Case Math Under Pressure

Branch: Casing & Quantitative Fluency (Consulting galaxy) · Node CN2.01 · ~8 focused hours · Mastery gate ≥ 85%

Picture the moment. You are two minutes into a live case (a client meeting, a partner's grilling, an interview) and you have just said "let me size that." The room goes quiet. There is no calculator on the table, no spreadsheet, no chat window you can quietly open. There is a pen, a blank page, and every eye waiting to see whether the structure you built a moment ago can actually produce a number. This is the instant that separates a candidate who sounds sharp from one who is sharp, and it has nothing to do with how elegant your issue tree was. It is arithmetic, live, unaided, narrated out loud, and it is a completely learnable skill that most people never deliberately practice.

Case math looks like harder classroom math, but it's scored on a different board. Classroom math rewards exactness, however long it takes, worked in silence and checked by a calculator. Case math rewards speed and a bounded, known error, roughly ±2%, sometimes a looser ±5% for a rough cut. It also rewards the part almost nobody drills, which is narration. You say your numbers, your rounding and your assumptions out loud as you produce them, so the room can follow your thinking instead of waiting in silence for a verdict.

Five families of computation do most of the work: percentages, growth rates, weighted averages, breakeven and contribution math, and ratio consistency checks. A consultant who can run those fast and out loud, rounding with care, has taken arithmetic off the list of things that limit their thinking. Everything else in a case (the structure, the hypothesis, the "so what") then gets 100% of your attention, because the numbers come out right, quickly, without a second thought.

Give it the hours ahead and five kinds of live computation get close to automatic. You'll know the correction that turns a naive "just add the percentages" guess into the correct result, and how big an error the guess makes when you skip it. You'll know why "up 20%, down 20%" never gets you back to where you started, and by how much it misses. You'll estimate a compound annual growth rate to within a point or two without a calculator, using nothing but the Rule of 72 and one direction fact that always holds. You'll build a blended average that's actually weighted. An eyeball average of two ratios is the most common silent error in case rooms and in real analyst work alike. You'll compute a breakeven point and a margin of safety, and see how sensitive that breakeven is to a small cost change. And you'll use a ratio to catch, on sight, when two numbers a source has handed you can't both be true.

The scope promise, restated. The galaxy's promise about scope applies here quite literally. Fluent mental math won't make you a partner. Knowing which number is worth computing carefully, which is fine to eyeball, and which one the whole recommendation hinges on comes from hundreds of live cases, and no course can install it. What you can fully learn is the arithmetic engine itself: the five operations, done fast, done right and narrated. That's the floor a working consultant needs on day one, and these pages take you to it. Most people never reach that floor without deliberate practice, and the next eight hours are that practice.

One more note before the content starts. The ground covered here is deliberately narrow. Market sizing and Fermi estimation from scratch get their own node next, which builds the assumption tree, chooses top-down or bottom-up and defends a range. Unit-economics modeling belongs to the Finance galaxy, where contribution margin, CAC, LTV and cohorts become a business-model diagnostic. DuPont and ROE are derived there too, in the ratio toolkit. You learn the raw arithmetic all three of those lean on, sped up, stripped of ceremony and tested for a room with no calculator in it.

Learning objectives

By the end you can:

1. Convert cleanly between "percentage of," percentage-point change, and percentage (relative) change, and never again confuse a point move with a relative move, out loud or on paper.
2. Combine successive percentage changes with the exact correction-term trick (not naive addition), including the equal-and-opposite trap (why "up 20%, down 20%" is a net loss) and the reciprocal recovery trap (why undoing a decline always takes a bigger percentage gain than the decline itself), and quantify precisely how large an error the naive shortcut makes.
3. Estimate a compound annual growth rate (CAGR) fast, using the Rule of 72 (and its 70 / 69.3 cousins) plus the one always-true direction fact about naive linear guesses, landing within a defensible point or two without a calculator.
4. Build a correctly weighted blended average from two or more segments, explain why an unweighted average of ratios is a live error, and predict, and quantify, the pure mix-shift effect when segment weights move but underlying rates do not.
5. Compute breakeven volume, breakeven revenue, and margin of safety from fixed cost, price, and variable cost; and predict (with a number, not just a direction) how a small price or cost change moves the breakeven point, and why that move is often disproportionately large.
6. Use a ratio as a live consistency check: reverse-engineer one disclosed figure from another to test whether a third claimed figure could possibly be true, and catch an order-of-magnitude mismatch on sight rather than after the meeting.
7. Apply a small, memorized toolkit of friendly-number decomposition (round-and-correct multiplication, the ×5/×15/×25/÷5/÷25 factor tricks, and the core fraction-to-percent table) to multiply, divide, and estimate fast, and know, case by case, how small the resulting error actually is.
8. Carry a rounding discipline that keeps cumulative error inside roughly 2% across a multi-step chain, state your roundings and assumptions out loud as you compute, and never let doing the math in your head quietly become doing the math in silence.
9. (AI-augmented objective.) Use an AI tool to generate a bottomless, leveled stream of case-math drills and to pressure-test your worked method between practice sessions, while keeping every ounce of the live arithmetic in your own head, because no case room, client meeting, or steering committee lets you open a chat window mid-sentence.

Objective 9 is a standing habit, not a separate gate: the mastery check and the mapped drill test objectives 1–8 only, and nobody grades how you practiced. But the discipline inside objective 9 (draft with AI, verify by hand, never depend on it live) is exactly what keeps the other eight fast and trustworthy over time.

Prerequisites & connections

Builds on. Nothing, formally. CN2.01 is the first node of the Casing & Quantitative Fluency branch, and under this galaxy's resolved progression rule every branch's opening module carries no hard prerequisite from anywhere else in the galaxy. You can start here cold with ordinary school arithmetic: fractions, decimals, percentages. Two things help it click faster, though neither is required. First, CN0.01's placement diagnostic already samples six items from this territory, so a strong score there is a signal, though not a requirement, that you'll move quickly here. Second, CN1.01 and CN1.02 (issue trees and hypothesis-driven problem solving) build the structuring habit this arithmetic will eventually work inside. If you've already met them, you'll see straight away where each of the five operations gets used in a real tree.

Leans on the Finance galaxy, never re-teaches it. If you've been through the Finance galaxy, you've already done slow, careful, on-paper versions of every operation you'll speed up here. M2.01 (Ratio Toolkit I. Profitability & DuPont) built the margin stack and the ROE/DuPont algebra, and a blended-margin question (objective 4) is a fast-forward version of it. M2.02 and M2.03 (Liquidity/Solvency/Efficiency; Returns on Capital) built the ratio and turnover fluency that objective 6's consistency checks lean on. M4.02 (Unit Economics) built contribution margin and breakeven from the ground up as a business-model diagnostic. Objective 5 is the same arithmetic with the diagnostic apparatus stripped off, sped up for live use. M3.01 (Time Value of Money) built compounding and discounting with full rigor, and objective 3's CAGR work is that same compounding, approximated in your head against a clock. None of that judgment gets rebuilt here. You take the raw arithmetic those modules lean on and drill it for speed, for reliability without a calculator and for out-loud narration, because a client meeting won't pause while you reconstruct a DuPont tree from first principles. If you've never touched the Finance galaxy, you don't need any of it. Each technique gets taught from scratch, with its own worked examples.

Feeds forward. CN2.02 (Market Sizing & Guesstimates) assumes this arithmetic is already automatic, so your attention can go to building the estimation tree and defending its assumptions, and the multiplication buried inside it takes care of itself. CN2.03 and CN2.04 (the core case formats: profitability, market entry, pricing, growth, M&A, operations) are the rooms where this math gets used. Every profit-tree isolation, every pricing bridge, every synergy stress-test is one of these five operations, chained. CN1.02's "one-day answer" only works if the arithmetic behind it doesn't cost you the day. And CN7.03, the timed Partner Capstone, runs an entire engagement candidate-led against a clock. By then none of this can still be a bottleneck.

In one sentence. Learn five families of arithmetic (percentages and their compounding, growth rates, weighted averages, breakeven and contribution, and ratio consistency checks) well enough that they run at conversation speed, with rounding habits that keep your error bounded and a narration habit that keeps the room with you.

Core teaching content

Why case math is a different skill from classroom math

Case math differs from the arithmetic you already know in three ways, and naming them now sets the standard for everything below.

First, there's no calculator. In real client work there rarely is one either. A partner who has to open a spreadsheet to sanity-check a number in the room has lost a beat of credibility.

Second, the acceptable error band is wide and known in advance. Roughly ±2% is the working convention used throughout, loosening to ±5% for a rough directional estimate. It tightens only when a recommendation rests on a knife-edge threshold, such as a breakeven test or a go/no-go cutoff. Chasing a fourth decimal place when the inputs are only known to ±10% may look like rigor, but it wastes the room's patience, and experienced people read it as a sign you don't know which digits matter.

Third, and almost nobody drills this on their own, the arithmetic is narrated out loud. Saying "call it fifty, so that's twenty-one hundred, plus a bit for the eighteen percent, round three-two-six" as you go is a different skill from getting the same answer in your head and reporting it a minute later. Silence reads as stuck even when you aren't. Narration turns a calculation into reasoning the room can see and trust.

Five families of computation cover the overwhelming majority of what shows up live: percentages and how they compound, growth rates, weighted (blended) averages, breakeven and contribution math, and ratios used as consistency checks. Under all five run two habits. One is a small toolkit of fast-multiplication and fraction shortcuts. The other is a way of rounding that keeps error bounded across a multi-step chain. Get the five families and the two habits down, and there's very little live arithmetic left that can catch you off guard.

One fence is worth putting up, because the line is easy to blur. None of this is market sizing. The market-sizing node teaches you to build the assumption tree: choosing top-down or bottom-up, defending each input's range, landing on a defensible band. What you learn here happens inside that tree once the assumptions are chosen. It's the multiplication, the percentage and the ratio, done fast and right. The same fence runs around unit economics, where the Finance galaxy builds the full CAC/LTV/cohort diagnostic, and around DuPont, which it builds as a full decomposition and judgment tool. Think of this as the engine, and of those as what you build with it.

Percentages: three words that are not the same word

Treating "percentage," "percentage point" and "percentage change" as interchangeable is the most common live error in a case room. They mean different things, and the confusion is expensive because nobody sees it until someone catches it. Suppose a brand's category share moves from 20% to 25%. That's a percentage-point (pp) change of +5. You subtracted one percentage from another, and the answer is in "points," not "percent." It's also a relative percentage change of +25%, because you divided the point change by the starting value: 5 ÷ 20 = 0.25. Both statements are true and describe the same event, yet the numbers are far apart, 5 against 25. A room that hears "share is up 5%" when the speaker meant "up 5 points" will draw a conclusion roughly five times too small.

So say which one you mean, every time. "That's five points, or a 25% relative jump, since we started from a smaller base" takes one extra breath and removes a whole category of miscommunication.

Reverse percentages need the same care. Here you're recovering an original figure from one that already has a percentage baked in. Instinct says subtract the rate. The correct move is to divide by (1 + rate). Say a shelf price of ₹590 already includes 18% GST. The pre-tax price is 590 ÷ 1.18 = ₹500, and you can check it: 500 × 1.18 = 590. The tempting shortcut of subtracting 18% from ₹590 (590 × 0.82 = ₹483.80) is wrong by ₹16.20, a 3.2% error. The 18% was always a percentage of the smaller, original number, never of the larger, final one, so taking 18% off the tax-inclusive price removes too much. That's a different operation from removing an 18% markup from the original price. Whenever you back an "of what" number out of a percentage that's already been applied, divide by (1 + rate). Don't subtract the rate from the final figure. It's an easy slip when the price tag is the only number in front of you, which is when it matters most to catch it.

Breakeven and contribution: where the fixed cost gets paid off

Breakeven analysis answers one question. How much volume, or revenue, do you need before fixed costs are fully covered and the business starts contributing to profit? The engine is the contribution margin (CM), price minus variable cost per unit. Contribution margin is the amount each extra unit sold puts toward paying off the pool of fixed cost. Once enough units have cleared that pool, every further unit is profit, at least until fixed cost takes a step up.

Breakeven volume = fixed cost ÷ contribution margin per unit. The most direct route to breakeven revenue is fixed cost ÷ contribution-margin ratio, where the ratio is CM as a fraction of price. It saves you rounding a fractional unit count. Margin of safety measures how far current volume is above breakeven, as a share of current volume: (actual volume − breakeven volume) ÷ actual volume. It tells you how much cushion a business has before a downturn turns it unprofitable.

Run it once, fully, on a subscription box business. Fixed costs are $840,000 a year. Each box sells for $45 and costs $27 in fulfillment and goods, so the contribution margin is $18 per box. Breakeven volume is 840,000 ÷ 18 = 46,667 boxes, rounded up because nobody sells a fractional box. Cross-check by the revenue route. The CM ratio is 18 ÷ 45 = 0.40, so breakeven revenue is 840,000 ÷ 0.40 = $2.10 million, and 46,667 boxes × $45 also comes to $2.10 million. Both routes agree. If the business actually sells 60,000 boxes, its margin of safety is (60,000 − 46,667) ÷ 60,000 = 22.2%. Volume could fall by just over a fifth before the business tips into loss.

The sharper insight, and the more valuable one, is sensitivity. Contribution margin is the denominator of breakeven volume, so a proportionally small change in cost or price can move breakeven by a disproportionately large amount. Take a firm with fixed cost ₹9,00,000, price ₹180 and variable cost ₹126. CM is ₹54, and breakeven is 9,00,000 ÷ 54 = 16,667 units. One supplier pushes variable cost up by ₹9, to ₹135. That's a 7% rise in variable cost, on a price that hasn't moved. Contribution margin falls to ₹45, and breakeven volume rises to 9,00,000 ÷ 45 = 20,000 units. Required volume jumped 20% from a cost move that was itself only 7% of price. Whenever a case touches a cost increase, say this asymmetry out loud: "a small move in variable cost is not a small move in breakeven, so check the denominator before you eyeball the impact." A cost change you'd barely notice on an invoice moved the volume target by more than three thousand units.

Ratios as a consistency check: reverse-engineering the other direction

You compute a ratio forward, and it's also one of the fastest live tools for testing whether two disclosed figures can both be true at once. When a source hands you a rate and an absolute figure, you can back-solve the implied third figure and check it against whatever else you've been told. A mismatch, especially one off by a clean multiple like 10×, is almost never a rounding artifact. It's a units slip, a mismatched time period, a different denominator, or a real red flag, and every one of those deserves a name before you build anything on top of the numbers.

Start with a clean case where the check confirms things. A company reports a profit of ₹9 crore and a 15% net margin. The implied revenue is 9 ÷ 0.15 = ₹60 crore. If the same source separately claims ₹60 crore of revenue, the two figures tie out and you can carry on with confidence.

The more common case, and the more instructive one, is where the check fails. A ride-hailing operator states 2.4 million rides a month, an average fare of ₹180 and a 22% platform take rate. Separately, it claims an annual platform revenue run-rate of ₹1,140 crore. Build the figure bottom-up: 2.4 million × ₹180 × 22% = ₹9.504 crore of platform revenue per month. Annualized, 9.504 × 12 = ₹114 crore a year, roughly one-tenth of the claimed ₹1,140 crore. A gap that clean, a near-perfect 10×, is the signature of a scale slip (crore quoted where lakh was meant, or vice versa). It doesn't describe a real business fact. Don't silently accept the bigger number, and don't silently accept the smaller one either. Say, out loud, "those two don't tie out. Before I use either figure, I want to know which base or which period is different." Questioning a client's own deck in front of them can feel awkward. Building a model on the wrong figure is far more awkward. Worked Example 5 runs this check from start to finish, and Worked Example 8 builds a second ratio check into a live market-share narrative.

{
  "skill": "Pressure-testing a case-math SETUP (the formula, the base, the fixed-vs-variable split) during solo practice, before the habit calcifies",
  "use": "When practicing alone between sessions, ask an AI to check whether your worked solution's setup is structurally correct — right numerator and denominator, right base for a percentage, right treatment of fixed versus variable cost — rather than asking it for the final number. Treat it as a second pair of eyes on your method, not a source of the answer.",
  "tools": ["Chat assistants — Claude, ChatGPT, Gemini (paste your full worked solution and ask it to audit the setup, not solve from scratch)"],
  "prompt": "Here is a case-math problem and my complete worked solution, step by step: <paste your work>. Do not just give me the final number — tell me whether my SETUP is correct (right numerator/denominator, right base for any percentage, correct fixed-vs-variable treatment), and name the exact step where it breaks down if it does.",
  "verify": "An AI auditing your setup can itself misapply the very conventions it is supposed to be checking — for instance, flagging a correctly weighted average as wrong because it silently expected a flat average. Treat its verdict as a second opinion, never a ruling: recompute the disputed step yourself from the underlying definition (fixed vs. variable cost; the actual base a percentage is measured against), and if your re-derivation and the model's verdict disagree, trust the re-derivation (Guardrail: Trace back to the definition, not the model; Recompute the disputed step by hand).",
  "diy": "The definitions themselves — what a breakeven volume actually divides, why a blended margin must be weighted and not averaged, what base a percentage change is measured against — are exactly what this module installs. AI can flag a step that looks suspicious; only you, holding the underlying definition, can say which of you is actually right.",
  "market": "IN"
}

Presence: narrate the arithmetic, do not perform it in silence

The last habit isn't arithmetic at all, and most technically sharp people skip it. Going silent while you calculate reads as being stuck, even when you aren't. It costs you the room's confidence at the very moment you're about to win it back with a good number. The fix is simple to state and takes real repetition to make automatic: talk while you work. It'll feel awkward the first few times you do it out loud, and the awkwardness fades with practice. "That's roughly a fifth… call it twenty percent… times sixty crore… so twelve crore, plus the base…" is no slower than working silently and announcing the total. It's the same calculation, made visible, and it lets the room correct you in real time if you've mis-stated an input, so you don't discover the error three steps later. State your assumptions as you make them ("I'll treat that as an average across the year, since we don't have quarterly detail"), and state your roundings as you take them. A confident, narrated approximation beats a silent, exact one every time, because the room can't grade a number it never saw you build.

Common mistakes & how experts think differently

1. Treating a percentage-point move and a percentage (relative) move as the same number. "Share is up 5%" when the speaker means "up 5 points" (from 20% to 25%, which is a 25% relative move) understates the actual change by a factor of five. Experts state which one they mean every time, out of habit more than caution.
2. Adding successive percentages instead of applying the exact correction. Two 10% rises make 21%, not 20%, because of the a × b ÷ 100 correction term that's always there. Experts either apply the correction explicitly or, for single-digit rates where it's under a point, say that they're knowingly ignoring a small, bounded error. They never just add without noticing the assumption.
3. Assuming equal-and-opposite percentage moves cancel. "Up 20%, down 20%" nets a 4% loss, and the loss grows with the square of the move. Experts recognize the shape on sight and never let a client's "well, it evened out" story pass unchallenged.
4. Assuming the percentage needed to recover a loss equals the loss. A 50% decline needs a 100% gain, not 50%, to fully recover. The recovery percentage is always larger than the decline, and the gap widens sharply as the decline grows. Experts reach for d ÷ (1 − d) by reflex whenever a client says "we lost X% and then made it all back."
5. Averaging ratios instead of weighting them by their base. A flat average of a 25% and a 10% margin (17.5%) isn't the blended margin of ₹60 crore and ₹40 crore of revenue at those margins. Weighted correctly, it's 19%. Experts ask "weighted by what?" before they average anything, and they separate a real margin improvement from a pure mix-shift effect before drawing any conclusion from it.
6. Confusing average cost with marginal (variable) cost inside a breakeven calculation. Breakeven divides fixed cost by contribution margin, which is price minus the cost of producing one more unit. Dividing by average total cost would double-count, because fixed cost is already baked into it. Experts pin down the fixed/variable split before touching the breakeven formula. They also know that a small change in variable cost can move breakeven volume by a disproportionately large percentage, because contribution margin is the denominator.
7. False precision, computing to the fourth decimal when the inputs are only known to ±10%. Grinding out "₹114.048 crore" from inputs that are themselves rough estimates doesn't make the answer any truer. It just spends the room's patience on digits that carry no information. Experts round to the precision the inputs support, and they say so.
8. Rounding silently, so small errors compound invisibly across a multi-step chain. Each rounding might be defensible on its own. Three or four of them in the same direction, unstated, can drift a final answer well outside the tolerance anyone would have accepted had they seen it happen. Experts round late, round out loud, and now and then sanity-check the final order of magnitude against a completely independent estimate.
9. Going silent while computing, then announcing a number. Silence reads as being stuck. It invites the room to lose confidence in the answer before it even arrives, and it gives no one a chance to catch a mis-heard input before it spreads. Experts narrate the arithmetic as routine. Thinking out loud and getting it right beats getting it right after a pause.
10. Not sanity-checking the final number's order of magnitude. An answer that's off by 10× (a lakh quoted as a crore, a million as a billion) is usually a units slip, and rarely a subtle modeling error. A five-second gut check ("does a mid-sized city's annual grocery spend really round to a trillion rupees?") would have caught it before it reached the page. Experts run this check as the very last line of every calculation, as automatically as they check their own name.

Run through all ten and one thread shows up. An expert holds a few exact identities firmly enough that they no longer re-derive them live: the correction-term formula, the equal-and-opposite square rule, the reciprocal-recovery rule, the weighted-average formula and the breakeven-sensitivity insight. That frees every ounce of attention for judging which number actually matters to the recommendation, instead of fighting the arithmetic itself.

Worked examples

Worked example 1: Layering percentages: distributor margin then GST (India, ₹)

Meera is walking a client through a pricing bridge live. A manufacturer's base price is ₹240. A distributor adds a 15% margin, and then 18% GST is charged on top of that marked-up price. The client, working fast, proposes adding the two rates: "15 plus 18 is 33, so the shelf price is roughly a third higher, call it ₹319 or so." Meera needs the true number, on the spot.

Solution. Step 1, apply the two percentages in sequence, one after the other. After the 15% distributor margin, ₹240 × 1.15 = ₹276. After 18% GST on that new base, ₹276 × 1.18 = ₹325.68. Step 2, compare against the naive additive guess. The client's shortcut (240 × 1.33) gives ₹319.20, which understates the true shelf price by ₹6.48, a 2.0% miss. Step 3, name the correction instead of only flagging that the shortcut is wrong. The correction-term trick from earlier gives combined % = 15 + 18 + (15 × 18) ÷ 100 = 33 + 2.7 = 35.7%, and ₹240 × 1.357 = ₹325.68, a perfect match. The 2.7-percentage-point correction is the interaction between the two rates. The GST layer taxes the distributor's margin too, and naive addition always drops that interaction.

Read it. Meera tells the client, "the two rates interact, and the interaction is worth 2.7 points, because the GST is being charged on an already-marked-up base, not on the original ₹240." A client can act on that sentence. It also generalizes to any stack of sequential markups, discounts or taxes. Multiply the factors together and don't add the rates. If you have to approximate fast, remember that the naive sum understates any all-positive chain by roughly the product of the rates divided by 100.

Worked example 2: Estimating a CAGR without a calculator (US, $)

Jordan is asked, mid-case, for the annualized growth rate behind a fact pattern: a client's revenue grew from $180M to $315M over five years. There's no calculator, and the interviewer is visibly waiting.

Solution. Step 1, compute the total change and the naive linear guess. Total gain is (315 − 180) ÷ 180 = 75%. Spread across five years, the naive guess is 75 ÷ 5 = 15% per year. Step 2, apply the one direction fact you can always trust. Compounding a constant rate over several years always produces more cumulative growth than the naive linear guess implies, so the true CAGR has to be a little below 15%, never above it. Step 3, cross-check with the Rule of 72. The rate that doubles a business in five years is 72 ÷ 5 = 14.4% per year. But $315M is only 1.75× the starting $180M, well short of a full 2.00× double. So 1.75× falls clearly below the 2.00× a 14.4% rate would deliver, and the naive guess of 15% is already known to overstate. Both signals point to the same answer: "roughly 12%."

Read it. The exact figure, computed afterward with a calculator, is 11.84%. The fifteen-second mental estimate misses by under two-tenths of a percentage point, comfortably inside the working tolerance. Nobody in the room needed to wait for a precise root extraction. They needed a confident estimate pointing the right way, defended with two independent anchors you can state fast (the naive guess's known bias, and a Rule-of-72 doubling comparison). That beats a single guess pulled from thin air.

Worked example 3: Blended margin and a pure mix-shift effect (India, ₹)

Rohan is reviewing a retailer's category economics. This year: Category A does ₹60 crore of revenue at a 25% margin; Category B does ₹40 crore at a 10% margin. Next year, Category A grows to ₹70 crore (same 25% margin); Category B holds flat at ₹40 crore (same 10% margin). The client's finance team reports "margin improved" year over year and wants to know why.

Solution. Step 1, build this year's correctly weighted blend. (60 × 0.25 + 40 × 0.10) ÷ 100 = (15 + 4) ÷ 100 = 19%. A flat average of 25% and 10% would have said 17.5%, already 1.5 points off before any mix shift happens. Step 2, build next year's blend, holding both underlying margins fixed. (70 × 0.25 + 40 × 0.10) ÷ 110 = (17.5 + 4) ÷ 110 = 19.5%. Step 3, attribute the move. Blended margin rose half a point year over year, and neither Category A's 25% nor Category B's 10% moved by a single basis point. The whole improvement is a pure mix-shift effect. Category A, the higher-margin segment, simply grew to carry more weight in the total.

Read it. Before you accept "margin improved" or reject it, ask "is this mix, or is this like-for-like?" A margin improvement driven entirely by mix says nothing about whether either underlying business got structurally better. A client making a capital-allocation decision on the strength of "margins are improving" needs to know which of the two stories is true before betting on it repeating.

Worked example 4: Breakeven, margin of safety, and a cost-sensitivity check (US, $)

Sam is evaluating a direct-to-consumer subscription box: fixed costs of $840,000 a year, a $45 price per box, and $27 of variable fulfillment-and-goods cost per box.

Solution. Step 1, find the contribution margin and the breakeven volume. CM = 45 − 27 = $18 per box. Breakeven volume = 840,000 ÷ 18 = 46,667 boxes, rounded up since you can't sell a fractional box. Step 2, cross-check via the revenue route. CM ratio = 18 ÷ 45 = 0.40, and breakeven revenue = 840,000 ÷ 0.40 = $2.10 million. Selling 46,667 boxes × $45 also lands at $2.10 million, so the two routes agree. Step 3, read the margin of safety at actual volume. At 60,000 boxes sold, margin of safety = (60,000 − 46,667) ÷ 60,000 = 22.2%. Volume could fall by just over a fifth before the business crosses into loss.

Read it. Now the sharper question. How fragile is that 22.2% cushion to a cost shock? Breakeven volume is fixed cost divided by contribution margin, and contribution margin is itself a small difference between price and variable cost, two larger numbers. So a modest percentage rise in variable cost produces an amplified percentage rise in required breakeven volume. You saw this earlier in the ₹9,00,000/₹180/₹126 example, where a 7% rise in variable cost pushed breakeven volume up 20%. A breakeven answer that stops at the point estimate, without asking "how sensitive is this to the inputs I'm least sure of?", has done only half the job.

Worked example 5: A ratio that does not tie out (India, ₹)

Ananya is diligencing a ride-hailing operator's investor deck. It states 2.4 million rides per month, an average fare of ₹180, a 22% platform take rate, and a claimed ₹1,140 crore annual platform-revenue run-rate.

Solution. Step 1, build the figure bottom-up from the three operating metrics. Monthly gross fare volume = 2.4 million × ₹180 = ₹43.2 crore. Platform revenue at a 22% take rate = ₹43.2 crore × 0.22 = ₹9.504 crore per month. Step 2, annualize. ₹9.504 crore × 12 = ≈₹114 crore a year. Step 3, compare against the claim. The deck states ₹1,140 crore, a near-perfect 10× the bottom-up figure.

Read it. A mismatch this clean is the signature of a scale slip, and you shouldn't treat it as a subtle modeling disagreement. Someone, somewhere, has quoted crore where lakh was meant (or the reverse). Or the ₹1,140 crore figure describes a different base entirely, such as total gross fare volume in place of platform revenue, though even that wouldn't close a full 10× gap on these numbers. Don't quietly accept the bigger figure just because it's more impressive, and don't quietly substitute your own. Say this out loud and straight away: "these two don't reconcile; before I build anything on either number, I need to know which base or which period is different." You built that ratio in your head in under thirty seconds. Catching the gap in the room, in real time, is worth more than an hour of downstream modeling built on the wrong figure.

Worked example 6: Fast multiplication under a clock (US, $)

Taylor is asked to size a category fast: 8.4 million units sold annually at an average price of $47. There's no calculator, and the interviewer wants a number in seconds, not minutes.

Solution. Step 1, round to a friendly number and correct. Round $47 up to $50: 8.4 million × 50 = $420 million. Step 2, subtract the overcount. The rounding added $3 per unit that was never there, so take off 8.4 million × 3 = $25.2 million. Step 3, combine. $420 million − $25.2 million = $394.8 million, the same answer direct multiplication would give, reached through two easy multiplications and one subtraction in place of one harder one.

Read it. Now suppose the inputs were messier: 8.37 million units at $46.80, numbers nobody can multiply cleanly in their head. The same friendly rounding (8.4 million at $47) gives $394.8 million against a precise $391.72 million. That's an error of under 0.8%, well inside the ±2% working tolerance, and it took a fraction of the time an exact calculation would cost. Neither "always compute exactly" nor "always approximate" is the rule. You need to know, case by case, that the rounding error is small enough to trust, and be able to say roughly how small.

Worked example 7: Fraction-to-percent shortcuts on a market question (India, ₹)

Kabir is asked to estimate how many households a brand reaches, given that its category penetration is described as "about one in six" of 340 million households nationally. Separately, he has to estimate a segment described as "roughly a seventh" of a ₹49,000 crore market.

Solution. Step 1, convert "one in six" instantly using the fraction-to-percent table. 1/6 ≈ 16.7%, so the estimate is 340 million × 1/6 ≈ 56.7 million households. You compute it as a single division by six, not as a percentage multiplication. Step 2, do the same for "roughly a seventh." 1/7 ≈ 14.3%, and ₹49,000 crore happens to divide cleanly by seven, so the segment is ₹49,000 ÷ 7 = ₹7,000 crore on the nose.

Read it. Both answers took seconds because you recognized the fractions and skipped the detour through a percentage. Dividing 340,000,000 by 6 is a much friendlier calculation than "340 million times 0.1667." When a case gives a share as "one in n" in place of a clean percentage, dividing directly by n is almost always the fastest route. A small memorized table of common fraction-percent equivalents (thirds, sixths, sevenths, eighths, ninths) means you never derive any of them from scratch under a clock.

Worked example 8: An integrated market-share bridge, narrated live (US, $)

Casey is mid-case when the interviewer adds a wrinkle: "the total market grew from $2.10 billion to $2.45 billion over the period, and the client's own revenue grew from $180 million to $220 million, what happened to their share, and how does that compare to the market's growth?" Answering it live and out loud means chaining several of the operations you've just learned.

Solution. Step 1, compute each growth rate separately, narrating as you go. "Market grew $2.10 billion to $2.45 billion, that's up $350 million on 2,100, so about 16.7%. Client grew $180 million to $220 million, up 40 on 180, call it 22.2%." Step 2, compute share at each end point. Starting share: 180 ÷ 2,100 = 8.57%. Ending share: 220 ÷ 2,450 = 8.98%. Step 3, state both the point change and the relative change, explicitly distinguishing them. "Share moved from about 8.6% to about 9.0%, that's roughly four-tenths of a point. It sounds small, but relative to where we started it's a 4.8% share gain (0.41 ÷ 8.57). And the client's growth rate is already running about a third faster than the market's own 16.7%."

Read it. Look at what happened across those three steps. A growth-rate calculation, a ratio calculation and a percentage-point-versus-relative distinction got chained in under a minute. All of it was narrated, so the interviewer could follow the logic instead of sitting and waiting for a number. Each operation on its own is easy. The hard part, and the part you're training, is running several of them back to back, out loud, without losing the thread or the room.

Practice set

Work each problem fully before you read its solution. Don't assume a calculator. Where one would obviously help, note that you used it. Numeric answers are accurate to about ±2% unless stated otherwise.

P1. Percentage point vs. percent. A category's penetration moves from 40% to 46%. State both the percentage-point change and the relative percentage change.

Solution. Percentage-point change: 46 − 40 = +6 points. Relative percentage change: 6 ÷ 40 = +15%. The two numbers describe the same event and aren't interchangeable, so always name the one you're stating.

P2. Successive percentage changes. Revenue rises 20% and then falls 15%. What is the exact net percentage change?

Solution. Using the correction-term trick: 20 + (−15) + (20 × −15) ÷ 100 = 5 − 3 = +2%. Check by multiplying the factors: 1.20 × 0.85 = 1.02, confirming a net +2%, not the naive guess of +5%, and not zero.

P3. Rule of 72, forward. A quantity doubles in 6 years. Estimate its annual growth rate using the Rule of 72, and state whether the true rate is a little above or below your estimate.

Solution. 72 ÷ 6 = 12%. The true rate for an exact 6-year double is about 12.25%. At this end of its range the Rule of 72 slightly underestimates the rate (equivalently, it overestimates the years needed at a given rate), though the miss is under a third of a point and well within case tolerance.

P4. Reverse percentage. A price of $460 already includes a 15% markup over cost. What was the original cost?

Solution. Divide, never subtract: 460 ÷ 1.15 = $400. Check: 400 × 1.15 = 460. Subtracting 15% of $460 would have wrongly given $391, because 15% of the final, marked-up price is a different quantity from 15% of the original cost.

P5. Weighted average, blended margin. A firm has three lines: ₹50 crore at 30% margin, ₹30 crore at 15% margin, and ₹20 crore at 5% margin. Compute the correctly weighted blended margin, and compare it to the (wrong) simple average of the three margin percentages.

Solution. Weighted: (50 × 0.30 + 30 × 0.15 + 20 × 0.05) ÷ 100 = (15 + 4.5 + 1) ÷ 100 = 20.5%. The naive simple average of 30%, 15%, and 5% is (30+15+5) ÷ 3 = 16.7%, nearly four points too low, because it ignores that the largest line also carries the highest margin.

P6. Mix shift, continued from P5. The same firm, one year later: the first line grows to ₹65 crore (still 30% margin), the second shrinks to ₹20 crore (still 15% margin), the third holds at ₹15 crore (still 5% margin). Recompute the blended margin and attribute the change from P5.

Solution. New blend: (65 × 0.30 + 20 × 0.15 + 15 × 0.05) ÷ 100 = (19.5 + 3 + 0.75) ÷ 100 = 23.25%, up from 20.5% in P5, a rise of 2.75 points, entirely from mix, since none of the three underlying margins moved at all. The improvement is real for the total, but it says nothing about whether any individual line got better.

P7. Breakeven volume. Fixed costs are ₹12,50,000/year; price is ₹95/unit; variable cost is ₹63/unit. How many units are needed to break even?

Solution. Contribution margin = 95 − 63 = ₹32/unit. Breakeven = 12,50,000 ÷ 32 = 39,062.5, rounded up to 39,063 units.

P8. Breakeven sensitivity to a cost change. A firm has fixed costs of ₹9,00,000, price ₹180, and variable cost ₹126 (so CM = ₹54, breakeven = 16,667 units). A supplier raises variable cost to ₹135. By what percentage does the required breakeven volume increase?

Solution. New CM = 180 − 135 = ₹45; new breakeven = 9,00,000 ÷ 45 = 20,000 units. Percentage increase in required volume: (20,000 − 16,667) ÷ 16,667 ≈ 20%. That comes from a variable-cost rise of only ₹9 on a ₹180 price, a 5% price-relative move. Contribution margin, the shrinking number in the denominator, absorbs the entire ₹9 hit.

P9. Ratio consistency check. A company reports a profit of ₹9 crore and states its net margin was 15%. It separately claims revenue of ₹60 crore. Do the three figures tie out?

Solution. Reverse-engineer revenue from profit and margin: 9 ÷ 0.15 = ₹60 crore, which matches the separately claimed revenue. The figures are internally consistent, so you can go on to use any of the three, confident they weren't independently made up.

P10. Fast multiplication. Estimate 42 × 68 using a friendly-number shortcut.

Solution. Round 68 up to 70: 42 × 70 = 2,940. Correct for the overcount: 42 × 2 = 84. 2,940 − 84 = 2,856: exact, in two easy steps.

P11. Fraction-to-percent conversion. A market segment is described as "about one in nine" of a ₹36,000 crore total market, and a second is "one in twelve." Estimate both segment sizes without long division.

Solution. 1/9 ≈ 11.1%, so the first segment ≈ ₹36,000 × 1/9 = ₹4,000 crore. 1/12 ≈ 8.3%, so the second segment ≈ ₹36,000 × 1/12 = ₹3,000 crore. Both were computed as a single clean division, not as a percentage multiplication.

P12. Rounding error across a chain. You need to multiply three successive growth factors: 1.083, 1.071, and 1.062. If you round each to two decimal places (1.08, 1.07, 1.06) before multiplying, how far off is your final answer from the unrounded chain, in percentage terms?

Solution. Unrounded: 1.083 × 1.071 × 1.062 ≈ 1.2318. Rounded: 1.08 × 1.07 × 1.06 ≈ 1.2249. The rounded chain is about 0.56% below the true figure. That's small and inside the ±2% tolerance, but it's a real drift in one direction, and it would compound further across a longer chain or coarser rounding. So round late, carrying the extra digit through intermediate steps, whenever the chain has more than two or three links.

P13. The reciprocal recovery trap. A price rises 25% during a promotion. What percentage decrease, applied afterward, exactly restores the original price?

Solution. Using the recovery formula for undoing a gain g: needed decrease = g ÷ (1 + g) = 25 ÷ 125 = 20%, not 25%. Check: 1.25 × 0.80 = 1.00. The percentage needed to undo a gain is always smaller than the gain itself. It's the mirror image of undoing a decline, which always needs a gain larger than the decline.

P14. CAGR sanity-check via near-doubling. A ₹250 crore business grows at a steady 9% a year for 8 years. Using the Rule of 72 as a sanity check, roughly what should the ending value be, and how close is the exact figure?

Solution. Rule of 72 at 9%: doubling time ≈ 72 ÷ 9 = 8 years, so an 8-year run at 9% should land very close to a full double, i.e., roughly ₹500 crore. The exact figure (250 × 1.09⁸) is ₹498.1 crore, within half a percent of the doubling estimate, confirming 9% sits inside the Rule of 72's most reliable band.

P15. Weighted average growth rate. A company's two segments: Segment A does ₹100 crore and is growing 30% a year; Segment B does ₹300 crore and is growing 8% a year. What is the company's overall (blended) growth rate?

Solution. Weighted by revenue: (100 × 0.30 + 300 × 0.08) ÷ 400 = (30 + 24) ÷ 400 = 13.5%. That's well below Segment A's headline 30%, because the much larger Segment B, growing far more slowly, dominates the blend. Quote "our fastest segment grows 30%" without this weighting and you'll badly mislead a reader about the company's overall trajectory.

P16. Integrative: weighted breakeven across a product mix. A company shares ₹18,00,000 of fixed costs across two products sold in a fixed 70/30 unit mix: Product X sells at ₹500 with ₹300 variable cost (CM = ₹200); Product Y sells at ₹800 with ₹560 variable cost (CM = ₹240). Find the total breakeven unit volume, and the breakeven volume of each product.

Solution. Build the mix-weighted contribution margin per "average" unit: (0.70 × 200) + (0.30 × 240) = 140 + 72 = ₹212 weighted CM/unit. Total breakeven volume = 18,00,000 ÷ 212 ≈ 8,490.6, rounded up to 8,491 units. Split by the fixed mix: Product X ≈ 0.70 × 8,491 ≈ 5,944 units; Product Y ≈ 0.30 × 8,491 ≈ 2,547 units. Notice that the whole calculation is itself a weighted average, of the two products' contribution margins weighted by their unit-mix share, sitting one layer inside a breakeven calculation. It's the same kind of chaining Worked Example 8 modeled live.

Applied mini-project

Run the clock: three timed case-math reps, cold.

Getting three more problems right is a side effect. The point is to feel the difference, under a real clock with no notes, between "I can eventually get this right" and "I can produce this fast, narrated, and defensibly close." Do all three parts in one sitting.

Do this, in order:

1. Assemble three fresh problems: one chained percentage/growth problem (in the spirit of Worked Examples 1, 2, or 8), one weighted-average-plus-breakeven problem (in the spirit of Worked Examples 3, 4, or Practice P16), and one ratio-consistency check (in the spirit of Worked Example 5 or Practice P9). Write your own numbers and keep them internally consistent, pull three from the Practice set you haven't solved yet, or generate a batch with an AI tool as the drilling callout above suggests, provided you haven't already seen the AI's worked key.
2. Set a clock: 90 seconds for the percentage/growth problem, 3 minutes for the weighted-average-plus-breakeven problem, 2 minutes for the ratio check. Start the clock, and narrate out loud (or into a recording) as you work. Silence fails this project even if the final number is right.
3. Write down your final answer and your rounding decisions the moment the clock stops, whether or not you finished. Do not go back and polish afterward.
4. Compute the exact answer afterward, untimed, and measure your error in percentage terms against it.
5. Self-grade against the rubric below, candidly, before moving on.

Rubric, score yourself out of 10, 2 points each:

• (a) Landed within tolerance. Each of the three final answers is within about 2% of the exact figure. A rough ratio-consistency catch like Worked Example 5's 10× mismatch counts as a full pass even without a precise number, because the point there is catching the mismatch, not computing a tight figure.
• (b) Correct method, visible. For each problem, the operation you ran (multiply the factors, weight by revenue, divide fixed cost by contribution margin, back-solve the ratio) was the right one, and you didn't just land a lucky number that happened to be close.
• (c) Rounding stated, not silent. You can point to at least one moment in each solve where you rounded, and you said so at the time (or would have, out loud, in a live room).
• (d) Narrated, not silent. You were talking (or writing a visible trail) throughout the calculation, not thinking silently and then announcing a number.
• (e) Finished inside the clock, or named where you'd have kept going. Either you landed a defensible answer inside the time budget, or, if you ran over, you can say which step cost you the extra time and what you'd drill to fix it.

Pass at 8/10, with no zero on any single item. Below that, rerun the project with three new problems after redoing the Practice set items that match your weakest rubric line. Don't just repeat the same three problems, since by then you'll have partly memorized their answers.

Drill: this project's gated companion is the mapped case-math drill pack for this node, the timed percentage / growth / weighted-average / breakeven / ratio-consistency problem bank in the Consulting Practice tab. Pass that drill pack at ≥85%, alongside the mastery quiz below, and this node is complete.

Reading & resources

Everything below sharpens the reflex. None of it replaces running the clock on real numbers.

The case-math and case-format canon:

• Marc Cosentino, Case in Point: the standard case-format reference, with worked math embedded throughout every case type. [Paid] [Beginner]
• Victor Cheng, Case Interview Secrets, plus the Look Over My Shoulder (LOMS) recordings and the free material at caseinterview.com: full case walkthroughs where the five operations show up live, under real time pressure, the way you've practiced them here. [Mixed, book and LOMS paid, site free] [Beginner–Intermediate]
• RocketBlocks, PrepLounge, CaseCoach, and Management Consulted, timed drill banks that isolate percentages, growth, weighted averages, and breakeven as standalone speed exercises, separate from full case structure. [Free tiers, paid upgrades] [Beginner–Intermediate]

Free, real practice material:

• The MBA consulting-club casebooks (Wharton, Kellogg, Fuqua/Duke, Darden, INSEAD, LBS, Ross, Booth, Columbia): free downloadable transcripts of real interview cases, where the arithmetic appears the way it does live and not as a cleaned-up textbook exercise. [Free] [Intermediate]
• Official practice cases published by McKinsey, BCG, and Bain on their own career sites. [Free] [Beginner–Intermediate]

Sanity-check references (for the order-of-magnitude gut check, Core Content and Worked Example 5):

• Aswath Damodaran's data library, useful for cross-checking whether an estimated ratio or figure is anywhere near real-world sector norms, the same instinct you've drilled here on a smaller scale. [Free] [Intermediate]

Elsewhere in the course, where the arithmetic gets applied at full depth (never re-taught here):

• M2.01 (Ratio Toolkit I. Profitability & DuPont) and M2.02–M2.03 (the rest of the ratio toolkit and returns on capital): the full, unhurried version of the weighted-average and ratio reasoning you speed up here.
• M4.02 (Unit Economics): the full contribution-margin, CAC/LTV, and cohort build, of which your breakeven arithmetic is the fast-forward version.
• M3.01 (Time Value of Money): the rigorous compounding and discounting machinery behind the CAGR shortcuts.
• CN2.02 (Market Sizing & Guesstimates), where the arithmetic gets put to work building and defending a full estimation tree.

Do this, not just read: for one week, whenever a percentage, a growth rate, or a ratio appears in a news article, a filing, or a conversation, do the mental-math check before you read the source's own interpretation of it. Compute the relative change from the point change, check whether two disclosed figures actually tie out, estimate the CAGR behind a multi-year figure. Ten minutes a day of this, on real numbers you didn't choose, will do more for your speed than another hour of solved practice problems.

Teach it back & journal

Feynman prompt. Explain to a friend who has never done a case why "our stock fell 50% last year and rose 50% this year" isn't a wash, using the fact that undoing a 50% decline needs a 100% gain, not 50%. Then, in the same breath, explain how "margins improved" can be entirely true and entirely uninformative at once, using the mix-shift example: a blend can rise even when neither underlying segment moved a single point. Next time your friend hears a blended number quoted, do they ask, unprompted, "wait, is that a real improvement, or just a mix shift?" If so, you've installed the reflex all of this was built to create.

Journal prompt. For one working week, apply the five operations to real numbers you didn't choose: a news article's growth claim, a company's disclosed margin, a ratio in a filing or an annual report. Log how many times you caught a percentage-point-versus-percent conflation, an unweighted average standing in for a blended one, or two disclosed figures that quietly don't tie out. Which single habit most changed how fast and how confidently you could produce a number this week: the correction-term trick, the Rule-of-72 sanity anchor, or narrating out loud as you compute? Then write down one real case, interview or meeting coming up in the next month where this arithmetic will show up live. Name which of the five families you're least confident will hold up under a clock, and go drill that one.

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