Business-Model Teardown & Unit Economics: Advisory Lens

Branch: Business Evaluation & Commercial Due Diligence (Consulting galaxy) · Node CN5.01 · ~10 focused hours · Mastery gate ≥ 85%

An equity analyst and an advisor can be handed the identical spreadsheet (the same contribution ladder, the same CAC and churn, the same cohort triangle) and produce two entirely different documents. The analyst writes "unit economics are weak, avoid," files it, and moves to the next name; nobody in the building is obligated to act on the sentence today. The advisor is in the room on Thursday, across the table from a founder who has already spent the number in their head, or a board that has to vote, or an investment committee that closes on Friday. What has to come out of your mouth there is an instruction for real capital rather than a verdict on a stock: scale it, fix it first, or walk away, plus the one number that would change your mind, because someone across the table is about to ask for it. That instruction, defensible under pressure, is what this module trains. The arithmetic is not new; the business-model and unit-economics modules already own it cold. What is new is the six inches between "the numbers say X" and "here is what you do about X, and here is what I'd need to see to change my mind", and that six inches is a partner's entire job.

Ten hours from now, one sentence should read as obvious: unit economics tells you whether the machine works; advisory judgment tells the client, this quarter, whether to fund it, fix it, or leave, and names the exact evidence that would flip the call. A business-model teardown that stops at "CAC is too high" or "cohorts are decaying" has done half the job, the mechanically correct half, and the half the programme's unit-economics module already trained to reflex. The other half is structural. Route the diagnosis to the right layer: is it the unit, the machine, or the market? Score the whole picture against an explicit rubric rather than one flattering or damning metric, and sequence the fix under a real budget and a real calendar. Then say the hard part out loud, in a way the room can actually hear and act on. Advisory work is not "more finance." It is finance with a decision welded onto it, delivered on a deadline, to a person who will spend real money based on what you say next.

One promise for the hours ahead. By the end you will be able to take a data room you have never seen: a founder's deck, a target's raw operational export, a board's growth ask. From it you recompute the four numbers that actually decide the question (never the deck's own numbers), route the diagnosis through Unit → Machine → Market, and score it against the five-dimension Scalability Verdict Rubric. Then you prioritize the fix among competing levers under a real constraint, and deliver a one-paragraph recommendation with a named condition that would change your mind. You will do this on a quick-commerce grocer, a subscription-apparel brand, a fitness-studio chain, a B2B SaaS company, a QSR chain, and a subscription box (three in rupees, three in dollars) and by the sixth you will have stopped needing the training wheels. That reflex (recompute, route, score, prioritize, recommend) is the purple belt of the Engagement Manager, and it is the exact thing a case team lead expects from you inside a single meeting.

Learning objectives

By the end you can:

1. State in one sentence why an advisor's use of unit economics differs from an analyst's, and explain what "judge quality and scalability, not just report them" concretely demands beyond arithmetic that happens to be correct.
2. Recompute (never accept on the deck's or the client's own say-so) a business's contribution ladder, CAC (three ways), honest discounted LTV, and cohort-level retention from data-room-grade raw inputs, applying M4.01/M4.02's mechanics without re-deriving them, and use only the recomputed figures as the basis for every judgment that follows.
3. Route any business-model question through the Unit → Machine → Market diagnostic tree, correctly identifying which layer is broken, and hand off cleanly to CN5.02 or the moat modules (M4.04–M4.05) when the true problem sits outside this module's remit rather than forcing a unit-economics fix onto a market-level disease.
4. Apply the Scalability Verdict Rubric (five dimensions, scored) to render one of three defensible verdicts (Scale, Fix-First, No-Go) each attached to the specific evidence that would flip it, rather than treating the rubric as a formula that outputs a verdict by itself.
5. Prioritize among competing levers (price, CAC efficiency, retention, cost structure) under a real constraint (a fixed budget, limited management bandwidth, a fundraising clock) using an explicit impact × cost × time-to-effect scoring, and defend a sequence, not just a ranked list.
6. Read operating leverage as a scalability risk, not only a margin fact: compute DOL and margin of safety, stress-test a demand shock in both directions, and translate the result into a phased-financing or covenant recommendation rather than a flat go/no-go.
7. Recognize, in a client's or founder's own presented numbers, the M4.02 §4.13 games catalogue in the wild, and raise it in the room by naming the specific definitional gap, never the person, while pairing the hard finding with a constructive, fundable alternative ask.
8. Compress a full teardown into an answer-first governing thought fit for a steering-committee slide (cross-link CN1.03/M10.03; this module never re-derives the pyramid) and state precisely where your job stops and deck craft (the future Presentation galaxy) begins.
9. (AI-Augmented Consultant: R10.) Use AI to accelerate a first-pass read of a data room, drafting a candidate rubric fill-in, flagging likely games-catalogue patterns, proposing diplomatic phrasing for a hard finding, while tracing every figure it touches to your own recomputation (Guardrail 1–5) and owning the verdict yourself; AI proposes, you dispose.

Prerequisites & connections

Builds on. You need two bodies of mechanics to be yours already, cold, by hand. From M4.01 you need business-model anatomy, the six revenue archetypes, the revenue-to-FCF waterfall and the formal treatment of operating leverage. From M4.02 you need unit economics in full: choosing the unit, the CM1/CM2/CM3 ladder, CAC three ways, the churn algebra, discounted LTV, cohort retention triangles, the two flagship unit-P&L builds and the games catalogue. If any formula below looks unfamiliar (DOL, LTV = m·r/(1+d−r), the CM ladder, the reconciliation test), stop and complete M4.01 and M4.02 first. Everything here cites those formulas at every turn and never re-derives them. You'll also lean on CN0.01's honest-scope contract and its structure-hypothesis-answer-first habit, plus the issue-tree, hypothesis and synthesis reflexes from CN1.01 to CN1.03. The routing framework below is an issue tree. And every worked example ends on a verdict delivered answer-first, because CN1.03 already made that the only acceptable way to deliver one.

M4.01 and M4.02 teach the mechanics and reward getting the arithmetic right. Here you lean on them without re-teaching them. The arithmetic is assumed to be right already (or you've just caught that it wasn't), and the question becomes what to do about it under a deadline, in a room, with a specific ask attached. When a worked example needs a number, it gets recomputed once and cited to its source formula. Then the page moves straight on to the judgment the number demands. The ratio between "here is the LTV/CAC" and "here is why that means Fix-First, and here is the churn target that flips it to Scale" is roughly one line to twenty.

Sometimes the diagnosis proves to be about market structure, size or a shifting profit pool, and the business's own machine is fine. That's CN5.02's question, and you hand off before you force a unit-economics answer onto a market disease. If it's about why good unit economics might persist against competition, that's the moat question of M4.04 and M4.05. You judge quality and scalability today, and durability over a decade belongs to them. When the diagnosis is one workstream inside a full engagement, CN5.03 is where it lives, alongside a market model, voice-of-customer and a formal go/no-go with risks and mitigations.

Feeds forward. CN5.02 picks up where the teardown deliberately stops, at market size, structure and competitive dynamics, and the routing framework below hands off to it by name. CN5.03 slots your recompute-and-verdict work in as one workstream of a full commercial due diligence, next to the market model and voice-of-customer that aren't taught here. CN6.01 and CN6.02 reuse the same lens, unit economics as advisory judgment, to assess a target's standalone quality before asking whether a deal or a synergy case makes it better. CN7.02's steering-committee craft is where the answer-first governing thought you learn to write here gets delivered, live, in a room, under pushback. You may already have met the taster in the finance track, E11.03 (the Consulting Case-Interview Lab) and CL-11.3 (Kite Foods). That was the on-ramp. The teardown is the first deep application of its instincts to a real engagement type.

The whole difference fits in one line. An analyst who finds broken unit economics writes "avoid" and closes the file; an advisor in the same seat has to tell a room, today, exactly what to do about it (scale, fix, or walk) and name the one number that would change the answer.

Core teaching content

The advisor's question is not the analyst's question

Put the same finding in front of both seats. A growth-stage company's LTV/CAC, properly recomputed, is 1.9×, well under the 3× bar, while its CAC payback is a healthy five months. An analyst's report can sit on that sentence indefinitely. It's true, it's complete, and nobody has to act on it by Friday.

An advisor can't leave it there, because the room's next question is always the same: so what do we do? You built that reflex when you studied the "so what" of synthesis, and a business-model teardown is one place you put it to work. "So what" forces four further commitments that a correct number never makes by itself.

First, name the layer that's broken. It might be a fixable unit-level problem or a machine-scaling problem, or it might be a market problem wearing a unit-economics costume. Sorting that out is the routing question below. Second, name the verdict. An adjective like "concerning" isn't one. The verdict is one of a small, named set of decisions: Scale, Fix-First, No-Go. Third, if there's more than one lever, say which goes first. Work to the budget and calendar the client actually has, and forget the unlimited one a model assumes. Fourth, name in advance the evidence that would flip the verdict.

A recommendation with no stated falsifier is an opinion wearing a recommendation's clothes.

Notice what's missing here. There's no new formula, no new arithmetic technique, nothing that competes with the unit-economics mechanics you've already done. Every number in every worked example below comes from methods you already know. All of the value is in the six inches after the number. Those six inches are worth the entire fee, so they get far more room here than a paragraph tacked onto the end of the mechanics.

The mechanics you already own: a one-screen crib, not a re-teach

Use this only as a lookup while you work through the examples. If a line needs explaining, go back to the business-model and unit-economics modules, because that's where the explaining happens.

On top of the crib you add two habits, and everything below depends on them. First, recompute before you read the deck's number, every time, even live in the room. The games catalogue you met in unit economics doesn't wait politely outside just because the person presenting the number is sitting across the table instead of buried in a filing. Second, every number above is an input to a decision, never the decision itself.

Everything from here on is about what you do with the row once the row is right.

Where this work actually happens: the engagement contexts

A teardown isn't one thing. The same recompute-and-verdict skeleton wears four different costumes depending on who's paying for the answer, and the costume changes what "done" looks like.

• In a growth-strategy diagnostic, a founder or board is asking "should we raise and scale?" The client controls the ask. Your job is to size what the ask should be and what it should be conditioned on. RozanaMart and Tapri Express below wear this costume.
• In commercial due diligence for an investor, a PE fund or growth-equity committee is deciding whether to fund someone else's company. You don't control the target's cooperation, and the data is whatever the data room contains. Your verdict feeds a term sheet instead of a strategy memo. Loomwell, Larkspur Metrics and Halcyon Box below wear this costume, and the full version gets its own treatment in the due-diligence toolkit later in the branch.
• In a turnaround or operational-improvement engagement, the business is already scaled and needs its economics repaired. Its size isn't the question. The unit questions are the same, but the constraint is usually cash runway instead of growth capital. FlexFit Studios below wears this costume.
• In board or founder advisory outside a formal engagement, a standing advisor gets a pointed question between meetings. It's the same method, compressed into a paragraph instead of a deck, and the Practice set's shorter items are written in this register.

Across all four, one thing never changes: the deliverable is a decision. A data room gives you spreadsheets. Public filings, the usual raw material in the finance track, give you disclosure notes. A founder's deck gives you a story that wants to be believed. All three go through the same recompute before a single word of advice gets written.

Every company you'll meet here and in the practice set is synthetic and internally consistent, because that's the defamation-safe way to practise. The method carries over unchanged to a real target with real, cited numbers.

The routing diagnosis: Unit, Machine, or Market?

Before you touch a lever, ask which layer is actually broken. Recommending a unit-economics fix for a market-level disease, or the reverse, is the most expensive diagnostic error an advisor can make. It's the mistake Rumelt's "diagnose before prescribe" spine exists to prevent. Ask the three questions in this order, because each one, answered yes, ends the search:

1. Is the unit itself broken? Recompute contribution, CAC, discounted LTV and payback for the smallest repeatable unit, chosen the way you learned in unit economics. If contribution is negative, or LTV/CAC is still well under the archetype's healthy band after a careful recompute, the problem is here. No amount of scaling, market tailwind or moat fixes a unit that loses money on its own terms. That's a Unit problem.
2. If the unit works, does the machine hold together at scale? Check operating leverage (DOL, margin of safety, breakeven) and the maturity mix across cohorts or stores. Then check whether the reconciliation test (unit × count + corporate ≈ reported) actually ties out. A business can have a profitable unit and still be a bad scale-up candidate. Its fixed-cost structure may mean growth compounds risk faster than it compounds profit, or its young vintages may be structurally worse than its old ones. That's a Machine problem.
3. If the unit and the machine both hold up, is the trouble in the market the machine sells into? A shrinking category won't show up in a unit P&L. Neither will a commoditising profit pool, or a structurally worse position than a rival's. No amount of unit-economics polishing fixes a market that's disappearing under the business. That's a Market problem, and it belongs to the market-attractiveness work that picks up where the teardown stops. The right advisory move is to say so and hand off. Don't force a unit-level fix onto a market-level disease. Practice P12 drills this hand-off.

The tree is ordered cheapest-to-check first, on purpose. Unit-layer arithmetic takes an afternoon with a spreadsheet. Machine-layer checks take a few cohort exhibits and a stress test. Market-layer work needs the full sizing and structural analysis that the market-attractiveness module owns.

Never skip a layer because the client's own narrative jumps straight to one.

A founder who says "we just need more marketing budget" is diagnosing a Unit problem as a Machine problem. Picture a board that says "let's just fix retention" while competitors are commoditising the category. That board is diagnosing a Market problem as a Unit problem. Run the tree yourself before you agree with either of them.

The Scalability Verdict Rubric

Once routing puts the question at the Unit or Machine layer, you render a verdict on five dimensions. Never let one flattering or damning metric stand in for the whole picture. Each dimension gets a Green / Yellow / Red score against named evidence:

Score all five before you read across them. The whole point of the rubric is that a single dimension never decides the verdict alone. It gives your judgment a structure, and it won't output an answer by itself.

Take a business that's Green on unit economics and cohorts, Yellow on operating leverage, Green on runway and Yellow on moat. It can still earn Scale, because each yellow is manageable with a named condition. A tranche structure protects against the operating-leverage risk, and a covenant protects against the moat uncertainty. Another business can be Green on four dimensions and Red on cohorts and still earn No-Go. A decaying cohort trajectory is a leading indicator that the other four are about to deteriorate too.

Read the five together and render one of three verdicts:

• Scale. Fund the ask, generally with named conditions attached to whichever dimension scored Yellow (a tranche, a covenant, a milestone). It's never a blank check.
• Fix-First. Don't fund the scaling ask yet. Name the specific metric and threshold that must move, and over what time horizon, before the ask comes back.
• No-Go. The economics don't support the ask at any plausible fix within the client's own timeline and constraints. Say so plainly, and say what would have to be different at the root for the answer to change.

{
  "skill": "Drafting a first-pass Scalability Verdict Rubric fill-in from a data room",
  "use": "Once you have recomputed the unit economics yourself, use a chat or document-grounded AI assistant to draft a first-pass read of the five rubric dimensions from the data room's own materials (cohort exhibits, board decks, KPI dashboards) — a fast way to make sure you have not overlooked a disclosed metric, before you commit to your own scores.",
  "tools": ["Chat assistants — Claude, ChatGPT, Gemini", "Document/PDF AI grounded on the data room's own files (never a bare model's memory)"],
  "prompt": "Here is a data room excerpt: <PASTE>. Using only what is in this material, draft a first-pass read of these five dimensions: unit economics health, cohort trajectory, operating leverage, reinvestment runway, and moat/durability. For each, cite the exact exhibit or line you drew it from, and flag anything you could not find evidence for rather than guessing.",
  "verify": "Every figure the assistant cites must be traced back to your own recomputation, never accepted as read off the deck (Guardrail 1-5: trace, match unit/scale/period/entity, recompute, date-check, cite). The five scores and the overall verdict are yours to render — AI can save you a re-read of the data room, never the judgment of what a Yellow on operating leverage means for this specific client's ask.",
  "diy": "You must already be able to compute all five dimensions by hand and know what evidence would flip each one — that is exactly what this module gates. AI accelerates the first pass through the paperwork; it cannot install the judgment of what the five scores mean together.",
  "market": "US"
}

Prioritizing the levers: impact × cost × time-to-effect

Once the verdict is Fix-First, or Scale with conditions, the client rarely has just one lever. There are usually two or three candidates: raise price, cut CAC, improve retention, fix a cost line. And there's a limited budget, a limited window of management attention, or both. The rookie move is to recommend whichever lever has the biggest steady-state number. The advisory move is to sequence under the constraint the client really has. Score every candidate lever on three axes.

• Impact is the incremental annual contribution once the lever is fully in effect, computed the unit-economics way: recomputed, never assumed.
• Cost is what it takes to execute, in money and in management bandwidth.
• Time-to-effect is how long until the impact is real and no longer just modelled. A lever that takes nine months to ramp competes for the same limited runway as one that takes a month.

A useful rough score is impact ÷ cost, read alongside time-to-effect as a separate risk adjustment. A high-ratio lever that also lands fast should almost always go first, because it funds everything after it. The lever with the largest steady-state impact is often correctly sequenced second or third. Its ramp is slow, and its near-term net effect on cash may be negative. The Tapri Express example below computes this.

What you deliver is the sequence: which lever goes this quarter, which follows once the first is funding it from cash flow, and which gets deferred entirely until a dedicated budget exists outside the current ask.

Advisory skepticism: reading a data room the way the games catalogue trained you to, then saying so in the room

Back in unit economics you catalogued five families of games that a company's own presented numbers play on a reader: allocation games, blended masking, growth-spend reclassification, denominator and survivor games, and the missing balance sheet. None of that scepticism gets switched off because the number now comes from a founder's mouth in a live meeting instead of a filing's footnote.

If anything, a live pitch is more exposed to the games. A founder pitching a raise has every incentive to show the flattering cut of the data. And the meeting format discourages the line-by-line interrogation that a written filing lets you do at your own pace.

Two habits make this workable in a room without torching the relationship. First, name the definitional gap, never the person. Try "this looks like it's blending organic customers into the CAC, which is common and not a sign of anything except that the marginal-paid-CAC number needs its own line." It comes across very differently from "your CAC number is wrong," even though both sentences describe the same recomputation.

Second, never deliver a hard finding without a constructive next ask. That might be a smaller proof-of-concept tranche or a specific two-quarter milestone. It might be a request for the raw cohort export in place of the summarised chart. The finding is technically the same one an equity analyst would write. The delivery isn't, because the room across the table is going to act on what you say in the next thirty seconds. A recommendation that says "no" without saying "here is what would make it yes" hasn't done the advisory half of the job.

{
  "skill": "Drafting diplomatic phrasing for a hard finding before a client meeting",
  "use": "After you have recomputed the real numbers and identified the specific definitional gap in a founder's or target's presented metric, use a chat assistant to draft two or three candidate phrasings that name the gap without naming the person — then choose and sharpen the one that is both accurate and deliverable in the room.",
  "tools": ["Chat assistants — Claude, ChatGPT, Gemini"],
  "prompt": "I recomputed a company's presented LTV/CAC and found it uses <SPECIFIC GAP, e.g., blended CAC while paid spend is scaling, or survivor-basis retention instead of whole-cohort>. Draft three ways to raise this in a client meeting that name the definitional gap, not the person, and each end with a specific constructive next step I could offer instead of a flat no.",
  "verify": "The recomputed number and the named gap are yours, verified against the raw data before the meeting (Guardrail 1-5) — never let the AI soften the finding itself, only the delivery. If a drafted phrasing understates what you actually found, discard it; diplomacy is about how the fact lands, never about what the fact is.",
  "diy": "You must be able to identify the games-catalogue pattern and recompute the honest number yourself before any phrasing question is even relevant — that is this module's and M4.02's gate. AI can smooth the sentence; it cannot decide what the sentence has to say.",
  "market": "IN"
}

From verdict to governing thought: the hand-off to answer-first communication

A finished teardown gives you the recomputed numbers, the routing diagnosis, the five-dimension rubric and the sequenced lever recommendation. A steering committee can't act on any of it until it's compressed into a governing thought, delivered answer-first. The verdict comes first, then the two or three reasons beneath it, the named risk and its mitigation, and the concrete next step.

That compression (finding versus insight, the MECE key line, answer-first delivery) is the whole subject of the "so what" synthesis work. The finance track's communication toolkit teaches it again as a writing craft, and neither gets re-derived here. Your job is to make sure the governing thought rests on a teardown that's correct and complete: the right layer routed, the right rubric scored, the right lever sequenced. Then the synthesis has something true to compress.

Once the governing thought and its key line exist, deck production (which chart, which layout, which slide sequence) belongs to the future Presentation track. The teardown stops at a titles-only storyline, which is the same point where the synthesis work hands off.

The honest limit: what this module cannot give you

The consulting track says this plainly at its very start, and it holds here too. The checklist above (recompute, route, score, prioritise, recommend) is fully teachable, and by the time you finish here it'll be yours cold. It won't be fast yet. You won't have the instant pattern-recognition of someone who has rendered this verdict five hundred times and can smell which Yellow is really a Red before the spreadsheet confirms it.

That judgment compounds only from live engagements. Real clients, real money, real consequences when the verdict is wrong. Nothing you read, however well built, replaces those five hundred reps. What you get here is the floor everyone eventually builds that judgment on: the structure that keeps your first ten verdicts defensible even before they're fast.

Common mistakes & how experts think differently

1. Reporting the numbers instead of rendering a verdict. "Unit economics look weak" is a sentence an analyst can leave on a page. "Recommend Fix-First, conditional on mature monthly churn clearing 5% within two quarters" is a sentence an advisor can hand to a room. If your teardown ends on an adjective instead of one of the three named verdicts, it isn't finished.
2. Reaching for "scale faster" before running the routing diagnosis. Recommending more growth capital before you've checked whether the unit works inverts Rumelt's diagnose-before-prescribe order. Of all the sequencing errors you can make in a teardown, it's the most expensive. Fund a broken unit at speed and it just loses money faster.
3. Accepting a client's or founder's own presented metrics because they arrived in a live meeting instead of a filing. The games catalogue doesn't know or care whether the number came from a footnote or a founder's mouth. Recompute both the same way, every time.
4. Treating the Scalability Verdict Rubric as a formula that outputs a verdict automatically. Two Yellows in different dimensions can be entirely fine together, like a manageable operating-leverage risk alongside a contestable-but-real moat. They can also be fatal together, like a decaying cohort trajectory alongside a narrowing runway. The rubric structures the judgment call. It doesn't make the call for you.
5. Recommending the biggest lever instead of sequencing under the client's real constraint. The lever with the largest steady-state annual impact is often the wrong one to fund first, because its ramp is slow and its near-term net effect on cash is negative. You'll watch this trap play out in the Tapri Express example.
6. Presenting a point estimate as if it were certain, instead of naming the range and what would move it. Say a recommendation states "the verdict is Scale" without adding "and here is the one number that would flip it to Fix-First." That's false precision wearing confidence's clothes.
7. Skipping the stress test before the recommendation goes in the room. A recommendation that only survives the base case is a hope with a spreadsheet attached. Always run the downside before you commit to a verdict. Where it's relevant, run the more pessimistic elasticity or the more aggressive rollout too.
8. Confusing a Market-layer problem for a Unit-layer one, or the reverse. Maybe you recommend a pricing or retention fix when the disease is a shrinking or commoditising category, which is a market question. Or you recommend a market pivot when the unit is simply mispriced. Either way you've spent an engagement's worth of the client's attention on the wrong layer.
9. Delivering the hard finding as an accusation instead of a fact pattern with a next step. Name the specific definitional gap: "blended versus marginal CAC," never "you were misleading us." And always pair bad news with a constructive, fundable alternative ask. That's what keeps the relationship, and the engagement, alive after the finding lands.
10. Forgetting the honest-scope line. Treating your first ten verdicts with the instant confidence of a partner on their five-hundredth engagement is itself a failure of judgment. You can learn the checklist today. The pattern-recognition that makes it fast isn't teachable yet, and pretending otherwise is the fastest way to be wrong with total conviction.

Pull one thread through all ten. An expert never lets a correct number stand in for a finished recommendation. Each number is a means to a verdict, a sequence and a named condition for changing your mind, delivered to a person who's about to act on it.

Worked examples

Worked example 1: Contribution and CAC/LTV as a scaling decision (India, ₹)

RozanaMart is a quick-commerce grocery app operating in four Indian cities. You're the board's advisor, and you've been asked for a verdict on a Series C ask to fund expansion from 4 cities to 20 cities. You get raw order-level and cohort data in the data room. The deck's own slide claims "unit economics are proven, CAC ₹280, payback under 6 months, ready to scale 5×."

Solution. Step 1, recompute the contribution ladder, never the deck's summary. Start with one order. Net revenue is ₹400 and product COGS is 65% (₹260), which leaves CM1 = ₹140 (35%). Fulfilment (delivery, packaging, payment gateway, support) costs ₹100 (25%), so CM2 = ₹40 (10%).

Marketing blends two very different rupees. Newly acquired customers place 30% of orders, and each of those carries a ₹120 CAC-equivalent marketing cost. Repeat orders make up the other 70%, carrying a ₹10 voucher. Blended marketing = 0.30×120 + 0.70×10 = ₹43, so CM3 = ₹40 − ₹43 = −₹3.

At the blended level, that's a loss.

Now split it. First-order CM3 = 40 − 120 = −₹80, and repeat-order CM3 = 40 − 10 = +₹30. Solve for the repeat share that would clear zero: f·30 + (1−f)·(−80) = 0 → f = 72.7%. Today's repeat share is 70%, so the order book is just under its own breakeven. It's a close call, and it isn't a clean pass.

Step 2, recompute CAC and LTV; do not inherit the deck's ₹280. The deck's figure blends organic customers into the number. Recompute the fully loaded, marginal paid CAC from the data room's own channel split and you get ₹450. A retained customer orders roughly 3 times a month, so monthly contribution (net of the ongoing repeat voucher) = 3 × (40 − 10) = ₹90.

For churn, read the flat part of the oldest cohorts and ignore the deck's "active user" definition. Mature-cohort monthly churn is 8%. Naive lifetime = 1/0.08 = 12.5 months, and naive LTV = 90 × 12.5 = ₹1,125. Discount it properly (r = 0.92, d = 1.5%/month): LTV = 90 × 0.92 / (1.015 − 0.92) = ₹872. That gives LTV/CAC = 872 / 450 ≈ 1.9×, well under the 3× bar.

Payback = 450 / 90 = 5.0 months, which really is fast. It's why the deck's slide isn't lying so much as quoting the flattering half of a picture that points two ways.

Step 3: route the diagnosis. Fast payback and weak LTV/CAC together mean one specific thing. RozanaMart recovers its cash quickly on the first few orders because churn is brutal, and the customer leaves before the lifetime value has a chance to compound. So it's a Unit problem, and the unit's problem is churn. Geography has nothing to do with it. No number of new cities repairs it. Scale a leaky unit and it just leaks in twenty places instead of four.

Step 4: verdict and recommendation, answer-first. Fix-First, not Scale. "Do not fund the full 20-city ask this quarter. Fund a smaller bridge round to run a retention-focused pilot in the existing four cities for two quarters. Unlock the geographic capital only if mature-cohort monthly churn falls from 8% to 5% or better and honest LTV/CAC clears 3×. The fast payback is real and reduces your downside risk during the pilot. It isn't evidence the unit works at scale, only evidence it's cheap to test." The named condition (churn ≤ 5%/month, LTV/CAC ≥ 3×) is what turns this from an opinion into a recommendation.

Read it. The deck wasn't fabricating a number. It quoted the metric that happened to look good, payback, and stayed quiet about the one that didn't, LTV/CAC. Your job was never to catch a lie. It was to recompute both metrics, notice that they disagree, and explain to the board why they disagree and what that means for a bet five times bigger.

Worked example 2: Cohorts and the survivor-basis trap in a live diligence (US, $)

Loomwell, a US direct-to-consumer apparel-rental subscription, is the target of a growth-equity firm's $40M investment. The deal team's investment memo says "Cohort NRR of 118%, we're comfortable with the growth story." Before the investment committee votes, you're asked to check that figure against the raw cohort exhibit.

Solution. Step 1, recompute contribution and the retention curve. ARPU is $90/month at a 60% contribution margin, so each active subscriber brings in $54/month of contribution. Track the January cohort of 4,000 new subscribers by logo retention and its path for M0 to M6 reads 100, 80, 70, 64, 60, 57, 55. Month-over-month churn runs 20%, 12.5%, 8.6%, 6.25%, 5.0% and then 3.5%. Plot it and you get a clean curve, flattening towards a habit floor. You can't see a diagonal calendar-shock stripe across the sibling cohorts' M1 columns. There's only a mild, ordinary vintage softening, which fits a growing acquisition base. It's the mechanism you already know from unit economics, applied here and not re-taught.

Step 2, reconcile the deck's 118% against the whole cohort, not just its survivors. At M6, 55% of the original 4,000 are still there (2,200 subscribers). Their average revenue really has risen, from $90 to $106.20, an 18% expansion via plan upgrades. Among survivors only, it's the 118% the memo cites. But NRR has to be measured on the whole original cohort, and the survivors are only part of it. Whole-cohort revenue retention = (2,200 × $106.20) ÷ (4,000 × $90) = $233,640 ÷ $360,000 = 64.9%.

Same cohort, two very different stories.

The deal team's 118% and the whole-cohort 64.9% are two different computations, both internally consistent, and only one of them is the metric its own name claims to be.

Step 3: route and name the consequence. Nobody fabricated anything. You're looking at the most common games-catalogue pattern in a live pitch, abuse #8, survivor-basis measurement, applied to a retention metric instead of an LTV. And the consequence is structural. Loomwell's headline subscriber count keeps climbing because ever-larger new cohorts arrive faster than the old ones melt, which is the blended-masking pattern. Both stories are real, the topline growth story and the true unit-economics story, and they point in opposite directions.

Step 4, verdict, delivered to the investment committee. Fix-First, structured as a condition on the deal instead of a flat no. "Do not approve the $40M at the valuation implied by 118% NRR. That figure measures expansion among survivors only. The whole-cohort figure is 65%, which means each cohort loses roughly a third of its value within six months, whatever upsell story is layered on top of the survivors. Recommend one of: (a) a materially lower entry valuation reflecting the whole-cohort 65% figure, (b) a retention-linked tranche structure releasing the back half of capital only if whole-cohort revenue retention at month six clears 85% for two consecutive quarterly cohorts, or (c) walk if management will not recompute and own the real number themselves."

Read it. An equity analyst reading a 10-K footnote would have written the same finding. What made it advisory was pairing the hard number with three specific, fundable structures instead of a bare rejection. Instead of a reason to be nervous, the investment committee has a decision it can execute on Monday.

Worked example 3: Operating leverage as a scale-up risk, not just a margin fact (India, ₹)

FlexFit Studios, a boutique fitness-studio chain, runs six locations in India. It wants advisory sign-off for a private-equity-backed expansion to 25 studios over three years. You're asked whether the operating model can absorb that pace.

Solution. Step 1, the unit's operating leverage. Take one studio. It has capacity for 800 seat-classes a week, at ₹500 a seat with a variable cost of ₹80 a seat, so each seat contributes ₹420. Fixed cost is ₹2,52,000/week, which puts breakeven at 600 seats/week, or 75% utilisation.

FlexFit's two mature studios run at 85% utilisation, 680 seats. P&L = 680×420 − 252,000 = +₹33,600/week, a 9.9% margin, and that's healthy. Its four young studios are still ramping at 55% utilisation, 440 seats. P&L = 440×420 − 252,000 = −₹67,200/week (≈ −₹2.91 lakh/month). That's a real cash drag, and it keeps going every week.

Step 2, the reconciliation test on the portfolio, not the mature-studio slide. Blend all six. Total seats = 2×680 + 4×440 = 3,120/week, a 65% portfolio utilisation. Portfolio contribution = 3,120×420 = ₹13,10,400/week and portfolio fixed = 6×252,000 = ₹15,12,000/week, so portfolio P&L = −₹2,01,600/week, a −12.9% margin today.

Two profitable mature studios don't yet make up for four young ones.

Management's "our unit economics are proven" pitch is true of the mature-studio slide and nothing else. Run the reconciliation test on the whole chain and you see the full picture, as the reconciliation work in unit economics predicts you will.

Step 3, route the diagnosis and stress the plan. The unit works, since mature studios clear a real if modest margin. So this is a Machine problem. Fixed costs per studio are large relative to contribution near breakeven, which makes it a high-DOL structure. And every new studio spends roughly a year climbing the same ramp the four young studios are still on.

Suppose you fund 19 new studios all at once, the aggressive rollout. If their early ramp looks like today's young-studio loss rate, peak combined monthly cash drag is roughly 19 × ₹2.91 lakh + the existing base's ₹8.7 lakh/month ≈ ₹64 lakh/month. That probably understates the peak, because a brand-new studio opens well below 55% utilisation.

Now phase the rollout at 6 studios/year instead. At any time roughly 8 studios are mid-ramp (the current year's 6 plus a tail from the prior batch), so peak exposure is roughly 8 × ₹2.91 lakh + ₹8.7 lakh ≈ ₹32 lakh/month. You've cut the aggressive scenario's cash risk roughly in half. Phasing also buys 12 to 14 months of real ramp data on each cohort before the next tranche is committed.

Step 4: verdict. Scale, but only phased, with a named financing structure. "The unit works. Mature studios clear ~10% margins on a well-understood 75% breakeven, so this is Scale, not Fix-First. But fund it as 6 studios/year, not 19 at once. The aggressive pace roughly doubles peak monthly cash exposure for the same eventual outcome, and phasing buys real ramp data before each new tranche is committed. Release each year's tranche only after the prior year's batch crosses 70% utilisation on schedule."

Read it. Operating leverage here was never just a margin statistic. It's the mechanism that turns "how fast do we grow" into "how much cash risk do we carry at once." So the recommendation it produced was a financing structure instead of a margin forecast.

Worked example 4: The full teardown: applying the Scalability Verdict Rubric end to end (US, $)

Larkspur Metrics is a B2B marketing-analytics SaaS company that sells to mid-market retailers. It's raising a $25M Series C from a growth-equity firm, and you're building the investment committee's combined "quality and scalability" verdict.

Solution. Step 1: unit economics (recomputed). ARPU of $1,800/month at a 78% contribution margin gives contribution of $1,404/month. Fully loaded, sales-led CAC is $10,500, so payback = 10,500/1,404 ≈ 7.5 months, comfortably inside the healthy band for SaaS. Annual contribution is $16,848. With mature logo churn at 14%/year (r = 0.86) and a discount rate of 12%/year, LTV = 16,848 × 0.86 / (1.12 − 0.86) ≈ $55,729. That makes LTV/CAC ≈ 5.3×, which clears 3× comfortably.

Step 2, cohort trajectory (recomputed from the data room's own triangle, with the cohort method you already know). Whole-cohort revenue retention recomputes to 108%. Logo churn is offset by real expansion across the whole original base. This time it isn't a survivor-basis artefact. You checked for Loomwell's trap and it came up clear.

Step 3, operating leverage (computed fresh, company level). Revenue is $30M and variable costs are 20% ($6M), so contribution is $24M (80%). Take off fixed costs of $19M and EBIT is $5M, a 16.7% margin. DOL = 24/5 = 4.8. Breakeven revenue = 19/0.80 = $23.75M, and margin of safety = (30 − 23.75)/30 = 20.8%.

Now stress a −15% revenue shock. Revenue drops to $25.5M, variable costs to $5.1M and contribution to $20.4M, and EBIT falls to $1.4M. You've lost 72% of EBIT on a 15% revenue miss, which is DOL × shock. It confirms a real downside risk, survivable and short of catastrophic. Run the +15% upside case and you get the same 72% swing the other way, with EBIT rising to $8.6M. It's the same coiled spring in both directions.

Step 4, reinvestment runway and moat, taken as given (inputs from the market and moat work, which aren't re-derived here). Assume the full diligence's market-model workstream has already sized a $2B+ addressable market with under 8% penetration, which is plenty of runway. Elsewhere in the diligence, the moat assessment rates the company's embedded-workflow switching costs as Moderate.

Step 5, score the rubric and render the verdict. Unit economics are Green, with 5.3× LTV/CAC and a 7.5-month payback. Cohort trajectory is Green too, at 108% whole-cohort retention. Operating leverage is Yellow. A DOL of 4.8 is real risk in a downside, though the business stays EBIT-positive through the stress. Runway is Green and moat is Yellow, both as given.

Verdict: Scale, with two named conditions. First, reserve or tranche-release a portion of the round, contingent on the operating margin of safety not falling below a named floor over the next four quarters. That protects against the DOL risk you just quantified. Second, add a covenant requiring quarterly disclosure of whole-cohort (never blended) NRR. It protects against the survivor-basis risk the Loomwell example showed in a sibling company.

Read it. Two Yellows didn't cancel each other into a No-Go. They didn't get waved away either. Each one earned a specific, named condition attached to the money itself. That's what "the rubric structures the judgment, it does not replace it" means in practice.

You end on Scale, but it's Scale with teeth, and nobody's writing a blank check.

Worked example 5: Prioritizing the levers under a fixed budget (India, ₹)

Tapri Express, an Indian QSR chain, has a ₹3 crore improvement budget. It wants to know which of three levers to prioritise this year: (A) improve marketing efficiency to drive more transactions, (B) improve repeat rate via a loyalty initiative, or (C) raise prices 4%.

Solution. Step 1, the baseline. Tapri does 2.4 million transactions/year at a ₹180 average ticket and a 42% contribution margin. Each transaction contributes ₹75.60, and baseline annual contribution = ₹18.14 crore.

Step 2, score each lever: impact, cost, time-to-effect. Option A is better targeting: 8% more transactions, ₹80 lakh of cost and a 3-month rollout. Full-year incremental contribution = 192,000 × 75.60 = ₹1.45 crore. In year one it's realised for roughly 75% of a full year (the remaining nine months), ≈ ₹1.09 crore, so net year-one ≈ +₹0.29 crore.

Option B is a loyalty programme that lifts repeat share: 11% more transactions, ₹1.2 crore of cost and a 9-month ramp. Full-year incremental contribution ≈ ₹2.0 crore, the largest steady-state number of the three. But the slow ramp means only ~30% of the full-year figure (₹0.60 crore) is realised in year one. Net year-one ≈ −₹0.60 crore. So B loses money in year one, even though its eventual annual run-rate (₹2.0 crore/year) beats both other options.

Option C is a 4% price rise, with an elasticity assumption of −0.3, ~1 month to take effect and ₹5 lakh of cost. Volume falls only 1.2%, and contribution per transaction rises by the full ₹7.20 price increase. Full-year incremental contribution ≈ ₹1.49 crore, realised for 11/12 of the year (₹1.37 crore). That leaves net year-one ≈ +₹1.32 crore, the highest near-term net of the three, at the lowest cost.

Step 3, stress the riskiest assumption before recommending. Option C's recommendation rides on the elasticity assumption, so re-run it at a far more pessimistic elasticity of −0.8. Volume now falls 3.2% instead of 1.2%. Incremental contribution is still ≈ ₹1.09 crore versus baseline. It's smaller, but still solidly positive, so the recommendation survives its own bear case.

Step 4, verdict: a sequence, not a ranking. Prioritise Option C first. It has the highest near-term net impact, the lowest cost and the fastest time-to-effect, and it holds up even under a materially worse elasticity assumption. Execute it this month. Run Option A in parallel. It's positive in year one at moderate cost, and it doesn't compete with C for operational bandwidth. Defer Option B to next year and fund it from the cash the first two levers generate.

B isn't a weak idea. It has the largest eventual annual impact of the three. But its slow ramp makes it cash-negative in year one, and funding the slowest, most bandwidth-hungry lever first would have starved the two that pay for themselves straight away.

Read it. The naive "fund the biggest number" instinct would have gone straight to Option B and produced a cash-negative year one for the largest-looking win. The advisory move was to sequence under this year's budget and bandwidth, the constraint the client actually had, instead of an unlimited multi-year model.

That sequence is the recommendation.

Worked example 6: Catching the games catalogue live, and delivering it diplomatically (US, $)

Halcyon Box, a US subscription box startup, is fundraising. The founder's deck says "LTV $410, CAC $85, LTV/CAC = 4.8×, we're ready to triple ad spend." You're advising the seed investor on whether to back the raise.

Solution. Step 1, recompute from the raw data room, not the deck. ARPU is $22/month at a 55% contribution margin, so contribution is $12.10/month. Read mature-cohort churn from the actual retention triangle, ignoring the deck's "active user" definition, and it's 9%/month. That implies a true lifetime of about 11 months, far short of the ~30 months the deck assumes. Naive LTV = 12.10/0.09 = $134.44. Discounted properly (r = 0.91, d = 1%/month), LTV = 12.10 × 0.91 / (1.01 − 0.91) ≈ $110.11.

Halcyon's blended CAC of $85 folds in organic customers. What you need is the fully loaded, marginal paid CAC, because that's the number that prices the next tranche of ad spend the founder wants to triple. It's $175.

Step 2, the honest ratio, and what it means. LTV/CAC = 110.11 / 175 ≈ 0.63×. That's a long way from the deck's 4.8×. At today's churn and marginal CAC, every extra paid customer destroys value. Triple paid spend at these unit economics and what you triple is the rate of value destruction. Payback = 175/12.10 ≈ 14.5 months, well outside the healthy 6 to 12 month band for consumer subscription.

Step 3, name what would have to change, quantified, not vague. Say mature churn fell from 9% to 5%/month. Discounted LTV would rise to 12.10 × 0.95 / (1.01 − 0.95) ≈ $191.58, which puts LTV/CAC ≈ 1.09×, still short of the 3× bar. Churn alone won't fix this. Marginal CAC almost certainly has to fall as well, towards roughly $110 or below, before the ratio clears a fundable bar.

Step 4, the verdict, delivered constructively. No-Go on tripling ad spend as proposed. Fix-First on the underlying model. "Using marginal paid CAC and mature-cohort churn in place of the blended numbers in the deck, the recomputed figures put LTV/CAC at roughly 0.6×. The deck shows 4.8×. This is a common conflation (blended vs. marginal CAC, and an optimistic assumed lifetime vs. the observed one), and it isn't a sign of bad faith. Recommend a two-quarter proof-of-concept phase, funded at current spend levels and not tripled, with two named, quantified targets. Mature monthly churn should be at or below 5%, and marginal paid CAC at or below roughly $110. Both would need to move together before the ratio clears a fundable 3×. Revisit the raise once both are in hand or trending clearly towards them."

Read it. An equity analyst reading the same data room would write the same technical finding, an LTV/CAC of 0.6× against a claimed 4.8×. What made it advisory was the framing. You presented the gap as a common, fixable definitional conflation, never as an accusation. And you turned a flat "no" into two specific, quantified, revisitable targets the founder could go and build towards.

Practice set

Work each problem fully before you read its solution. Write before you peek. Numeric answers are to about ±2%. All companies are synthetic and internally consistent unless a primary source is named, so recompute every figure yourself before trusting a "Solution."

P1 (guided). Route the diagnosis. A packaged-foods company's contribution per unit is healthy and stable, and its cohorts of repeat buyers aren't deteriorating. But each new geography it enters grows more slowly than the last one, and profit pools in its category are visibly shifting towards private-label competitors. Is this a Unit, Machine, or Market problem? Which module owns the next step?

Solution. The unit (contribution) and the near-term machine (cohort stability) both check out, so the routing tree's first two questions both answer no. The slowing geographic returns and the shifting profit pool are symptoms of the Market layer, meaning category structure and competitive dynamics. This company's own unit economics aren't where the trouble is. The next step belongs to Market Attractiveness & Competitive Dynamics, so hand off. Don't recommend a pricing or retention fix, because no unit-level lever repairs a market that's structurally reallocating its profit pool to a different kind of competitor.

P2 (guided). Build a fresh contribution ladder and route it. Basantpur Bites, an Indian cloud kitchen, reports these figures per delivered order. Net revenue is ₹350 and product COGS ₹210. Fulfilment (shipping, pick-pack, gateway, support) costs ₹70. Marketing blends a 40% new-order share at ₹100/order with a 60% repeat share at ₹20/order. Compute CM1, CM2, CM3 (blended) and the breakeven repeat share, then say whether this passes the Unit-layer test.

Solution. CM1 = 350 − 210 = ₹140 (40%). CM2 = 140 − 70 = ₹70 (20%). Blended marketing = 0.40×100 + 0.60×20 = ₹52, so CM3 = 70 − 52 = ₹18 (5.1%), which is positive. First-order CM3 = 70 − 100 = −₹30 and repeat-order CM3 = 70 − 20 = +₹50, so the breakeven repeat share comes from f·50 + (1−f)·(−30) = 0 → f = 37.5%. The current 60% repeat share is comfortably above the 37.5% breakeven, so this passes the Unit-layer test cleanly. Compare Worked Example 1's RozanaMart, with a 72.7% breakeven against a 70% actual share, which was a near-fail. If Basantpur Bites has a growth problem, it isn't in this ladder. Check the Machine or Market layer next.

P3 (guided). CAC three ways, and which one to use for which decision. At Ferngully Foods, paid ads cost ₹4.0 crore and bring in 80,000 customers. Referrals cost ₹0.3 crore and bring in 15,000. Organic brings in 45,000 customers. Compute paid CAC and blended CAC. Then say which one you'd use to answer "what will next year's growth cost us," if next year's plan is to double new customers to 280,000 by scaling paid spend. Organic caps near 50,000, referrals reach 25,000, and the marginal paid customer now costs ₹650.

Solution. Paid CAC = 4.0 cr / 80,000 = ₹500. Blended CAC = 4.3 cr / 140,000 = ₹307. For the growth decision, don't use either one on its own. Compute the marginal, blended-forward cost of the incremental plan. Paid must deliver 280,000 − 50,000 − 25,000 = 205,000 customers at ₹650 each (₹13.325 cr), plus referrals' ₹0.5 cr, spread over 280,000 total customers. That gives new blended CAC ≈ ₹493.75, a 61% rise, even though paid CAC itself only rose 30%, from ₹500 to ₹650. The mix shift towards paid amplifies the increase. Tell the board the true marginal cost of the incremental plan, because last year's blended average of ₹307 won't hold.

P4 (independent). Churn algebra into an honest verdict. Fernglen Supply Co. (US) has $60/month ARPU, a 65% contribution margin, mature churn of 6%/month, CAC of $210 and a discount rate of 1%/month. Compute contribution/month, naive LTV, discounted LTV, LTV/CAC and payback. Then find the churn rate that would clear a 3× LTV/CAC bar, holding CAC fixed.

Solution. Contribution = 60 × 0.65 = $39/month. Naive LTV = 39/0.06 = $650. Discounted (r = 0.94, d = 0.01), LTV = 39 × 0.94 / (1.01 − 0.94) ≈ $523.71. LTV/CAC = 523.71/210 ≈ 2.49×, just under 3×. Payback = 210/39 ≈ 5.4 months, which is excellent. Solve for churn c at LTV/CAC = 3 (target LTV = $630) and churn needs to fall from 6% to ≈ 4.9%/month, a modest and plausible ask. Verdict: Fix-First with a narrow, credible target. The cash maths already works, and a small retention improvement clears the lifetime-value bar too.

P5 (independent). Read a small triangle before committing to a churn estimate. Here's the triangle. Jan's cohort of 2,000 reads 100, 70, 58, 52 for M0 to M3. Feb's cohort of 2,400 reads 100, 68, 55 for M0 to M2. Mar's cohort of 2,900 reads 100, 65 for M0 to M1. Is there vintage deterioration? And what's the risk of estimating mature churn from this data today?

Solution. Read down the M1 column and you get 70, 68, 65, a mild, real softening of roughly one to two points per cohort as cohort size grows, which fits the marginal new customer being slightly weaker. You can't see a diagonal calendar-shock stripe in this small window. But Jan's own row, going from 100 to 70 to 58 to 52, hasn't flattened into a habit floor by month 3. Churn is still decelerating and hasn't settled: 70 to 58 is a 17.1% monthly churn, and 58 to 52 is 10.3%. So any "mature churn" figure you read off this data today would be extrapolated from a cohort that hasn't reached its flat part yet. Unit economics raised the same caveat about young-book churn estimates. The sound answer is to hold back a firm mature-churn number and ask for two more months of the Jan cohort before committing an LTV to it.

P6 (independent). Operating leverage stress test. DeskShare, an Indian coworking chain, has capacity of 300 seats per centre and charges ₹8,000/seat/month. Variable cost is ₹2,000/seat and fixed cost is ₹12,00,000/centre/month. Compute breakeven utilisation, current P&L at 58% utilisation, P&L and DOL at 75% utilisation, and the effect of a 10% demand shock at 75%.

Solution. Contribution/seat = ₹6,000, so breakeven seats = 12,00,000/6,000 = 200 seats, or 66.7% utilisation. At 58% (174 seats), P&L = 174×6,000 − 12,00,000 = −₹1,56,000/month. At 75% (225 seats), P&L = 225×6,000 − 12,00,000 = +₹1,50,000/month, and DOL = 13,50,000/1,50,000 = 9.0. That's very high, as you'd expect near breakeven. Now hit it with a −10% demand shock from 75%, down to 202.5 seats. P&L = 202,500... = 202.5×6,000 − 12,00,000 = ₹15,000/month, a 90% EBIT decline, and DOL × shock = 9.0 × 10% = 90% confirms it. A business running close to its breakeven line carries enormous downside sensitivity even while it's EBIT-positive. Flag this explicitly in any Scale verdict instead of reporting the 75%-utilisation profit figure on its own.

P7 (independent). Score the rubric as a whole. Millbrook Analytics is a US B2B SaaS company. LTV/CAC is 2.1× and payback is 9 months. Whole-cohort NRR is 96%, flat and not improving. DOL is 6.5 with a margin of safety of only 9%. TAM is $600M, and 40% penetration is already achieved. Score the five rubric dimensions and render an overall verdict.

Solution. Unit economics are Yellow. Payback is good, but LTV/CAC at 2.1× is meaningfully under 3×. Cohort trajectory is Yellow too. NRR is flat, so it isn't declining, but it isn't improving either, and there's no positive trend to lean on. Operating leverage is Red, because a 9% margin of safety against a DOL of 6.5 means even a modest downside pushes towards a loss. Reinvestment runway is Red as well: 40% penetration of a modest $600M TAM leaves limited room to grow into. Moat isn't given. Treat it as unscored or unknown. It's a gap to flag, and you shouldn't assume Green.

That's two Reds, operating leverage and runway, alongside a below-bar unit-economics score. The overall verdict is No-Go on scaling as proposed. No single number is catastrophic. The combination leaves very little room for anything to go wrong, and the runway itself is nearly exhausted. Recommend revisiting only if a credible new addressable segment turns up, which is a runway question for the market-attractiveness work, or if the cost structure is de-risked first, which is a Machine-layer fix.

P8 (independent). Score two levers and choose a sequence. Option X brings $2.0M of incremental annual contribution for $300k of cost, with a 2-month time-to-effect. Option Y brings $3.5M of incremental annual contribution for $1.2M of cost, with a 10-month time-to-effect. Which goes first, and why?

Solution. Impact ÷ cost gives X = 2.0/0.3 ≈ 6.7× and Y = 3.5/1.2 ≈ 2.9×. X is far more capital-efficient, and it lands in a fifth of the time. Recommend X first and fund it immediately. It's cheap, it's fast, and it's plainly high-return. Look at Y again once X's cash gain is funding it or a dedicated budget exists. It's the same logic as the C, then A, then B sequence in Worked Example 5. Y's larger steady-state number doesn't automatically earn it first place.

P9 (independent). Catch the games-catalogue pattern live and state the advisory next step. A founder tells the board, "Our CAC is a remarkably low ₹307, driven by our incredible referral engine." The data room shows the same split as P3's Ferngully Foods. Paid ads cost ₹4.0 crore for 80,000 customers, referrals cost ₹0.3 crore for 15,000, and 45,000 came in organically. What's the specific issue, and what would you say to the board?

Solution. ₹307 is the blended CAC (₹4.3 cr / 140,000). It folds in 45,000 essentially free organic customers alongside the 80,000 the company actually paid ₹500 each to acquire. It's no evidence that the "referral engine" is driving the low number either, since referrals are only 15,000 of the 140,000. Frame it around the gap and leave the founder out of it: "The ₹307 figure blends a large organic base into the paid number. It's common and not misleading in itself, but it understates what the next tranche of paid growth will actually cost, since paid CAC alone is ₹500 and organic doesn't scale on command. Recommend the board ask for CAC broken out by channel before approving next year's paid budget on the strength of the blended figure."

P10 (independent). Run the reconciliation test adversarially and name the advisory ask. A retail chain's deck claims "60 stores, 14% store-level EBITDA margin, so consolidated EBITDA should be about ₹840 crore on ₹6,000 crore revenue." Reported consolidated EBITDA is ₹360 crore. Run the reconciliation test and state two likely explanations and what you'd request.

Solution. Implied EBITDA at the claimed margin = 6,000 × 14% = ₹840 crore. Reported is ₹360 crore. So the residual is ₹480 crore, or 133% of the reported figure, vastly above the 10 to 15% threshold at which a story is flagged as not reconciling. That's no rounding error. There are two likely explanations. (1) The 14% figure is "mature stores only," and it leaves a large tranche of younger, loss-making stores out of the average, which is blended masking. (2) "Store-level" excludes real store costs (pre-opening expenses, warehousing, last-mile) that belong in the unit P&L but have been allocated to "corporate" instead. That one's an allocation game. For the advisory ask, request the store-vintage split, the full definition of "store-level EBITDA," and a written bridge from the store-level figure to the consolidated number. A company confident in its store economics can produce all three without difficulty.

P11 (independent). Payback vs. LTV/CAC divergence as a financing-structure question. Two customer segments at the same US company both show LTV/CAC of 3.0×. Segment A's CAC is repaid in 6 months and Segment B's in 24 months. Which would you tell the client to fund with a fast, debt-like growth facility, and which needs patient equity capital? Why?

Solution. Identical LTV/CAC means identical eventual wealth creation per customer, but the cash timing and risk are very different. Segment A's 6-month payback means the acquisition loan is repaid quickly and predictably. That makes it a good candidate for a debt-like or revenue-based financing facility, since the lender's risk window is short. Segment B's 24-month payback means the company carries four times the "float" of not-yet-repaid CAC per customer for the same eventual return. (That's the float identity from unit economics.) It's a bet on a retention curve that has to hold for two full years before it pays off.

So B is a risk for patient equity, and a poor fit for debt-like money. A downturn or a capital-markets freeze midway through the 24 months leaves the company holding unrecovered CAC with no fast way to unwind it. Advise financing each segment differently, even though their headline ratio is the same.

P12 (independent). Distinguish a Market hand-off from a Unit fix. A specialty retailer's unit economics and cohorts are both healthy. But every new store cannibalises an existing one's sales, same-store growth has flattened across the category, and two new low-cost entrants have started taking share nationally. Route the diagnosis and name the hand-off.

Solution. The Unit (contribution, CAC/LTV) and the Machine (cohort and store trajectory) both check out, so the disease isn't there. Cannibalisation, flattening category growth and new entrants taking share nationally are all Market-layer symptoms. The category itself is maturing, and its competitive structure is shifting. Route it to market attractiveness and competitive dynamics. Don't recommend a pricing or retention fix at the unit level, because no unit-economics lever repairs a maturing, more contested category.

P13 (independent). Compress a jumble of findings into an answer-first governing thought. You have an LTV/CAC of 1.2×, a CAC payback of 22 months, whole-cohort NRR falling three quarters running, a DOL of 6.0 and a very large total addressable market. Write a two-sentence answer-first governing thought for a steering-committee slide. Don't build the deck, just the sentence. The craft itself comes from your synthesis and communication work, so it isn't re-derived here.

Solution. "Recommend No-Go on the scaling ask as proposed: at a 1.2× LTV/CAC, a 22-month payback, and a whole-cohort retention curve that has worsened for three straight quarters, the unit itself is losing money in a way that scale would only accelerate, the large addressable market is real but irrelevant until the unit is fixed. Revisit only once whole-cohort NRR has stabilised for two consecutive quarters and honest LTV/CAC clears 3×." Notice the shape. The verdict comes first. Next come the two or three reasons, compressed into a single sentence, then the large-but-irrelevant TAM, named and dismissed, and last a concrete re-entry condition. Your synthesis work teaches that answer-first shape as a full craft.

P14 (timed, ~10 minutes). Mini-teardown 1 (US, $). Northfield Shave Co. is a DTC subscription razor brand with $18/month ARPU and a 58% contribution margin. CAC is $95 (paid, fully loaded), mature churn is 4%/month and the discount rate is 1%/month. Compute contribution, discounted LTV, LTV/CAC and payback, then render a verdict.

Solution. Contribution = 18 × 0.58 = $10.44/month. Naive LTV = 10.44/0.04 = $261. Discounted (r = 0.96, d = 0.01), LTV = 10.44 × 0.96 / (1.01 − 0.96) ≈ $200.45. LTV/CAC ≈ 2.11×, below 3× but not by a wide margin. Payback = 95/10.44 ≈ 9.1 months, comfortably inside the healthy band. Verdict: Fix-First, narrow gap. The cash maths is already sound. Holding CAC fixed, churn would need to fall from roughly 4% to about 2.6%/month to clear 3×. It's a real but achievable ask, and it doesn't call for a fundamental redesign.

Set it beside Worked Example 6's Halcyon Box. The shape of the problem is the same, with LTV/CAC under the bar and payback fine, but the size of the gap is vastly different. The verdict format is identical and the required fix isn't. Naming the size of the gap is itself part of the advisory judgment.

P15 (timed, ~10 minutes). Mini-teardown 2 (India, ₹). PixelPlex is an Indian gaming-lounge chain. Each lounge has capacity of 12,000 seat-hours/week and sells them at ₹150/seat-hour against a variable cost of ₹50/seat-hour. Fixed cost is ₹9,00,000/week per lounge. Compute breakeven, current P&L at 68% utilisation, the value of one utilisation point, and P&L/margin at 80% utilisation.

Solution. Contribution/seat-hour = ₹100, so breakeven = 9,00,000/100 = 9,000 seat-hours, or 75% utilisation. At 68% (8,160 seat-hours), P&L = 8,160×100 − 9,00,000 = −₹84,000/week. One utilisation point is worth 1% × 12,000 × ₹100 = ₹12,000/week, and since fixed costs don't move, it's almost pure profit. At 80% (9,600 seat-hours), P&L = 9,600×100 − 9,00,000 = +₹60,000/week, and margin = 60,000/(9,600×150) ≈ 4.2%. PixelPlex needs roughly 7 more utilisation points to flip from a weekly loss to a modest profit. That's a specific, trackable target for the next quarter's operating review, and far more useful than a vague "improve utilisation" instruction.

P16 (synthesis). State your own procedure. You've worked through six worked examples and fifteen practice problems. In four to five sentences, state the general decision procedure you now follow whenever you're handed a business-model teardown, from a data room's raw numbers to a client-ready recommendation. Then name the one thing that would make you escalate a verdict from Fix-First to No-Go on any deal.

Solution. A representative procedure: "Recompute every headline metric from raw data before reading the deck's own summary of it, because the games catalogue applies just as much live as in a filing. Route the diagnosis through Unit, then Machine, then Market, in that order, and stop at the first layer that is actually broken instead of defaulting to whichever layer the client's own narrative points at. Score the five-dimension rubric in full before rendering one of three verdicts, and if the verdict is Scale or Fix-First, name the specific evidence and threshold that would flip it. If more than one lever is available, sequence by impact-per-cost and time-to-effect under the client's real constraint, and never by which lever has the biggest steady-state number. Deliver the whole thing answer-first, with the hard finding framed as a fact pattern and a next step, never an accusation." The one thing that escalates Fix-First to No-Go is a cohort trajectory that's actively deteriorating (not merely flat) with no credible, evidenced lever to reverse it within the client's own timeline. A business whose unit economics are merely thin can often be fixed. One whose unit economics get worse every quarter is telling you that the fix, whatever it is, is losing the race against decay right now.

Applied mini-project

Produce a business-model teardown and advisory memo. The project gives you a portfolio piece, and it works as a realistic simulation of a single case-team meeting. Take one data-room-style dossier and turn it into a one-page advisory memo. It needs recomputed unit economics, a routed diagnosis, a scored Scalability Verdict Rubric, a sequenced lever recommendation if one is relevant, and an answer-first verdict with a named condition for changing your mind.

The dossier, BrightPath Learning (India, ₹), a live online-tutoring subscription. ARPU is ₹1,500/month at a 48% contribution margin. Fully loaded paid CAC is ₹3,200, and the discount rate is 1.2%/month. A six-month-old cohort keeps 61% of its original logos, and average revenue per surviving student has risen 9% through add-on courses. At company level, revenue is ₹42 crore, variable costs are 45% of revenue and fixed costs are ₹19 crore. You're given two more facts and don't have to derive them. India's online-tutoring addressable market is roughly ₹18,000 crore, so the company's ₹42 crore of revenue is well under 1% penetration. And BrightPath holds some brand and curriculum IP but faces low switching costs.

Do this, in order:

1. Recompute contribution per student per month, discounted LTV, LTV/CAC and CAC payback. Separately, compute the whole-cohort revenue retention at six months. Watch for the survivor-basis trap you've already drilled twice (Loomwell and P9).
2. Compute company-level operating leverage: contribution, EBIT, DOL, breakeven revenue and margin of safety at the current revenue base.
3. Route the diagnosis. Is this a Unit, Machine, or Market question? Or does it pass all three checks cleanly? If it does, say so.
4. Score all five rubric dimensions Green, Yellow or Red. Use the given runway and moat inputs where you have no basis to compute them yourself.
5. Render the verdict (Scale / Fix-First / No-Go) with the specific evidence that would flip it.
6. Write the one-page memo, answer-first. Put the verdict first, then two or three reasons beneath it, one named risk and its mitigation, and one concrete next step. It's the answer-first shape from your synthesis work, applied to your own recompute.

Deliverable and self-check. One page, plus your recompute working. Score yourself against the rubric below. Pass at 8/10 with no zero on any row:

Keep the memo. The recompute-route-score-recommend sequence you just ran is what a case team lead in commercial due diligence will expect inside a single meeting, and every later module on business evaluation builds on that shape.

Case Lab: the mapped case for this teardown is the CN5 branch's forthcoming Business-Model Teardown case, a commercial-due-diligence-style, candidate-led teardown of a growth-stage target. Once it's authored, it becomes the natural business-model-teardown entry in the Consulting Case-Engine's case bank, alongside CC-CDD-01 in the due-diligence toolkit. It follows the same interim pattern already named for the deeper CN5/CN6 labs it defers to. Until that case ships, your case-equivalent is the Applied Mini-Project above, self-graded against its rubric at the same ≥85% bar the quiz below carries.

Reading & resources

Treat each item below as an anchor to learn from and check against, never as a primary data source to quote. Everything is free unless marked, and the level tags follow the ramp from intuition to mastery.

The judgment spine (read these first):

• Richard Rumelt, Good Strategy / Bad Strategy. It's the diagnose-before-prescribe kernel the routing tree is built on, so read the diagnosis chapters before anything else here. [Paid] [Intermediate–Advanced]
• Charles Conn & Robert McLean, Bulletproof Problem Solving, on sizing a decision before you recommend one and naming what would have to be true. [Paid] [Beginner–Advanced]
• Maister, Green & Galford, The Trusted Advisor, for the trust equation and low self-orientation. It's the source for delivering a hard finding without it turning into an accusation, which the advisory-scepticism section above leans on. [Paid] [Beginner–Advanced]

Business quality and unit economics, applied at scale:

• Tim Koller, Marc Goedhart & David Wessels, Valuation (McKinsey). Read the chapters on the link between ROIC, growth and value. They explain why a broken unit destroys value faster once it's scaled. [Paid] [Advanced]
• Bill Gurley, "A Rake Too Far", the canonical essay on marketplace take-rate economics and the ceiling on a platform's own pricing power. Read it alongside any marketplace-shaped teardown. [Free] [Intermediate]
• Aswath Damodaran's data library has free benchmarks for industry margins, growth and multiples. Use it to sanity-check a rubric's runway dimension before you hand the sizing proper to CN5.02. [Free] [Beginner–Advanced]

Free official practice, for reps at case pace:

• McKinsey / BCG / Bain official practice cases (on their career sites) and the free MBA consulting-club casebooks (Wharton, Kellogg, Fuqua/Duke, Darden, INSEAD, LBS, Ross, Booth, Columbia). Several contain a profitability or growth case that's a unit-economics teardown underneath. Practise routing and rendering a verdict against a clock. [Free] [Beginner–Advanced]
• Victor Cheng, Case Interview Secrets and caseinterview.com, for the timed delivery that the timed practice items (P14 and P15) rehearse. [Mixed free/paid] [Beginner–Intermediate]

Cross-links inside the program (do not re-read as new material: connect):

• M4.01 (Business Model Anatomy) and M4.02 (Unit Economics) hold every formula cited here, and neither gets re-derived. Go there for the mechanics. [Program] [Intermediate–Advanced]
• CN5.02 (Market Attractiveness & Competitive Dynamics) is where a Market-layer routing hand-off lands. [Program: forthcoming] [Intermediate]
• CN5.03 (The CDD Toolkit + Voice-of-Customer) is the full commercial due diligence, and your recompute-and-verdict becomes one workstream inside it. [Program: forthcoming] [Intermediate–Advanced]
• M4.04 and M4.05 (Moats) answer the durability question that the teardown deliberately takes as a given input. [Program] [Intermediate–Advanced]
• CN1.03 (The "So What") and M10.03 (Communication Toolkit) own the answer-first compression and deck craft that your finished verdict is handed to. [Program] [Intermediate]
• HBR, for the practitioner literature on unit economics and growth-versus-profitability trade-offs. Search "unit economics," "growth at all costs," "cohort analysis". [Mixed free/paid] [All levels]

Don't just read, do this: find one real, public growth-stage company's investor deck or shareholder letter (India or US) that states an LTV/CAC, an NRR, or a "unit economics" claim. Run the recompute on it using only what the company itself discloses. If the disclosed metric looks survivor-based or blended, restate it on a whole-cohort or marginal-CAC basis. Then write the one-paragraph verdict you'd deliver if you were advising the board. Ten minutes of that will teach you more than an hour of reading.

Teach it back & journal

Feynman prompt. Explain to a smart friend why knowing the formulas isn't the same as being able to advise a client. Your friend has just read the business-model and unit-economics material and can already compute contribution, CAC, LTV, and DOL correctly. Use this sentence: "the client is not paying you to confirm the arithmetic; they are paying you to tell them what to do about it, and what would change your mind." Then run the routing tree with them. Take a business they know (their own employer, a favourite app, a local chain) and walk it through Unit, then Machine, then Market. Have them try to name which layer is actually broken, if any. Listen for whether your friend can, unprompted, tell "the unit doesn't work" apart from "the unit works but scaling it is risky" and from "this isn't really a business-model question at all, it's a market question". If they can, you've installed the routing reflex.

Journal prompt. Take one real, public, growth-stage company's own disclosed unit-economics or cohort claim, from India or the US. It could come from an investor letter, an S-1/DRHP, or an earnings call. Run the full recompute-route-score-recommend sequence on it, using only what the company itself discloses. Log three things. First, what you had to recompute instead of accepting. Second, which layer (Unit, Machine, or Market) the diagnosis actually routed to, and whether that surprised you. Third, what verdict you'd deliver, with the one number, named in advance, that would flip it.

Then write a paragraph on this idea. The advisory edge is the habit of never leaving a correct number without a decision attached, delivered to someone who's about to act on it. Access to the formulas is no edge at all, since the unit-economics module already gave those to everyone who completed it. Which single habit from the teardown would most change how a case-team lead reads your first memo? It might be routing before recommending, scoring the rubric before naming a verdict, or naming the falsifier before finishing the sentence. And what will you do this week to make it automatic?

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