expense. Now you know why Jubilant FoodWorks, the real-life (a) with ~₹8,000 cr revenue (illustrative FY25), built the stores. And you know how Domino's Inc., the real-life (b), reached ~45%+ margins on royalties while its own biggest revenue line, supply chain at ~11% margin, carries the low-margin volume.

(c) The delivery marketplace (40% of sales flow through it): GMV = 240 × 40% = ₹96 lakh. A take rate of ~22% including fees gives revenue of ₹21.1 lakh. Variable serving cost (riders net of delivery fees, payments, support) comes to ~13% of GMV = ₹12.5 lakh. That leaves contribution of ₹8.6 lakh (9% of GMV). Whether this store's flow is profitable to the platform depends on its share of fixed costs (tech, brand).

The lesson. One customer's ₹480 pizza order feeds three P&Ls with three archetypes. The operator's is re-earned transactional: thin, capital-heavy, execution-graded. The franchisor's is royalty, a thin slice with near-infinite ROIC. The marketplace's is take-rate, a toll on the flow that's profitable only at network scale. When these three negotiate over royalty renewals and commission disputes, you're watching the value chain argue over one pie. Once you've built all three P&Ls, you know who holds the leverage. It's the party the customer would follow.

Worked example 2: The same shock, two cost structures (operating leverage, both directions)

Use the twins from the operating-leverage section: HighFix (fixed ₹600, variable 30%) and LowFix (fixed ₹100, variable 80%), both at revenue ₹1,000 and EBIT ₹100.

Step 1: DOL. Contribution is 700 (70%) at HighFix and 200 (20%) at LowFix. So DOL = 700/100 = 7.0 and 200/100 = 2.0.

Step 2: a +20% year. HighFix: 1,200 − 600 − 360 = 240, up 140%. Check it: 7 × 20% = 140%. LowFix: 1,200 − 100 − 960 = 140, up 40%, which also checks.

Step 3: a −20% year. HighFix: 800 − 600 − 240 = −40, down 140% and into losses. LowFix: 800 − 100 − 640 = 60, down 40%.

Step 4: breakeven & safety. HighFix breaks even at 600/0.7 = ₹857, a 14.3% margin of safety. LowFix breaks even at 100/0.2 = ₹500, a 50% margin of safety.

Step 5: flow-through (the practitioner's phrasing). HighFix converts incremental revenue to EBIT at 70 paise per rupee. Hotel and multiplex managements call this "flow-through" on earnings calls. When PVR-Inox adds ad revenue or IndiGo fills the last five seats, contribution ≈ 90 to 100%. The same arithmetic ran in reverse in FY21's shutdowns.

Step 6: the modelling rule. Forecast HighFix with a stable EBIT margin across a ±15% demand band and you mis-state EBIT by roughly ±90% (7 × 15% vs the assumed 0). For any HighFix name, model fixed + variable explicitly, or use flow-through ratios from history. Margins come out of the model. They don't go in. File this away too, because your three-statement model template gains a cost-split tab thanks to this example.

Worked example 3: GMV vs revenue: the presentation trap, computed

Here are two quick-commerce firms with identical economics and different revenue recognition. Both have GMV of ₹1,200 cr, a product margin of 18%, warehouse + rider costs of ₹150 cr and corporate fixed costs of ₹60 cr.

KwikKart (1P, the principal). It owns the inventory and sets prices.

• Revenue = ₹1,200 cr (gross). COGS = 1,200 × 0.82 = ₹984 cr, leaving gross profit of ₹216 cr.
• EBITDA = 216 − 150 − 60 = ₹6 cr. EBITDA margin on revenue = 0.5%.

ZipBasket (3P, the agent). Sellers own the inventory, and it charges an 18% total take.

• Revenue = 1,200 × 18% = ₹216 cr (net). The operations are the same, so EBITDA = 216 − 150 − 60 = ₹6 cr. EBITDA margin on revenue = 2.8%.

The businesses are identical, yet reported revenue differs 5.6× and margins differ 5.6× the other way. Now spring the screen trap. Both trade at EV ₹2,400 cr. On EV/Revenue, KwikKart is at 2.0× and ZipBasket at 11.1×. A multiples screen flags KwikKart as "cheap". It's pure illusion. Restate both on GMV and EV/GMV = 2.0× each. Restate on gross profit-equivalent (₹216 cr) and it's 11.1× each. They're identical, as they must be.

The professional protocol has three steps. (1) Read the revenue-recognition note for principal/agent language: control, inventory risk, pricing discretion. (2) Put marketplaces and 1P retailers on a common base, using EV/GMV for flow and EV/gross profit for economics, because gross profit is presentation-proof. (3) Treat any 1P/3P mix shift as a restatement event, whether it's Amazon's perpetual state or a company "platformising". Growth and margin trends break mechanically at the seam. There's a bonus check, too. ZipBasket's model holds seller float. It collects on day 0 and settles on day 7. At ₹1,200 cr GMV and a 7-day cycle, that's ~₹23 cr of permanent free funding (1,200 × 7/365). It's the marketplace's quiet balance-sheet gift, and the 1P twin doesn't get it.

Worked example 4: The real profit engine: IRCTC, segment by segment

IRCTC has four segments. The figures are illustrative, modelled on ~FY24 scale, so pull the AR's Ind AS 108 note and re-run them.

Step 1: name the machine. The headline says "catering company", because catering is the largest line at 45.6%. The profit table tells a different story. It's a usage-based toll, a ₹15 to 30 convenience fee per online ticket at ~84% margins, wearing a catering uniform. Three-quarters of profit comes from under a third of revenue. Profit intensity is 2.47 there, against 0.26 to 0.37 everywhere else.

Step 2: model the engine, never the average. Consolidated revenue growth is nearly meaningless here. A ±10% move in catering shifts total EBIT by ~₹25 cr (17% of it × ~12% margins). A ±10% move in ticketing volumes or fee rates shifts EBIT by ~₹111 cr, 4.5× the sensitivity on a line a third the size. So your forecast drivers are online ticket volumes (structural: railway capacity + online penetration) × fee per ticket, and that fee is regulated. Forget "revenue growth of 12%". Step 3: locate the risk where the profit is. In October 2021 a government directive proposed sharing ~50% of convenience-fee revenue with the Railways. The stock fell ~25% intraday before the order was reversed within a day (check the episode). It's instructive because nothing about catering, water, or tourism moved. When 76% of profit comes from one administered fee, the segment note is the risk register. Monopoly + regulated toll = a profit engine that serves at the pleasure of the regulator. The multiple debate about IRCTC is really a debate about one line. Step 4: generalise the protocol. Give any segment note sixty seconds: shares, intensities, engine, the engine's KPI, the engine's kill-switch. Run it on Amazon (AWS), Apple (services), Qualcomm (QTL), Info Edge (Naukri funding three cash-burning classifieds) and Domino's (royalty vs supply chain). It's a habit that compounds for life.

Worked example 5: The hybrid stack and the sum-of-the-parts trap (India)

Sangam Digital (synthetic) is filed, screened, and discussed as "an Indian e-commerce company", because 60% of its revenue is a first-party online store. Build the stack from its revenue-disaggregation and segment notes. Find the engine. Then let the engine price the company through a sum-of-the-parts.

Step 1: name the machine before the company. The biggest room on the ship, the 1P store at 60% of revenue, throws off just 5.4% of EBIT at a 2% margin. The recruitment annuity is 30% of revenue and 81.1% of EBIT at a 60% margin, profit intensity 2.70. Sangam is a recruitment-annuity business wearing an e-commerce coat. The commerce line buys the traffic and the growth headlines, while the listings line pays the bills.

Step 2: why the blended lens lies. The blended EBIT margin is 22.2% and the blended gross margin 42.3%. Both are weighted averages that describe no activity that actually exists. Feed the low-margin store faster and the blend falls with zero deterioration in any engine. That's the mix-shift point from the stack section, and P10 computes it. Never model or judge Sangam on a consolidated margin.

Step 3: let each engine carry its own multiple (SOTP). Different archetypes deserve different multiples, and one number on the whole stack mis-prices it. Use EV/EBIT multiples suited to each archetype. Here they're illustrative (build the real comps in Phase 3): annuity 25×, transactional 12×, advertising 18×.

• Recruitment: 540 × 25 = ₹13,500 cr
• Online store: 36 × 12 = ₹432 cr
• Ads: 90 × 18 = ₹1,620 cr
• Sum-of-the-parts EV = ₹15,552 cr.

Now the trap. A screen calls Sangam "e-commerce" and stamps its commerce peers' 12× onto the whole ₹666 cr of EBIT. That gives ₹7,992 cr. It prices the ₹13,500 cr annuity engine as if it were more thin-margin retail, and so it throws away ₹7,560 cr (a ~1.9× undervaluation). The SOTP is the only thing that stops the biggest revenue line from setting the multiple for a profit stream it doesn't generate.

The lesson. Revenue share shows where the noise is. Profit intensity shows where the value is. And the multiple belongs to the engine's archetype, never the company's label. With one stack table plus one SOTP, a company the market has mis-sorted becomes yours. Info Edge is the living specimen, with Naukri funding 99acres, Jeevansathi and Shiksha. Run this on its actual Ind AS 108 note and verify.

Worked example 6: When growth funds itself: the negative cash-conversion model (US)

Harbor Provisions (synthetic) is a US membership-retail operator. Customers pay by card at the till, and suppliers are paid on terms. The numbers are illustrative, at ~annual scale: revenue $2,000m, COGS $1,600m, inventory $130m, receivables $16m, payables $320m.

Step 1: the cash cycle. DIO = 130 ÷ 1,600 × 365 = 29.7 days. DSO = 16 ÷ 2,000 × 365 = 2.9 days, because card payments settle in days. DPO = 320 ÷ 1,600 × 365 = 73.0 days. Cash conversion cycle = DIO + DSO − DPO = 29.7 + 2.9 − 73.0 = −40.4 days. Harbor sells and collects roughly forty days before it pays for the goods, so its suppliers finance the shelf. It's the cash conversion cycle again, this time as a design choice.

Step 2: net working capital is negative. Net operating working capital = inventory + receivables − payables = 130 + 16 − 320 = −$174m. A negative number on the "capital tied up" line means capital is being supplied. The balance sheet hands Harbor $174m of interest-free funding that scales with the business.

Step 3: grow 25% and watch cash appear. Grow revenue and the working-capital lines 25% together, and net working capital moves to −$217.5m. The change is −$43.5m, so growth releases $43.5m of cash into the year, before a rupee of profit. Amazon used this trick for years, and it's the engine inside every subscription-prepay model. You met it in the value chain and in the "+float" row of the waterfall. Collect-before-you-serve turns expansion into a cash source.

Step 4: the mirror image. Give an otherwise identical twin the opposite sign. Picture an EPC contractor or exporter carrying net working capital of +$174m. That's inventory and receivables it funds while suppliers demand cash. The same 25% growth now consumes $43.5m. Growth is a cash furnace even as the P&L reports profit, which is Phase 2's standing warning. Same income statement, opposite cash life.

The lesson. Two of the five mechanical questions decide whether growth is self-funding or self-immolating: what does the next unit cost, and when does cash move relative to cost. So compute the cash conversion cycle before you admire a growth rate. A negative cycle is a feature engineered into the model. A positive one is a bank loan the income statement forgot to mention.

Practice set

Work in order. P1 to P3 are guided and P4 to P10 independent. P11 to P12 are timed. Figures are in ₹ crore with 25% tax unless stated. Every problem is answered in full, so give each one a real attempt before you read the solution.

P1 (guided, model ID). Classify each into archetype(s) and name the single KPI you'd request first. (a) A company charging AMCs 2.5 to 3.5 bps annually on mutual-fund assets it services. (b) A hotel aggregator taking 18% of each booking. (c) A tyre maker selling through dealers. (d) A credit bureau charging lenders per credit report pulled. (e) A listed company earning 4.5% of the retail sales of 1,400 stores run by others under its brand. (f) An OTT platform at ₹299/month with a mobile-only ad-supported tier.

Solution. (a) Recurring/annuity, usage-flavoured because the fee floats on AUM. The KPI to request is AUM growth + the fee-rate (bps) trend. (b) Marketplace/take-rate. Ask for the take rate and booking (GMV) growth, with contribution per booking as your second request. (c) Transactional (B2B2C, episodic replacement cycle). Request volume growth & realisation, plus channel inventory. (d) Usage-based. Here you want report volumes, which are a credit-cycle derivative. (e) Licensing/franchise. Start with system sales growth (the royalty base), then store adds/closures. (f) A recurring subscription + advertising hybrid. Ask for net adds & churn on the sub side, and engagement/ad-ARPU on the ad side. Notice that each KPI is the live variable in the revenue formula. Classification has to come first.

P2 (guided, waterfall). A retailer has revenue of ₹8,000, gross margin 15%, other opex ₹560, D&A ₹120 and tax at 25%. ΔWC is −₹60, which absorbs cash. Capex is ₹400, of which maintenance is ₹120. Compute EBITDA, EBIT, NOPAT, FCF (total and maintenance-basis), and both FCF conversions against NOPAT.

Solution. GP = 8,000 × 15% = 1,200. EBITDA = 1,200 − 560 = 640 (8.0%). EBIT = 640 − 120 = 520. NOPAT = 520 × 0.75 = 390. FCF = 390 + 120 − 60 − 400 = 50, a conversion of 50/390 = 13%. Maintenance-basis FCF = 390 + 120 − 60 − 120 = 330, a conversion of 85%. So the model converts fine (85%). It's the strategy, ₹280 of growth capex, that consumes the cash. Judge that growth spend with an ROIIC test, because the conversion ratio can't tell you whether it's worth it.

P3 (guided, GMV vs revenue). A food-delivery platform has GOV of ₹12,000. Commissions are 16% of GOV, customer fees 4% of GOV, variable serving costs 12% of GOV, and fixed costs ₹600. (a) Find revenue, take rate, contribution (₹, and as % of GOV and of revenue) and EBITDA. (b) A rival with identical economics runs 1P and reports revenue of ₹12,000 with EV ₹9,600, the same as this firm's EV. Which one "looks" cheaper on EV/Revenue, and what is true?

Solution. (a) Revenue = (16% + 4%) × 12,000 = 2,400, a take rate of 20%. Contribution = 2,400 − 1,440 = 960 = 8% of GOV = 40% of revenue. EBITDA = 960 − 600 = 360 (3% of GOV, 15% of revenue). (b) On EV/Rev, the 3P firm is at 9,600/2,400 = 4.0× and the 1P rival at 9,600/12,000 = 0.8×. The rival "looks" 5× cheaper. In truth they're identical, at EV/GOV 0.8× both, with identical gross economics. Any screen that ranks them differently is measuring accounting policy.

P4 (independent, operating leverage). Fixed costs are ₹240, variable costs 60% of revenue, and revenue ₹800. Compute contribution, EBIT, DOL, EBIT at revenue ±10%, breakeven revenue, and margin of safety. Then recompute DOL at revenue of ₹1,000 and explain the change.

Solution. Contribution = 800 × 40% = 320. EBIT = 320 − 240 = 80. DOL = 320/80 = 4.0. At +10%: 880 × 0.4 − 240 = 352 − 240 = 112, up 40%, which matches 4 × 10%. At −10%: 288 − 240 = 48, down 40%. Breakeven = 240/0.40 = 600, and margin of safety = 200/800 = 25%. At ₹1,000 the contribution is 400 and EBIT 160, so DOL = 400/160 = 2.5. Leverage fades as you climb away from breakeven, so the same business is riskier at ₹800 than at ₹1,000. DOL is a location on the curve, and it moves.

P5 (independent, subscription mechanics). A SaaS firm opens with ARR of ₹500. During the year it loses 8% of opening ARR to churn, expands existing customers by ₹30, and signs new customers worth ₹150 (all ARR). It bills annually upfront. Cash collections are ₹680 and recognised revenue ₹580. Compute closing ARR, gross revenue retention, net revenue retention and ΔDeferred revenue, then state the CFO effect.

Solution. Churn = 40. Closing ARR = 500 − 40 + 30 + 150 = 640. GRR = (500 − 40)/500 = 92%. NRR = (500 − 40 + 30)/500 = 98%, with new logos excluded by definition. ΔDR = collections − revenue = 680 − 580 = +100, so CFO exceeds accrual profit by ₹100 before other items. Customers are financing the firm. Now interpret it. NRR < 100% means the installed base shrinks without new sales, so this "recurring" business must re-earn 2% + growth every year. Compare CAMS-style annuities, at NRR ≈ 100%+ by construction because AUM growth does the expanding.

P6 (independent, advertising incrementals). An ad platform has revenue of ₹1,000, TAC at 21% of revenue, infrastructure & content of ₹190 and opex of ₹280. (a) Find EBIT and margin. (b) Next year revenue rises ₹100, with TAC unchanged in % and zero incremental infra/opex. Find the incremental EBIT margin and the new blended margin, and reconcile them with operating leverage.

Solution. (a) EBIT = 1,000 − 210 − 190 − 280 = 320 (32%). (b) Incremental EBIT = 100 − 21 = 79, an incremental margin of 79% against the blended 32%. New EBIT = 399 on 1,100 = 36.3%. To reconcile, note that TAC is the only truly variable cost here. That makes the contribution margin 79%, so DOL = 790/320 = 2.5, and each 1% of revenue growth adds ~2.5% to EBIT while costs sleep. That's the mechanism behind ad platforms' margins marching up in expansions and slumping disproportionately when ad budgets, the most cyclical spend in commerce, are cut.

P7 (independent, licensing deal value). An Indian pharma out-licenses a molecule. The upfront of ₹80 is certain and paid today. An approval milestone of ₹200 arrives at end-year 4 with probability 35%. If approved, royalties of 8% on partner sales of ₹1,500/yr follow for years 5 to 10. That's six years, which you should treat as year-end flows, with the same 35% probability. Discount at 12%. Value the deal as an expected PV.

Solution. Upfront = 80. Milestone: 200/1.12⁴ = 200/1.5735 = 127.1 × 0.35 = 44.5. Royalties: 8% × 1,500 = 120/yr for 6 years. The annuity factor @12% = (1 − 1.12⁻⁶)/0.12 = (1 − 0.5066)/0.12 = 4.111, so PV at end-yr 4 = 120 × 4.111 = 493.4. Today that's 493.4/1.5735 = 313.6 × 0.35 = 109.7. Total ≈ ₹234. Look at the anatomy. One-third is certain cash and two-thirds is probability-weighted hope. And the P&L in year 4 will either print a ₹200 windfall or nothing. That windfall is rung 5, so don't annualise it. Read every "deal pipeline" slide this way from now on.

P8 (independent, concentration). A precision-components maker has revenue of ₹2,000, an EBIT margin of 14% and a contribution margin of 35%. The segment note discloses one customer = 62% of revenue. Receivable days from that customer are 90 against 45 for others. The contract renews in 18 months. (a) If the customer insources 40% of its volumes, compute the EBIT impact, assuming no fixed-cost relief in year 1. (b) List three questions this disclosure forces, and the pricing consequence.

Solution. (a) Revenue loss = 62% × 40% = 24.8% of revenue = ₹496. EBIT loss = 496 × 35% = ₹173.6, because the contribution vanishes and the fixed costs stay. Against current EBIT of ₹280, new EBIT is ₹106.4, −62% on a 25% revenue hit. Concentration × operating leverage is multiplicative. (b) First, is the company spec-locked or price-locked? That decides who holds redesign costs. Second, what do the 90-day receivables say about bargaining power? They tell you the company is funding its customer. Third, what is the customer's own health and dual-sourcing policy? The pricing consequence is that you underwrite it as a captive supplier. The thesis must survive a renegotiation (annual cost-downs). The multiple must also embed cliff risk, so no rung-1/2 multiple on rung-3 revenue with a single counterparty, whatever today's 14% margin.

P9 (independent, windfall normalisation). In FY24 a chemicals firm has revenue of ₹4,900, including a one-time supply contract of ₹1,400 at 55% EBITDA margin. Core revenue is ₹3,500 at 18% EBITDA margin. Market cap = 20× reported EBITDA. Compute reported EBITDA and the implied valuation error, if the correct approach is 14× core EBITDA plus the after-tax one-time cash (tax 25%).

Solution. Reported EBITDA = 1,400 × 55% + 3,500 × 18% = 770 + 630 = 1,400, a flattered 28.6% blended margin. The market pays 20 × 1,400 = 28,000. The correct value is 14 × 630 = 8,820, plus the one-time cash after tax of ≈ 770 × 0.75 = 578, for ≈ 9,400. So the market cap embeds a ~3× overstatement: peak multiple × peak earnings. It's the Laurus/Zoom/Pfizer pattern reduced to arithmetic, and the whole error is in the classification. The calculation was fine. Rule: growth rates and multiples attach only to revenue that will exist next year.

P10 (independent, profit engine & mix maths). "Trishul Industries" reports three segments. Cables has revenue of 5,200 and EBIT of 780. EPC has revenue of 3,800 and EBIT of 114. Consumer has revenue of 1,000 and EBIT of −50. (a) Build shares and profit intensities, and name the engine. (b) Management guides "revenue to grow 38% by doubling EPC". Compute the new blended EBIT margin with segment margins held constant. Then explain why reported "growth + margin compression" would be neither good nor bad news without this table.

Solution. (a) Total revenue is 10,000 and EBIT 844 (8.44%). Revenue shares are Cables 52%/EPC 38%/Consumer 10%, and EBIT shares are 92.4% / 13.5% / −5.9%. Intensities come out at 1.78 / 0.36 / negative. The engine is Cables, a 15% margin line carrying ~92% of EBIT. (b) When EPC doubles, revenue is 13,800 and EBIT = 780 + 228 + (−50) = 958. The margin is 958/13,800 = 6.94%, down from 8.44%. Optically it reads "+38% growth, −150bp margins". Actually it's just mix, and every segment performed as before. Without the segment table you can't tell mix from deterioration. With it, you also see the question that matters: why is capital going to the 0.36-intensity business? That's the capital-allocation lens meeting the segment note.

P11 (timed, 12 minutes, model-ID sprint). Classify each one, giving archetype(s) + one-line economics + the characteristic leak. (a) A depository charging companies annual issuer fees per folio plus per-transaction charges. (b) A D2C mattress brand that sells 70% via its website and 30% via a marketplace paying 25% commission. (c) An airline. (d) A stock exchange. (e) A movie production house selling satellite + streaming rights per film. (f) A payments firm earning 18bps MDR on card volumes.

Solution. (a) Recurring annuity (issuer fees, rung 1) + usage (transaction charges). The leak is regulation of tariffs, which SEBI administers. That's CDSL's anatomy. (b) This is a seller on a marketplace, which makes it a different animal from a marketplace. It's transactional D2C + a 25% take-rate cost on 30% of sales. Leaks: CAC on its own site + platform dependence, which is concentration read sideways. (c) Transactional and extreme HighFix (high DOL), with perishable inventory, since a seat expires at takeoff. Leaks: fuel (pass-through lag) and fixed costs in downturns. (d) A usage/take-rate toll on trading flow + recurring listing and data fees. Leak: regulated tariffs + volume cyclicality. Beautiful economics, with an administered ceiling. (e) One-off/episodic per-film sales (rung 4 to 5). They aren't recurring even if every year has releases, because each film re-earns from zero. Leaks: hit risk, and receivables from broadcasters. (f) Usage-based, bps on flow, a take-rate on money. Leaks: regulated MDR (India's UPI-zero-MDR lesson) and network fees. Score yourself 2 points per item (archetype + leak). You need ≥10/12 to match the mastery gate's bar.

P12 (timed, 10 minutes, flywheel verification). You have five years of data. ValueMart's revenue goes from 100 to 260 and gross margin from 15.2% to 14.6%. Opex/revenue falls from 9.0% to 7.4%, sales/store rises from 40 to 52, and ROIC from 14% to 19%. Market share goes from 6% to 11%, and marketing spend stays flat in ₹. GlamKart's revenue also goes from 100 to 260, but its gross margin falls from 42% to 36%. Marketing/revenue rises from 18% to 24%, the repeat-purchase rate drops from 61% to 52%, and contribution/order is falling. Give a verdict on each, using the flywheel tests.

Solution. ValueMart = scale economies shared, verified. Gross margin is slightly down by policy while share nearly doubles, so it's passing savings on and isn't losing pricing power. The opex leverage is real (−160bp). Throughput (sales/store +30%) and ROIC (+500bp) are rising, and growth comes without marketing escalation. The loop does the work. GlamKart = buying growth. It surrenders margin (−600bp) while marketing intensity rises (+600bp) and repeat rates fall. The discounts rent customers who don't stay. Falling contribution per order means scale is making unit economics worse. Identical revenue lines, yet one is a compounder and the other a treadmill. The pair shows the lesson in miniature. The P&L totals can't tell them apart, and the model anatomy can.

P13 (guided, SaaS waterfall + SBC realism). A mature SaaS firm has, per ₹100 of revenue, COGS of ₹22, cash operating costs (S&M + R&D + G&A) of ₹50, stock-based compensation of ₹14 and D&A of ₹2. It bills annually in advance (ΔDeferred revenue +₹4), spends ₹3 on capex and pays tax at 25%. (a) Walk revenue to NOPAT. (b) Compute the "adjusted FCF" the company markets by adding back the non-cash D&A and SBC, and its margin. (c) Restate FCF treating SBC as the cost it is. What share of the marketed FCF was actually stock handed to employees?

Solution. (a) GP = 100 − 22 = 78. EBIT = 78 − 50 − 14 − 2 = 12, and tax = 3, so NOPAT = 9.0. (b) Adjusted FCF = 9 + 2 (D&A) + 14 (SBC) + 4 (ΔDR) − 3 (capex) = ₹26, a 26% "FCF margin" the deck will boast about. (c) SBC is compensation settled in shares, an owner cost paid through dilution. Real FCF = 26 − 14 = ₹12, a 12% margin and under half the marketed figure. So ₹14 of the ₹26 (53.8%) was stock handed to employees. It's the Snowflake pattern again, where "adjusted FCF" that adds SBC back counts dilution as profit. Rule: subtract SBC or model the dilution. Never neither.

P14 (guided, operating × financial leverage). A firm has revenue of ₹1,000, variable costs of 40% of revenue, fixed operating costs of ₹450 and interest of ₹50. (a) Compute contribution, EBIT, pre-tax profit, DOL, DFL, and the degree of total leverage DTL = DOL × DFL. (b) Push a ±10% revenue year through and confirm the EBIT and pre-tax swings.

Solution. (a) Contribution = 1,000 × 60% = 600. EBIT = 600 − 450 = 150. Pre-tax = 150 − 50 = 100. DOL = 600/150 = 4.0, and DFL = EBIT/pre-tax = 150/100 = 1.5, so DTL = 4.0 × 1.5 = 6.0. (b) At +10%, contribution is 660 and EBIT = 660 − 450 = 210, up 40%, which is DOL × 10%. Pre-tax = 160, up 60%, which is DTL × 10%. At −10%, EBIT is 90 (−40%) and pre-tax 40 (−60%). The two leverages multiply, so a 10% demand wobble becomes a 60% swing in owner profit. Put a HighFix cost base under financial leverage, as an airline that also borrowed does, and you hold a coiled spring on a coiled spring. Underwrite the down-case first.

P15 (independent, the cohort-maturity mix). A quick-commerce platform runs three vintages of cities. Contribution here means order economics after variable serving cost, before corporate fixed costs of ₹400. Mature cities have GOV of ₹4,000 at +12% contribution. Growing cities have GOV of ₹2,500 at +2%. New cities have GOV of ₹2,000 at −11%, with heavy launch spend. (a) Find blended contribution (₹ and % of GOV) and consolidated EBITDA. (b) What would EBITDA be if the mature book stood alone against corporate fixed costs? (c) Give the unit-economics verdict.

Solution. (a) Contribution = 480 + 50 − 220 = ₹310 on GOV of ₹8,500, or 3.65% of GOV. Consolidated EBITDA = 310 − 400 = −₹90, a reported loss. (b) The mature book alone makes 480 − 400 = +₹80, a profit. (c) The units work fine. The consolidated loss is the cost of opening new cities, laid over a profitable mature book, because a P&L is units × maturity mix. The bear reads "loss-making platform". The analyst reads "profitable at maturity, spending the mature cohort's surplus and more on land-grab." So skip "is it profitable?" and ask the thesis question: "do the new cohorts mature into the +12% the old ones reached?" Cohort curves, coming in the next module, answer it.

P16 (independent, why 'subscription' is not 'safe'). Two US consumer-subscription apps each have 1,000,000 members at $10/month, and both are filed as "recurring revenue." AnnuityCo loses 0.5% of members per month. LeakyCo loses 5% per month. (a) Annual retention of each (compounded). (b) Average customer life (≈ 1 ÷ monthly churn, in months). (c) Gross lifetime revenue per member (ARPU × life). What multiple separates two identical-looking revenue lines?

Solution. (a) AnnuityCo keeps (1 − 0.005)¹² = 94.2% of members, and LeakyCo keeps (0.95)¹² = 54.0%. After a year with no new adds, ~942k vs ~540k members remain. (b) Lives ≈ 1/0.005 = 200 months (16.7 yrs) vs 1/0.05 = 20 months (1.7 yrs). (c) Gross LTV = $10 × 200 = $2,000 vs $10 × 20 = $200, a 10× gap in the value of one dollar of "recurring" revenue. Billing frequency isn't durability. A leaky monthly sub is rung-3 transactional wearing a rung-1 badge. Price the churn and the switching reality, never the word "subscription."

P17 (independent, the rake ceiling). A food-delivery marketplace has GOV of ₹10,000 and a take rate of 18% (commission 13% + consumer/ad fees 5%), for revenue of ₹1,800. Management wants +3 percentage points of take. Route (a) raises commission to 16% (take 21%), and squeezed restaurants push 10% of GOV off-platform. Route (b) adds the 3pp as consumer platform fees and restaurant ads, leaving commission unchanged (take 21%). Sellers feel less pain, so GOV falls only 2%. Compute revenue under each and name the principle.

Solution. (a) Revenue = 21% × (10,000 × 0.90) = 21% × 9,000 = ₹1,890. That's only +₹90 for a 3pp headline hike, and the suppliers are angrier. (b) Revenue = 21% × (10,000 × 0.98) = 21% × 9,800 = ₹2,058, or +₹258, nearly 3× the gain. Disintermediation sets a ceiling on the rake (Gurley's "A Rake Too Far"). Pushing commission invites direct ordering and rivals, which is ONDC's explicit purpose, so the extra rupees leak out through GOV attrition. Where you take the rake matters more than the headline take rate. Elastic sellers walk. Ads and consumer fees are stickier. The attrition figures are assumed for the exercise. The direction is the lesson.

P18 (timed, 15 minutes, stretch, the one-page teardown). "Trident Everyday (synthetic)" is reported as a ₹6,600 cr retailer. Its notes show three streams. Retail stores (own inventory) have revenue of ₹6,000 and segment EBIT of ₹300. The marketplace (3P, a net take on third-party GMV of ₹2,000) has revenue of ₹400 and EBIT of ₹180. Brand licensing (royalty from partner-run outlets) has revenue of ₹200 and EBIT of ₹170. Consolidated variable costs run 82% of revenue, and one institutional customer is disclosed at 11% of revenue. Now work against the clock. (i) Give the archetype of each stream. (ii) Compute revenue-share vs EBIT-share and profit intensity, then name the engine(s). (iii) Explain why the ₹400 marketplace revenue can't be compared with the ₹6,000 retail revenue, and give the correct common base. (iv) Compute consolidated contribution margin and DOL. (v) Say what the 11% disclosure forces and where else to read it. Finish with a one-line verdict.

Solution. (i) Retail is transactional (1P, gross). The marketplace is take-rate (net/agent). Licensing is royalty/franchise. (ii) Total EBIT is ₹650. Retail is 90.9% of revenue but 46.2% of EBIT, at a 5% margin and an intensity of 0.51. The marketplace is 6.1% of revenue and 27.7% of EBIT, an intensity of 4.57. Licensing is 3.0% of revenue and 26.2% of EBIT, at an 85% margin and an intensity of 8.63. So the engines are the marketplace and licensing lines, where 9.1% of revenue (₹600) earns 53.8% of EBIT (₹350). The retail shell is the theatre. (iii) The ₹400 is a net take on ₹2,000 of third-party GMV, while the ₹6,000 is gross own-store sales, so the bases differ. Compare marketplace to retail on GMV (₹2,000 vs ₹6,000) or on gross profit, never revenue-to-revenue. (iv) Contribution = 18% × 6,600 = ₹1,188, and DOL = 1,188/650 = 1.83. It's modest because the thin-margin, variable-heavy retail line dominates the blend. (v) An external customer ≥10% must be disclosed with amount and segment (Ind AS 108 / ASC 280). At 11% it's flag-and-monitor. Cross-read the risk factors and the receivables-concentration note to size the true exposure. Verdict: despite the label, a licensing-plus-marketplace profit engine using a ₹6,000 cr retail operation as its shop window. Value the engines on their own multiples (SOTP), and watch the royalty base and 3P GMV more than consolidated sales. Score (i) to (v) at two points each. Getting ≥8/10 in fifteen minutes matches the Phase 8 teardown bar.

Applied mini-project

The Business Model X-Ray: run the full anatomy on one Indian and one US company, from primary filings.

Choose. In India, the choices are Eternal (Zomato), Info Edge, CAMS, IRCTC*, Jubilant FoodWorks, Sun TV, Dixon, Trent, IEX. In the US, they're Costco*, Netflix, Uber, Alphabet, McDonald's, Apple, Amazon, Snowflake, Otis. (*If you pick IRCTC or Costco, both heavily worked above, you must go beyond the numbers used here to the actual current filings and add at least one insight not in this text.) Pick one company you find boring. Anatomy skills grow fastest without a story to lean on.

Pull (all free). For India, get the latest annual report. You want its revenue disaggregation note under Ind AS 115, its segment note under Ind AS 108 including the ≥10% customer paragraph, and the related-party note for royalties. Add the latest investor presentation and five years of numbers from screener.in. For the US, get the 10-K (Item 1, Item 1A concentration language, revenue note, segment note), the latest 10-Q and the investor day or shareholder-letter KPIs (GMV, ARR, membership, take rate). For both, read the two latest earnings-call transcripts.

Build (per company).

1. The stack table: every revenue stream | archetype | revenue share | estimated gross-profit share | growth | its KPI | cash-timing note.
2. The ₹1/$1 waterfall: walk revenue to FCF with actual numbers and mark the two biggest leaks. Compute FCF conversion (maintenance-basis where splittable) and compare it with the archetype norm.
3. Revenue-share vs profit-share table from the segment note, with profit intensities. Name the engine, then name the engine's KPI and its kill-switch.
4. Quality & concentration audit: place each stream on the five-rung ladder with filing evidence (deferred revenue/RPO/order book/renewals). Extract any ≥10% customer disclosure and translate it into thesis language.
5. Operating-leverage estimate: classify the cost base as fixed or variable (state your assumptions), estimate DOL, and stress EBIT at revenue ±15%.
6. Flywheel verdict: does management's claimed loop pass the four-number verification? Show the numbers.
7. The one-page memo: conclusion first (what is this machine, what is the engine, what would you watch, what would kill it), and max one page per company.

Rubric. Score each criterion 0 = missing/wrong · 1 = partial · 2 = complete and correct, for 16 points in all. You pass at ≥ 13.

File both X-rays in your company knowledge base. The Phase 8 teardown will ask you to produce this analysis in ~25 minutes. This week it may take you six hours. That gap closes with practice.

Reading & resources

• Bill Gurley, "All Revenue Is Not Created Equal: The Keys to the 10X Revenue Club" (Above the Crowd blog, 2011): the canonical essay behind the revenue-quality ladder, with ten characteristics that make a revenue dollar valuable. Read it twice. [Free] [Beginner]
• Bill Gurley, "A Rake Too Far: Optimal Platform Pricing Strategy" (Above the Crowd, 2013): take-rate strategy and the disintermediation ceiling on marketplaces. [Free] [Intermediate]
• Nick Sleep & Qais Zakaria, the Nomad Investment Partnership Letters (2001–2013): the "scale economies shared" letters on Costco and Amazon (the Costco analysis begins in the 2002–2004 letters); the full compilation circulates free (IGY Foundation PDF). Read the Costco sections now; the rest belongs to Phase 6. [Free] [Intermediate]
• Jeff Bezos, Amazon shareholder letters, 1997 and 2001–2005 (aboutamazon.com / SEC): the flywheel and the customer-first cost logic, from the operator who ran it. [Free] [Beginner]
• Ben Thompson, Stratechery, "Aggregation Theory" (2015) and the Business Models compendium: how internet economics reshape the six archetypes. The free weekly articles are enough. [Free/Paid] [Intermediate]
• Damodaran on the Zomato IPO, "The Zomato IPO: A Bet on Big Markets and Platforms?" (Musings on Markets blog, July 2021): a full valuation built from GMV, take rates, and unit economics of an Indian marketplace; disagree with his numbers, learn his structure. [Free] [Intermediate]
• Gassmann, Frankenberger & Csik, The Business Model Navigator (2014; 2e 2020): 55 recurring model patterns; use as a pattern reference, not a bible. [Paid] [Beginner]
• Acquired podcast: "Costco" (2023) and "Amazon.com" (2021): book-length teardowns in audio; the Costco episode is the best single companion to the transactional archetype and the flywheel. [Free] [Beginner]
• Business Breakdowns podcast (Colossus): one company's model per episode; start with the episodes on any company you picked for the mini-project. [Free] [Beginner]
• Filings practice set: Costco 10-K (membership fee line + MD&A gross-margin discussion) · Alphabet 10-K (TAC + revenue disaggregation) · Uber 10-K (gross bookings-to-revenue reconciliation, segment take rates) · Netflix 10-K (content amortisation note) · Eternal quarterly shareholder letters (GOV, take rate, contribution disclosures; the best-disclosed marketplace in India) · Avenue Supermarts AR (MD&A on EDLC-EDLP) · IRCTC AR (Ind AS 108 segment note) · Info Edge investor presentation (billings by vertical) · McDonald's 10-K (franchised vs company-operated margin tables) · Qualcomm 10-K (QCT vs QTL) · Cirrus Logic 10-K (Item 1A concentration) · Otis 10-K (new equipment vs service). All on EDGAR / company IR / BSE-NSE. [Free] [Intermediate]
• screener.in and Tijori (India: segments, KPIs, ten-year numbers) · stockanalysis.com / TIKR (US equivalents): for the mini-project's five-year pulls. [Free] [Beginner]

Going deeper.

• Michael J. Mauboussin & Dan Callahan, "Measuring the Moat" (Morgan Stanley Counterpoint Global, 2016). It's the rigorous, professional version of the value chain and the profit-engine hunt: value-chain mapping, profit-pool and value-added analysis, and how to locate where industry economics actually sit. Read Part I now for the anatomy method. The moat sections belong to M4.04 to M4.05. [Free] [Advanced]
• Madhavan Ramanujam & Georg Tacke, Monetizing Innovation (2016): a book-length treatment of the central lever, how the charge is structured. It shows why the monetisation model is a deliberate design decision and no mere by-product of the product, and how each of the six archetypes gets chosen, packaged, and priced. [Paid] [Intermediate]

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": "Decomposing a hybrid revenue stack into its engines and pricing each on its own multiple (SOTP)", "use": "Feed an LLM the revenue-disaggregation note and segment note (Ind AS 108 / ASC 280) from a filing and ask it to draft the stack-table shell -- stream, archetype, revenue, gross margin, segment EBIT -- so you spend your time on the SOTP judgment call instead of retyping tables from a PDF.", "tools": ["Claude or ChatGPT with the filing PDF or 10-K/annual-report text pasted in", "screener.in / Tijori Finance (India) or stockanalysis.com / TIKR (US) for the historical segment series", "a spreadsheet or notebook for the SOTP arithmetic"], "prompt": "Here is [company]'s segment note and revenue-disaggregation note from its FY[xx] annual report. Build a table: stream | archetype (recurring/annuity, transactional, advertising, etc., per the M4.01 taxonomy) | revenue | revenue share | gross margin | segment EBIT | EBIT share | profit intensity (EBIT share / revenue share). Flag any stream where the filing does not disclose EBIT separately and state what proxy you used.", "verify": "Trace every number in the AI output back to the specific line and page of the filing yourself -- segment notes are exactly the kind of dense, tabular disclosure a model can silently mis-map columns on. Recompute at least the two largest streams' share and profit intensity by hand before trusting the table.", "diy": "The mastery check hands you a segment note with no tool in the room. You must build the stack table, name the profit engine, and compute the SOTP enterprise value by hand -- the AI-augmented version above only removes the retyping, never the judgment of which multiple belongs to which engine." }

Modern Data Analysis

By hand first. The four-step decomposition protocol (revenue-disaggregation note, segment note, MD&A KPIs, one stack table) is unchanged, and it's what the mastery check gates.

Today's workflow. Once you've built one stack table by hand, turn the arithmetic into a small reusable function. Never hand over the classification. A real company rarely has just three streams, and re-deriving revenue/GP/EBIT shares and an SOTP by hand for a six-segment conglomerate is how errors creep in. The python block below is that function, run against Sangam Digital's own numbers as the check.

Tools & sources (IN + US). In India, use the Ind AS 108 segment note and Ind AS 115 revenue-disaggregation note in the annual report, screener.in's segment-wise data, and BSE/NSE filings. In the US, use SEC EDGAR full-text search and the XBRL companyfacts API. There, us-gaap:RevenueFromContractWithCustomerExcludingAssessedTax broken out on the srt:ProductOrServiceAxis / srt:StatementBusinessSegmentsAxis dimensions is the machine-readable version of the same segment note. Use stockanalysis.com or TIKR for the historical series.

# recomputed in-session -- R3. Reproduces Sangam Digital's worked stack + SOTP as a function,
# then re-runs the shares. All inputs are the module's own Sec 4.10 numbers.
streams = {
    "Recruitment listings": dict(archetype="Recurring/annuity", revenue=900,  gm=0.80, ebit=540, mult=25),
    "Online store (1P)":    dict(archetype="Transactional",     revenue=1800, gm=0.18, ebit=36,  mult=12),
    "On-platform ads":      dict(archetype="Advertising",       revenue=300,  gm=0.75, ebit=90,  mult=18),
}

def stack_table(streams):
    total_rev  = sum(s["revenue"] for s in streams.values())
    total_gp   = sum(s["revenue"]*s["gm"] for s in streams.values())
    total_ebit = sum(s["ebit"] for s in streams.values())
    rows = {}
    for name, s in streams.items():
        rev_share  = s["revenue"]/total_rev
        gp_share   = (s["revenue"]*s["gm"])/total_gp
        ebit_share = s["ebit"]/total_ebit
        rows[name] = dict(rev_share=rev_share, gp_share=gp_share, ebit_share=ebit_share,
                           intensity=ebit_share/rev_share, sotp_ev=s["ebit"]*s["mult"])
    return rows, total_ebit

rows, total_ebit = stack_table(streams)
sotp_total = sum(r["sotp_ev"] for r in rows.values())
blended_stamp = total_ebit * streams["Online store (1P)"]["mult"]   # the screen's mistake
# rows["Recruitment listings"]["intensity"] -> 2.70 ; sotp_total -> 15,552 ; blended_stamp -> 7,992

Verify. Running the function above reproduces every number in the Sangam Digital example exactly. Revenue shares come out at 30.0% / 60.0% / 10.0%, gross-profit shares 56.7% / 25.5% / 17.7%, EBIT shares 81.1% / 5.4% / 13.5% and profit intensities 2.70 / 0.09 / 1.35. SOTP EV is ₹15,552 cr against a blended-multiple mis-price of ₹7,992 cr, a 1.95× gap, which matches the earlier "~1.9× undervaluation." The function generalises the by-hand method. It can't make the judgment of which multiple belongs to which stream.

Quantitative lens

DOL across a full revenue range. Practice-set P4 shows DOL moving between two revenue levels. Compute the same HighFix twin (fixed ₹600, 30% variable, breakeven ₹857) in-session across a wider range and you can see how sharply DOL decays as revenue moves away from breakeven:

The ₹1,000 row reproduces the DOL = 7.0 from the operating-leverage section (verified in-session). Here's the practical use. A HighFix business trading near its own breakeven (₹900 cr, DOL 21×) is a wholly different risk object from the identical business at ₹2,000 cr of revenue (DOL 1.75×). One bad quarter is existential for the first and a rounding error for the second, even though nothing about the cost structure changed. So don't screen on a single trailing DOL number without asking how far from breakeven the company is. That's the same error as screening on trailing P/E at the top of a capital cycle. Both readings depend entirely on where you're standing.

Where the method breaks. The DOL/breakeven split assumes a clean fixed/variable partition. Real cost structures have step-fixed costs, like a new distribution centre or a second shift, that turn the smooth curve above into a staircase the formula doesn't see. Semi-variable costs (partly-fixed staffing) blur the split further. Treat any single DOL figure as an estimate, bracketed by your fixed/variable classification choices. It's never a measured constant. The SOTP method is even more exposed. It's only as reliable as the peer multiples fed into it, and picking "the right" multiple for each engine is a judgment call this arithmetic can't automate. The function above computes the SOTP correctly for whatever multiples you give it. But assigning 25× to the recruitment engine is an assertion about that engine's durability, which the CAP machinery of the moats work is built to test. The segment note doesn't contain it as a fact.

Where this goes next: galaxy cross-links

• G3 · CN5.01 (business-model-teardown): the consulting-side teardown protocol that the stack-and-engine method here feeds directly into.
• G2 · DA1.02 (pandas-data-wrangling): turns the python block above from a three-stream toy into a real multi-segment pipeline over ten years of filings.
• X1 · AI0.07 (ai-augmented-analyst): the general discipline behind the AI-Augment section above: what to delegate, what to keep, and how to verify either way.

Teach it back & journal

Feynman prompt. Take a smart 15-year-old to your school's canteen. Using only things she can see, explain all six ways the canteen could charge. It could sell samosas one by one. That's transactional, and it has to earn her ₹15 again tomorrow. It could sell a monthly meal card, a subscription where her not cancelling is the product. It could let outside stalls sell and keep 10% of each sale. That's a marketplace, selling footfall-liquidity instead of food. It could put a poster wall by the queue and charge brands for it. That's advertising. You manufacture attention, and it's ~free at the margin. It could let another school use your famous canteen's name and recipes for 5% of their sales. That's licensing, with their capital and your margin. Or it could install a juice machine that charges per ml, and with usage pricing the meter is the price. Then ask her the killer question and make sure she can answer it: if the meal-card line is only a tenth of the money coming in but three-quarters of the profit, which line should the canteen's owner lie awake protecting? And what single rule change by the principal could break it? If she names the card line and comes up with the fee-cap risk unprompted, you've taught it well. Finish by writing the one-page explainer: "Every business is one of six charging machines, or a stack of them. Each machine leaks money in its own places on the way from revenue to cash, breaks in its own way, and hides its engine in a different room. Anatomy first; admiration later."

Journal reflection. Choose a company whose product you personally used today, paid or "free". Before opening any filing, write down from memory its stack (which archetypes), your guess at the revenue shares, and your guess at where the profit engine is. Then pull the actual revenue-disaggregation and segment notes and score yourself. Record three things. (1) The biggest gap between your intuition and the filing, and why the surface of the product concealed it. (2) Which archetype you instinctively over-admire: subscriptions, flywheels, asset-light royalties? That's the beginning of the bias inventory Phase 6 will formalise. (3) One sentence you couldn't have written a week ago. File the stack table in your company notes. The Phase 8 teardown template has a slot named "model stack" waiting for it.

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