Market Sizing & Guesstimates

Branch: Casing & Quantitative Fluency (Consulting galaxy) · Node CN2.02 · ~10 focused hours · Mastery gate ≥ 85%

A client, an interviewer, or your own curiosity asks a version of the same question with no textbook answer: how big is this? How many diagnostic tests get run in India each year? What does the US spend on residential solar? Nobody hands you a spreadsheet. Nobody has ever measured it precisely, often nobody can, because the category is new, the data is fragmented across thousands of private firms, or the number simply does not exist in any database you could look up even if you had a week. And yet the business decision in front of you, enter this market, price this product, justify this acquisition, cannot wait for someone to measure it properly. So you build the number yourself, in the open, from assumptions you can defend one at a time, and you hand over not a fact but a defensible range with the seams showing. That is what a market-sizing question, also called a guesstimate, actually tests: not trivia, not a memorized statistic, but whether you can turn "I don't know" into "here is a number, here is exactly how I got it, and here is how confident you should be in it."

Boil the whole subject down and you get this, and the next ten hours are about why you can trust it: a market size is a number you construct, and you aren't finished constructing it until you've built it at least two independent ways and watched whether they agree. When two builds land close together, that's the closest thing to proof you'll get in a world with no ground truth. When they land far apart, you've found the fastest route to the one bad assumption hiding in your logic, and catching it before a client does is what all of this is for. Everything else you'll learn here serves that one habit: the four ways to build an estimate, segmenting a market without double-counting it, defending an assumption instead of guessing at it, and turning a point estimate into a range you can stand behind. Build it twice, reconcile the two, and only then say the number out loud.

By the end of these hours you'll be able to take almost any "how big is X" question and produce a stated range within minutes, whether it's a familiar category or one you've never given a second's thought. You'll build it top-down from an aggregate and bottom-up from a unit, and cross-check it a third way when the first two disagree by more than a rounding error. When two routes to the same number disagree by 3x, you'll know which assumption to interrogate first. And you'll be able to say, out loud and with a straight face, "my range is ₹22,800 to ₹24,000 crore, and the number I am least sure of is X, here is what would change my mind." Construct, cross-check, know your own weak link: that reflex is the orange-belt skill these hours are meant to install. It's also one of the most-practiced moves in every case interview and every early client engagement you'll ever run.

Learning objectives

By the end you can:

1. State, in one sentence, what a market-sizing question actually tests (structured estimation under uncertainty, not a memorized statistic) and explain why a stated range is always the right unit of answer, never a bare point figure.
2. Distinguish a stock (an installed base (how many exist right now) from a flow (a rate) how many transact or get sold per year), convert cleanly between them given a replacement cycle, and treat "what's the market for X" as an ambiguous question until you have pinned down which one is being asked.
3. Classify TAM, SAM, and SOM for any sizing question, the total addressable market, the serviceable slice a specific business model could realistically reach, and the obtainable share it could realistically win this year, and explain why quoting TAM alone, without SAM/SOM, is a classic novice error.
4. Build a top-down estimate (a large aggregate, successively filtered by defensible ratios) and a bottom-up estimate (an atomic unit, scaled up by a chosen driver tree) for the same market, each with every assumption stated aloud and every unit carried through the arithmetic.
5. Run two more sizing techniques beyond the basic pair, a supply-side / capacity cross-check (what could the industry's installed capacity to deliver actually support?) and an analogy / comparables build (borrowing a penetration ratio from a genuinely comparable category when no direct anchor exists), and know when each extra method earns its place.
6. Segment a market MECE-ly for sizing purposes specifically (choosing a cut (demographic, geographic, channel, or use-case) that neither overlaps nor omits) and catch the sizing-specific double-counting traps a generic MECE tree does not warn you about (bundled offerings, multi-sided marketplaces, stock/flow leakage across segments).
7. Triangulate: run two (or three) independent builds, reconcile them into one credible band when they agree, and (when they disagree by a large factor) diagnose the single broken assumption that explains most of the gap, rather than splitting the difference or picking whichever number you prefer.
8. Defend every assumption with a named anchor (a benchmark, a logical bound, an analogy, or a stated illustrative figure flagged for verification) instead of asserting a number from nowhere, and build a best/base/worst range with a one-variable sensitivity check that names which single assumption most deserves further research.
9. Deliver the sizing conclusion answer-first (the range, the input you are least sure of, and the "so what" for the decision at hand) in the consulting register, while recognizing that the deck/storyline craft itself is out of scope here and belongs to the Presentation galaxy.

Prerequisites & connections

Builds on. CN2.01 (Case Math Under Pressure) is the hard prerequisite inside this branch. Every technique here assumes you can already do fast percentage, ratio, and order-of-magnitude arithmetic in your head, round sensibly, and sanity-check an answer without a calculator. What gets added here is the structure on top of arithmetic you should already own cold. CN1.01 (Issue Trees & MECE) and CN1.02 (Hypothesis-Driven Problem Solving) are the galaxy's recommended lead-ins. They're advisory and don't gate anything. The cuts you'll use to size a market MECE-ly belong to the same family CN1.01 teaches for any problem tree. You won't see the algebraic/process/segment/stakeholder theory or its general MECE tests derived again here. The segment cut gets applied to sizing specifically, along with the traps unique to that use. And CN1.02's habit of stating your assumption and then designing the test that would confirm or kill it is what turns a wild guess into a defensible one. If you haven't run those two yet, nothing here is blocked, but you'll notice the family resemblance straight away.

Two on-ramps feed in from outside the Consulting galaxy. E11.03 (The Consulting Case-Interview Lab), the Finance galaxy's phase-11 elective, is the taster. This is the "whole kitchen" for it. Its section 4.4 already introduced top-down, bottom-up, and a five-step sizing routine, and worked two examples: online food delivery in India, and electric-toothbrush replacement heads in the US. If you completed that taster, you already own the vocabulary, so it isn't taught again here. This goes past it. You get a third and fourth sizing method, a real MECE segmentation trap with a numeric reconciliation, and a divergence you have to diagnose yourself instead of a convergence handed to you. You'll also pick up the habit of a stated range with a sensitivity check. If you haven't run E11.03, nothing here depends on it, but its worked examples are good extra reps once you finish. CL-11.3 "Kite Foods Profitability," the app case E11.03 seeds, is a profitability case. So it isn't the mapped drill here. It gets a mention only because it shares the on-ramp.

The math here reuses finance fluency you already hold instead of deriving it again. M2 (ratios) supplies the percentage and share arithmetic. M3 (valuation) supplies the habit of stating an assumption and its source before computing with it. M4.02 (Unit Economics) supplies the idea of an atomic "unit" (a customer, a cohort, an order) that a bottom-up build scales up from, and CN5.01 applies that same lens with full advisory judgment later in the galaxy. Sometimes a top-down build needs a real industry aggregate and not an illustrative placeholder. For that, go to the Finance galaxy's seventeen sector playbooks (M5.01–M5.10, banks through airlines/hotels/media), which aren't taught again here. You learn the sizing technique here. M5.0x supplies the sector's actual, current numbers when you're working a real mandate and not a practice guesstimate.

Feeds forward. CN2.03 (Core Case Formats I, Profitability & Market Entry) uses a market-sizing build as the first block of every market-entry case, "how big is the prize before you ask whether you can win it", and takes the top-down/bottom-up machinery from here as given. CN2.04, CN3.01, and CN3.02 each reuse the triangulation and defensible-range habits directly: sizing a growth branch, a profit-pool revenue estimate for a chain link you can't observe, a job-to-be-done's non-consumption segment. None of them repeats a worked example from here. CN5.02 (Market Attractiveness) and CN5.01 (Business-Model Teardown) both reuse sizing as one skill inside a larger advisory judgment. CN5.03 (the CDD Toolkit) builds its market model directly on the top-down/bottom-up mechanics, adding only the CDD-specific step of reconciling both estimates against a seller's own business plan. Sometimes a sizing conclusion has to be presented as well as computed, turned into a governing thought and a slide. That craft is CN1.03's answer-first storyline and, beyond it, the Presentation galaxy's deck craft. Here you learn only the consulting-specific logic of saying the range and the "so what" out loud. The visual and narrative craft is handed off, and isn't built here.

Core teaching content

1. What a guesstimate actually tests

A market-sizing question, or guesstimate, asks you to produce a number nobody handed you: the annual market for X, the number of Y transacted per year, the total addressable spend on Z. There's deliberately no single correct answer graded against a key. In the real engagements this drill rehearses you for, no answer key exists either. The client is paying you because the number has never been measured cleanly. Knowing a statistic won't get you far. The test is whether you can impose a defensible structure on a question with no data attached, hold every assumption up to the light as you make it, and land on a number a skeptical partner would sign their name to.

Three things separate a strong guesstimate from a weak one, and none of them is getting the exact number. First, you state the structure before the arithmetic starts. Say what you're about to build and why. The second is that every assumption gets a name and a defense instead of being asserted. "I'll assume 20% penetration" is weak. Compare "roughly 20% penetration, in line with adoption of comparable durable goods at this price point in urban India". That one's defensible, because a listener can argue with the anchor as well as the number. Third, and this is the habit everything here is built to install, the answer is a range, built at least two independent ways. A single-method point estimate is a guess wearing a costume. A range built top-down and bottom-up is the closest thing to evidence a guesstimate can produce. Reconcile the two, or explain why they disagree.

Keep one caveat in mind the whole way through. Guesstimates are for decisions that can't wait for real data. When real data exists, use it. If a market-research report, a regulator's published statistic, or a company's own disclosed figure answers your question directly, use it (cited, dated, traced to its source) and skip the guesstimate entirely. The techniques below are for the far larger set of questions nobody has actually measured. That's most of them.

7. Segmenting a market MECE-ly: cuts and sizing-specific traps

The issue-trees course teaches four general-purpose MECE cuts (algebraic, process, segment, stakeholder) for any problem tree. Sizing reuses the segment cut constantly. Most markets aren't homogeneous, and a single blended assumption hides a mix story the same way a single blended margin hides one in a profitability case. The natural segment cuts for a sizing question are demographic (age, income, household type), geographic (urban/rural, region, country), channel (organized/unorganized, online/offline, direct/retail), and use-case or purpose. Worked Example 3's school-tuition vs entrance-exam-coaching vs study-abroad-prep split is a use-case cut. Pick the cut that carries the mix story your question actually cares about. A "should we enter the premium segment" question needs an income or channel cut. A "how many total units" question may not need any segmentation at all.

Sizing brings double-counting traps a generic MECE tree won't warn you about, because real offerings straddle categories that look separate on paper. Three are worth naming. First, bundled products. An Indian coaching institute selling an "integrated school-plus-entrance-exam-coaching" program blurs the line between a school-tuition segment and an entrance-coaching segment. The same rupee can land in both buckets unless you fix a rule (e.g., "count each bundled fee once, in the segment of its primary stated purpose") and apply it consistently. Second, multi-sided markets. A marketplace's GMV, its take-rate revenue, and each side's independent spend are three different numbers. Add a buyer-side estimate to a seller-side estimate of "the same" transaction and you've counted it twice. Pick one side of the transaction, or the platform's own revenue, and stay there. Third, stock/flow leakage across segments. If one segment's data point is a stock and another's is a flow, summing them won't give you a coherent market. Apply the stock-or-flow test from section 2 inside every segment as well as once at the top.

The issue-trees course asks a generic question, "could one thing sit in two branches?" For sizing there's a better test that a segmentation really is MECE, and it's a numeric reconciliation. If you've segmented a top-down total into named buckets, the buckets must sum, in rupees or dollars, to the whole. They must sum exactly, by construction. Approximately isn't good enough. Worked Example 3 shows it: four segments summing to the ₹90,000 crore they were carved from. A segmentation that feels MECE but whose parts don't sum to the stated whole has an arithmetic error or a hidden overlap or gap that the numbers are quietly pointing at, and the second is more common.

8. Defending an assumption: anchors, not guesses

Every sizing build is a chain of assumed numbers, and the chain is only as strong as its weakest defended link. "I'll assume 15% penetration" is an assertion. A defensible assumption names an anchor the listener could check or argue with. Four kinds of anchor do almost all the work. A benchmark cites a known rate from a comparable, better-measured situation, such as a similar-income adjacent category's ownership rate or a well-known population or household count. A logical bound reasons from a hard constraint instead of a memorized figure: "penetration cannot exceed the share of households with reliable electricity access, which puts a ceiling on the number regardless of what any survey says". An analogy borrows a ratio from a comparable earlier category, the method from section 6, named openly as a borrowed comparison. And a stated illustrative figure, flagged as an approximation to verify, is legitimate only when it's labeled that way. Every "approx, verify" figure in the worked examples here is this fourth kind. It's a placeholder plausible enough to reason with, and it isn't a claim to quote to a client unchecked.

Nothing in a guesstimate is certain, so certainty can't be what makes an anchor good. A good anchor is one a skeptical listener can name a way to check. "About 15% of urban households own an AC, similar to reported ownership rates for comparable mid-tier durables at this income level" gives a partner or interviewer something to push back on. "Probably around 15%, I think" gives them nothing but your confidence, which carries zero information about whether the number is right. So build the habit of narrating the anchor in the same breath as the number, every time, until it's automatic. Do it because the anchor is the evidence, and a number without one is just a guess with better posture.

One caution before you go further. Every macro-level benchmark in the worked examples and practice set here is illustrative, flagged "approx, verify" throughout. Don't quote any of them to a real client, a real interviewer, or in a real memo until you've re-anchored it to a current, named, primary source. The AI-Augment callout below is there to make that re-anchoring fast without letting it turn into an unverified habit.

{
  "skill": "Structuring and stress-testing a market-sizing / guesstimate build",
  "use": "Brainstorm a top-down AND a bottom-up segmentation for a market you need to size, generate the 4-6 inputs each route needs, and have the assistant name a plausible primary source for each input — then build and reconcile the estimate yourself.",
  "tools": ["Chat assistants — Claude, ChatGPT, Gemini", "Research/agentic tools — to locate a candidate reference statistic, which you then verify", "Document AI — to digest an industry or sector report you already hold and trust"],
  "prompt": "I'm sizing <MARKET> in <India/US>. Propose one top-down segmentation and one bottom-up segmentation. For each, list the 4-6 inputs it needs, and for every input name a primary source that could confirm it (a census, a regulator's data, a sector report, a company disclosure). Do NOT invent figures — leave every input as a labeled variable for me to fill in.",
  "verify": "Treat every number the assistant proposes as a candidate, not a fact — trace each input to the named primary source before it enters your build (Guardrail #1 Trace, #2 Match on unit/scale/period/scope), and recompute the final arithmetic yourself rather than trusting the assistant's multiplication (Guardrail #3 Recompute). An AI-fabricated 'penetration is X%' looks identical to a correctly recalled one until you check it — the model optimizes for a plausible-sounding input, not a verified one.",
  "diy": "You must already be able to build both routes, defend each assumption with a named anchor, and reconcile or diagnose a divergence entirely by hand (§4-9 of this node) — the assistant only accelerates finding candidate anchors and stress-testing your segmentation for gaps; it never supplies the judgment about which anchor is actually defensible or owns the final number.",
  "market": "IN"
}

9. Triangulation: reconciling two builds, or diagnosing a divergence

Build a market two independent ways and one of two things happens. Both are useful. Convergence means the two land within, say, 10–15% of each other, and it's the strongest evidence a guesstimate can produce. That's because the two builds share no common assumption. A top-down chain (aggregate × category share) and a bottom-up chain (households × penetration × price) fail in different ways when an input is wrong. Two independently wrong builds landing in the same place is far less likely than two correct ones doing so. When they converge, state the band that spans both and say so plainly. Two independent routes agreeing is the finding, and it belongs up front, not in a footnote.

Divergence means the two land a large factor apart: 3x, 5x, more. When that happens the exercise has done its job, catching the mistake before a client does, and that's the reason you build twice in the first place.

The reflex runs in order. (1) Don't average the two and move on. A number halfway between a correct answer and a wrong one is a worse answer dressed up as a compromise. (2) Restate each chain's assumptions in one line and look for a scope mismatch. The most common cause of a large divergence is that the two builds are quietly answering different questions, far more often than a single wrong arithmetic step. In Worked Example 5, the naive bottom-up answers "what if everyone bought packaged water," while the top-down answers "what does the packaged-water category actually spend". That's a population-base mismatch, and no multiplication was wrong. (3) Fix the mismatch you found and recompute. The two should now land far closer, though real diagnostic work often takes more than one round to close fully. Practice P11 walks through this. An initial 3x gap narrows to under 2x on the first correction, and that's progress worth reporting even before the gap is fully closed.

Which assumption broke? There are four usual suspects, listed roughly by how often each one is the culprit. A population-base mismatch applies a per-buyer rate to a denominator that includes non-buyers. It's by far the most common, and Worked Example 5 is given over to it. A stock/flow mix-up (section 2) happens when one chain quietly measured the installed base and the other the annual sale. A definitional/scope mismatch (section 7) uses "organized" in one chain and blended organized-plus-unorganized in the other, as in Practice P11. A unit-price or unit-scale error is ₹ lakh mistaken for ₹ crore, or a monthly figure carried forward as if annual. It's the same unit-and-scale check the AI-Augment callout's "Match" step runs on any AI-sourced figure, turned on your own arithmetic. Say out loud which family a divergence belongs to. That's worth more than the corrected number, because it's the skill that transfers. The next divergence you meet will belong to one of these same four families.

One caution against overcorrecting: not every gap is a mistake to eliminate. A gap of 10–20% between two properly built independent routes is a normal, workable band for a guesstimate. Worked Example 2's three-way solar triangulation spans about 4%, which is tight. Practice P14's US EV-charger cross-check spans about 17%. That's wider, but still the same order of magnitude and still usable, with the softest assumption flagged for a follow-up instead of being treated as disqualifying. Save the full "hunt the broken assumption" diagnostic for gaps of 2–3x or more. Smaller gaps are simply the real width of an estimate built from stated approximations, and papering over that width with false precision is a mistake of its own (section 10).

11. The routine, end to end, and the answer-first close

Pull sections 1 to 10 together and you get one routine to run every time, in this order. (1) Clarify what's being asked: stock or flow (section 2), TAM/SAM/SOM (section 3), units, geography, time period. (2) Choose and state your approach. That means top-down or bottom-up. Supply-side or analogy come in when neither has a clean anchor (sections 4 to 6). (3) Build the tree, segmenting MECE-ly where a mix story matters (section 7), with every assumption named and defended (section 8). (4) Do the arithmetic, carrying units through every step. (5) Build the second (or third) independent route and triangulate. Reconcile a convergence or diagnose a divergence (section 9). (6) State a range with the input you're least sure of named alongside it (section 10). Six steps, every time, whether you have thirty seconds or thirty minutes.

In a real engagement, a defensible range becomes the raw material for an answer-first close. It's the same pyramid, running from governing thought to key line to support, that the so-what lessons teach for any recommendation, applied here to a number instead of a strategic call. Lead with the range. Give the one or two reasons it's credible, meaning the two routes that produced it and how tightly they agree. Name the input you're least sure of, and say what would tighten it. Here's a complete, client-ready version: "The market is ₹22,800–24,000 crore, built top-down from category spend and bottom-up from addressable buyers, agreeing within 5%; the number I would revisit first is the addressable-buyer share, since it is the softest input in the chain; a week of retailer interviews would tighten it by half." It's in the same register as the 60-second recommendation you've met, or will meet, elsewhere in this branch. Turning that sentence into a slide, a chart, or a deck's narrative arc is real craft, and it isn't taught here. Storylining and deck construction belong to the Presentation galaxy, and the consulting-specific answer-first logic that feeds into it is covered in the so-what lessons. Your job here ends at the range, the reasons, and the one caveat, said out loud, in one breath, the moment the arithmetic is done.

Common mistakes & how experts think differently

1. Quoting TAM when SAM or SOM was wanted. The novice sizes the whole category and presents it as "our opportunity." The expert asks which of the three was actually requested. Planning a year-one sales target against a national TAM overstates it by whatever factor SAM and SOM would have filtered out, and Worked Example 1's TAM/SOM gap is roughly 37x. If you don't ask the filtering question first, a partner will.

1. A point estimate with no range. "The market is ₹10,237 crore" sounds more rigorous than "₹10,000–10,500 crore" and is actually worse, because it claims a precision the underlying assumptions never supported. The expert always states a band and resists the pull toward extra significant figures.

1. Building only one route and calling it done. A single top-down or bottom-up chain, however carefully built, is a guess wearing a costume. The expert treats the second, independent build as mandatory. It's the only thing that turns a plausible number into evidence.

1. Averaging two divergent estimates instead of diagnosing. Splitting the difference between a 3x-apart pair produces a number with no defensible logic behind it at all, worse than either input alone. The expert restates each chain's assumptions, hunts for the specific scope mismatch, and recomputes only once the mismatch is named (section 9).

1. Confusing stock and flow. If "the market for smartphones" gets an installed-base number when a flow was wanted (or vice versa), you get an accurate calculation that answers the wrong question. The expert restates stock-or-flow explicitly before building anything, the way a case's clarify beat restates scope.

1. Asserting an assumption instead of anchoring it. "I'll just say 15%" gives a listener nothing to push back on but your confidence. The expert names a benchmark, a logical bound, an analogy, or an explicitly flagged illustrative figure, every single time, even under time pressure, because the anchor is the evidence.

1. Non-MECE segmentation that hides a double-count. Bundled offerings straddling two segments, multi-sided GMV summed against take-rate revenue, a stock figure summed against a flow figure inside the same tree: each of these looks like careful segmentation while quietly counting the same rupee twice. The expert runs the numeric reconciliation test (do the named segments sum exactly to the stated whole?) instead of trusting that the buckets "feel" separate.

1. Chaining too many filters off one distant macro aggregate. Sizing a specific niche by starting from "global GDP" and applying six multiplicative guesses is technically top-down and practically useless, because each filter compounds uncertainty instead of resolving it. The expert stops at the smallest aggregate they can genuinely defend a share of, usually one or two filters from a real anchor.

1. Treating an analogy-based estimate with the same confidence as a well-anchored bottom-up. Borrowing a penetration curve from a comparable category is a legitimate last-resort method. But its largest single source of error, whether the analogy is even valid, sits outside the arithmetic. The expert states a visibly wider range for an analogy-based number (section 6) and uses the method only when the two core techniques both have no anchor to build from.

1. Silent unit or scale errors. ₹ lakh mistaken for ₹ crore, a monthly figure carried forward as if it were annual, a per-household number applied per person. Each one looks like careful arithmetic and is really a scale error that multiplies straight through to the final answer. The expert carries the unit label through every line of the calculation, the same check the AI-Augment "Match" step runs on any externally sourced figure.

1. Reaching for the fancy method before the basic pair, or chasing a decimal instead of widening the range. Capacity and analogy sizing exist for when top-down and bottom-up have no anchor at all. Reaching for them first, or for a fourth significant figure of "precision," shows a misunderstanding of what a guesstimate is for. The expert exhausts the two core methods first and spends the saved effort on a wider range that tells the truth about the uncertainty, instead of a narrower, falsely precise one.

1. Burying the number instead of leading with it. A brilliant triangulation followed by a rambling explanation of every input, with the range mentioned somewhere near the end, reads as no analysis at all. The expert leads with the range, gives the reason it's credible, names the weak link, and stops. That's the answer-first close (section 11), every time.

Worked examples

Six workups in the case style: structure named, every number recomputed in Python for this module, units carried through every line. Every figure marked "approx, verify" is an illustrative anchor and not a researched statistic. That labeling is part of the habit you're learning, and none of it is a claim about any real company's actual results.

Worked example 1: Room air conditioners: stock vs flow, TAM/SAM/SOM, and a first triangulation (India, ₹)

Prompt: "Estimate the annual market for room air conditioners sold in India: top-down and bottom-up, triangulated." No further detail given.

Solution. Step 1: clarify. "Market for air conditioners" is ambiguous in the way section 2 warns about. Installed base (stock), or annual unit sales (flow)? A manufacturer's-revenue or market-entry question wants the flow. That's this year's unit sales and their ₹ value, so it's what this build targets. The stock comes out as a side quantity, both because it feeds the replacement-flow calculation and because it shows the TAM/SAM/SOM ladder from section 3 afterward.

Step 2: bottom-up. Start from urban households, the natural buying unit for a home appliance. India has roughly 30 crore households (approx.), of which ~35% are urban, so 10.5 crore urban households. Current AC penetration among urban households is about 15% (approx, check, an illustrative benchmark for a market where AC ownership is still a minority even in cities). That makes the installed base (stock) 10.5 crore × 15% = 1.575 crore units, or 15.75 million ACs in use today. Annual flow has two parts. New adopters come from rising penetration. Assume it's climbing roughly 1.5 percentage points a year, so 1.5% × 10.5 crore = 15.75 lakh units. Replacements come from the stock-over-life arithmetic in section 2. At an average 10-year AC life, that's 1.575 crore ÷ 10 = 15.75 lakh units. Total bottom-up flow = 15.75 + 15.75 = 31.5 lakh units (3.15 million). At an average unit price of ₹32,000 (approx.), that's 31,50,000 × ₹32,000 = ₹10,080 crore.

Step 3: top-down. India's organized/branded consumer-durables-and-electronics retail market (TVs, ACs, refrigerators, washing machines and the like) is roughly ₹1,60,000 crore/year (approx.). AC's share of that basket is about 6.5%. A mid-single-digit share is defensible, since ACs are a meaningful durable category but smaller than television and general appliances combined. So ₹1,60,000 crore × 6.5% = ₹10,400 crore, which implies ₹10,400 crore ÷ ₹32,000 ≈ 32.5 lakh units.

Step 4: triangulate. Bottom-up says ₹10,080 crore (31.5 lakh units). Top-down says ₹10,400 crore (32.5 lakh units). They're within ~3% of each other, which gives a tight, credible band: ₹10,000–10,500 crore, roughly 31–33 lakh units, about $1.2 billion (at ~₹85/$). One chain starts from households and behavior, the other from a retail-spend aggregate and a category share, so they share almost no assumption. Their agreement is real evidence, and arithmetic coincidence can't explain it.

Step 5: TAM, SAM, SOM. The triangulated ₹10,000–10,500 crore figure is the category TAM. Now take a specific mid-premium split-AC brand that sells only through organized urban retail. Split ACs are roughly 65% of the category, and organized urban channels carry roughly 80% of split-AC volume. So this brand's SAM ≈ ₹10,000–10,500 crore × 0.65 × 0.80 ≈ ₹5,200–5,460 crore. That's a real market, but only about half the TAM, because most of the category sits outside this brand's segment and channel. Suppose the brand's realistic obtainable share this year, given its distribution footprint, is 5%. Its SOM is then ≈ ₹260–273 crore. That's the number that should actually anchor its sales target, and it sits three filters below the headline TAM.

Read it. One build, three lessons. Stock (15.75 million installed ACs) and flow (31–33 lakh units/year) differ by roughly 5x, for the reason section 2 predicts: a 10-year replacement cycle. A tight top-down/bottom-up convergence is strong evidence because the two chains don't share an assumption. And the three TAM/SAM/SOM numbers really are different numbers, all correctly derived. A brand that quoted the ₹10,000 crore TAM as "its" market would be answering a different question from the one that matters for planning.

Worked example 2: Residential solar installations: a three-way triangulation (US, $)

Prompt: "Size the current-year US market for new residential rooftop solar installations, in dollars. Build it demand-side two ways and add a supply-side check."

Solution. Step 1: clarify. This is a flow question: installations this year, in dollars, new systems only. The cumulative installed base isn't what's being asked.

Step 2: top-down (demand, aggregate-and-filter). Start from total US home-improvement and renovation spend, roughly $500 billion/year (approx.). Residential solar's share of that basket is about 5%, so $500bn × 5% = $25.0 billion.

Step 3: bottom-up (demand, household funnel). US households number about 130 million (approx.). The owner-occupied share is about 65%, which gives 84.5 million. Of those, about 40% are solar-suitable (adequate roof, sun exposure, not already installed), leaving 33.8 million addressable homes. This year's adoption rate in that pool is about 3%, so 33.8m × 3% = 1.014 million installations. At an average system price of $24,000, 1,014,000 × $24,000 = $24.336 billion. Bottom-up and top-down already agree within ~3%, a tight demand-side band of $24.3–25.0 billion.

Step 4, the supply-side cross-check. Two demand-side builds agreeing is good. A third, structurally independent check is better, because both demand-side chains still share one hidden assumption: that whatever demand exists can actually be delivered. So estimate installer capacity directly. There are roughly 80,000 workers (approx.) in residential solar installation, working in crews of 4, which makes 20,000 crews. A realistic pace is 50 installations per crew per year, about one installation completed every five working days once permitting, scheduling, and on-site work are all counted. That gives 20,000 × 50 = 1,000,000 installations, or $24.0 billion at the same $24,000 average price.

Step 5: triangulate all three. $24.336bn (bottom-up), $25.0bn (top-down), $24.0bn (supply-side). That's a spread of only about 4% across three methods that share essentially no common assumption. State the range as $24.0–25.0 billion, roughly 1.0–1.04 million installations.

Read it. The supply-side check does two jobs. Here it independently backs up the demand-side pair. In a category where demand-side inputs are softer, it would have worked as a hard ceiling instead. If the installer workforce could physically complete only 400,000 installations a year, a demand-side estimate of 2 million would be self-evidently wrong, however carefully its funnel was built. No amount of buyer willingness turns into a sale the industry can't deliver.

Worked example 3: Private test-prep and coaching: MECE segmentation with a numeric reconciliation (India, ₹)

Prompt: "Size India's private supplementary-education ('shadow education') market and segment it MECE-ly by purpose. Then bottom-up cross-check the entrance-exam-coaching segment specifically."

Solution. Step 1, the top-down aggregate and a MECE segmentation. Total Indian private supplementary-education spend is about ₹90,000 crore/year (approx, check, real estimates for India's tuition-plus-ed-tech market vary widely; this is an illustrative anchor, not a researched figure). Segment it by primary purpose, which is a use-case cut: school-level academic tuition (55%), engineering/medical entrance-exam coaching for JEE/NEET (25%), other competitive-exam coaching such as UPSC, banking, SSC and similar (15%), and study-abroad test-prep for IELTS/TOEFL/GRE/GMAT (5%). In rupees that's ₹49,500 crore, ₹22,500 crore, ₹13,500 crore, and ₹4,500 crore. Now run the numeric MECE check from section 7: 49,500 + 22,500 + 13,500 + 4,500 = 90,000 crore, the whole they were carved from. A segmentation whose parts didn't sum to the stated total would be hiding an overlap or a gap. This one passes.

Step 2, bottom-up on the JEE/NEET segment alone. Don't try to cross-check the whole ₹90,000 crore at once. Cross-check the segment that matters most for a specific engagement. Annual JEE Main plus NEET test-takers come to about 30 lakh (3 million) (approx.). Roughly 70% of them pay for some form of coaching instead of self-study, which gives 21 lakh paying students. The average annual coaching fee, blended across online, offline, and city tiers, is about ₹1,00,000 per student. That's a wide blend, since flagship-brand offline coaching runs several times this and pure online options a fraction of it. So 21,00,000 × ₹1,00,000 = ₹21,000 crore.

Step 3: triangulate the segment. Top-down, from the MECE split, gives ₹22,500 crore for this segment. Bottom-up gives ₹21,000 crore. They're within ~7%, a workable band of ₹21,000–22,500 crore for the JEE/NEET coaching segment specifically.

Read it. The double-counting trap from section 7 sits one layer beneath this clean arithmetic. Many coaching institutes sell an integrated school-plus-foundation-coaching program starting in grades 8–10. A fee for that bundle could legitimately be booked in either the school-tuition or the entrance-coaching segment, depending on which analyst you ask. Cleverer arithmetic won't fix that. The fix is a stated allocation rule ("each bundled fee counts once, in the segment of its primary stated purpose"), applied consistently before a single rupee goes into any bucket. Segmentation traps are rarely arithmetic. They're definitional, and the fix is always a rule stated in advance. A recalculation after the fact won't do it.

Worked example 4: A nascent wearable category: analogy sizing plus a channel-based bottom-up (US, $)

Prompt: "A new smart-ring wearable category has no direct sales history to anchor on. Size this year's US market using an analogy-based estimate, then cross-check with a bottom-up build from retail channel data."

Solution. Step 1, clarify and pick the addressable base. No direct precedent exists for this exact product, so the analogy method from section 6 applies. Find a comparable earlier category and borrow its early-stage adoption ratio. Apply it to the most natural first-buyer population: adults who already own some fitness wearable (smartwatch or band), the people most likely to try a new wearable form factor. US adults number about 260 million. The existing-wearable-owner share is about 30% (approx.), which gives a 78 million enthusiast base.

Step 2, the analogy-derived top-down estimate. Borrow the order-of-magnitude early-adoption penetration a comparable earlier wearable category reached at a similarly early stage. Illustratively, that's ~2% of the enthusiast base in a given year. It's an analogy, named as one, and not a historical statistic about any named product. That gives 78,000,000 × 2% = 1.56 million buyers this year. At an average price of $350, 1,560,000 × $350 = $546 million.

Step 3: bottom-up via channel sell-through. Now build a different route, from observable retail distribution instead of household behavior. Assume the category sells through roughly 2,000 specialty-electronics retail doors (approx.), each moving about 23 units a month, plus direct-to-consumer online sales. Say retail doors account for 35% of total volume, with DTC taking the rest, which is typical for an early-stage wearable brand. Retail units = 2,000 × 23 × 12 = 552,000, so total units = 552,000 ÷ 0.35 ≈ 1,577,000. At $350, that's ≈$552 million.

Step 4: triangulate. $546 million from the analogy top-down, against $552 million from the channel bottom-up. They're within ~1%, tight enough to state a confident $540–560 million range.

Read it. Don't let the tight convergence hide the method's real weakness. Both routes ultimately lean on an assumption about how fast a new category adopts. Step 2 borrows it from a comparable category. Step 3 asserts it directly, in a channel sell-through rate that in practice gets calibrated to land near the same order of magnitude. So these two are less independent than the pairs in Worked Examples 1–3, because no anchor here is a hard, well-measured fact. In real work, state this range wider than the arithmetic alone suggests, using the ±50–100% guidance from section 6 and not the ±10–15% a well-anchored bottom-up would earn. Treat the tight convergence shown here as an illustration of the method. It doesn't license you to trust an analogy-based number as tightly as a demand-funnel one.

Worked example 5: Packaged drinking water: a divergence, and hunting the broken assumption (India, ₹)

Prompt: "A junior analyst sizes India's packaged/bottled drinking water market bottom-up at ₹1,68,000 crore. A colleague's top-down estimate says ₹24,000 crore: seven times smaller. Diagnose the gap."

Solution. Step 1: restate both chains. Naive bottom-up: India's population is about 140 crore. Assume average annual spend on packaged water of about ₹1,200 per person (approx.), so 1,40,00,00,000 × ₹1,200 = ₹1,68,000 crore. Top-down: India's total organized packaged-beverages market is about ₹1,20,000 crore a year (approx.), and bottled water's share of that basket is about 20%, so ₹1,20,000 crore × 20% = ₹24,000 crore. The two disagree by 7x. That's squarely "diagnose, don't average" territory (section 9).

Step 2: do not split the difference. Splitting it gives ~₹96,000 crore, a number with no defensible logic behind it at all. Calling that a compromise between a strong and a weak estimate doesn't rescue it, because it's worse than either input alone. Restate each chain's assumptions and look for a scope mismatch first, following the ordered checklist in section 9.

Step 3: find the broken assumption. The top-down chain implicitly covers only people who actually buy packaged water: largely urban dwellers, travelers, and institutional or event buyers. Most of rural India, and a meaningful share of urban India too, drinks tap, filtered, hand-pump, or community water and buys little or no packaged water at all. The naive bottom-up chain applied a per-buyer spend rate to the entire national population, buyers and non-buyers alike. That's a population-base mismatch. It's the most common cause of a large guesstimate divergence, and section 9 lists it first as the most frequent culprit.

Step 4: correct and recompute. Restrict the per-capita spend to the people who actually buy: urban dwellers plus travelers and institutional or event buyers, illustratively ~19 crore people (about 14% of the population). Reapply the same ₹1,200 per person: 19,00,00,000 × ₹1,200 = ₹22,800 crore. That's now within 5% of the top-down ₹24,000 crore, a defensible band of ₹22,800–24,000 crore.

Read it. One fix, restricting the denominator to people who actually buy the product, closed almost the entire 7x gap. Population-base mismatches come first among the four broken-assumption families in section 9 for this reason. Applying a per-buyer or per-user rate to a denominator full of people who don't buy or use the product at all is the fastest way to overstate a bottom-up build by an order of magnitude. It's also the fastest one to catch, because the fix is a single question: "does everyone in my denominator actually buy this?" Ask it before the arithmetic, so a 7x divergence doesn't have to force it on you later.

Worked example 6: Standing desks: a full range, a sensitivity check, and the answer-first close (US, $)

Prompt: "Size the current US annual market (units and $) for home-office standing desks, then build a defensible range and identify which single assumption most deserves a follow-up before you quote this number to a client."

Solution. Step 1: base case, both ways. Bottom-up: US remote/hybrid workers number about 60 million (approx.). About 4% of them buy a standing desk for home use this year, so 60,000,000 × 4% = 2.4 million units. At an average price of $450, that's $1.08 billion. Top-down: the US office-furniture retail market is about $18 billion a year (approx.), and standing desks' home-office segment share is about 6%, so $18bn × 6% = $1.08 billion. The two are set up to land on the same base case here by construction. That frees this example to spend all its effort on the range and the sensitivity instead of another convergence check.

Step 2: best/base/worst. The input with the most real uncertainty is the 4% adoption-share assumption. How durable the remote-work trend is, and how many of those workers actually buy this year, is a true unknown. The worker-base count and the average price are both comparatively well anchored. Hold everything else fixed and let adoption range from 2.5% (worst) to 6% (best). Worst case = 60m × 2.5% × $450 = $675 million. Best case = 60m × 6% × $450 = $1.62 billion. Stated range: $0.7–1.6 billion, centered near the $1.08 billion base.

Step 3, confirm it really is the highest-leverage input. As section 10 showed, every factor in a pure product moves the output by its own percentage change, so there's no purely mathematical reason adoption "matters more." Check the other two inputs at the same kind of move. Price ±10% moves the answer ±10%. The worker-base count ±10% moves it ±10%. But adoption share's plausible range (2.5% to 6%, a swing of −37.5% to +50% relative to the 4% base) is far wider than ±10% on either of the others. Nothing about adoption share is mathematically special. Its uncertainty is simply the largest, and that's what makes it the one worth a follow-up.

Step 4, the answer-first close. "The market is $0.7–1.6 billion this year, centered near $1.1 billion, built from a worker-base times adoption-share times price funnel that matches an independent top-down check exactly at the base case. The number I am least sure of is the adoption share. A swing from 2.5% to 6% moves the answer 2.4x, dwarfing the ±10% moves that price or the worker-base count could plausibly contribute, so that is the one figure worth ten minutes with an office-furniture retailer before this range goes in front of a client."

Read it. Notice what this close leaves out. There's no spreadsheet, no chart, no slide. Turning this sentence into a deck is the Presentation galaxy's craft and gets handed off (section 11), repeated here once more because this is the last worked example. What the close does is lead with the range, give the reason it's credible, and name the one weak link. A client can act on those four sentences straight away, and they carry every habit you've practiced here.

Practice set

Work each problem fully before reading its solution. Numeric answers are to about ±5% given the illustrative anchors involved; every figure not attributed to a named source is illustrative: approx, verify before quoting externally.

P1 (guided). TAM/SAM/SOM. A US startup sells expense-management SaaS to mid-market companies (500–5,000 employees). It estimates roughly 45,000 such US firms, at an average contract value of $18,000/year, and expects to realistically win about 0.5% of this base in year one. Compute the SAM and the SOM, and say which one belongs in a year-one sales plan.

Solution. SAM = 45,000 firms × $18,000 = $810 million. SOM (year one, 0.5% share) = $810 million × 0.5% = $4.05 million. Build the year-one sales plan on the SOM ($4.05 million). The SAM is the real addressable prize the company could grow into over several years, but planning this year's target against it would badly overstate what a young sales team can actually capture.

P2 (guided). Stock to flow. India's installed smartphone base is roughly 75 crore units, with an average 3.5-year replacement cycle. Ignoring first-time adopters entirely, what annual unit flow does replacement alone imply?

Solution. 75 crore ÷ 3.5 years ≈ 21.4 crore units/year (about 214 million) from replacement alone, before a single new (first-time) buyer is added. That's the core arithmetic from section 2: stock ÷ average life = the replacement component of annual flow. Add new adopters and the real annual flow would be higher still. So a "market for smartphones" question has to say stock or flow before either number means anything.

P3 (guided). Top-down only, India ₹. India's total food-services spend is roughly ₹4,80,000 crore/year, of which about 18% is genuinely organized (the rest informal/unorganized), and organized quick-service restaurants (QSR) are about 35% of that organized slice. Size the organized QSR market top-down.

Solution. Organized food-service spend = ₹4,80,000 crore × 18% = ₹86,400 crore; organized QSR = ₹86,400 crore × 35% = ₹30,240 crore. That's a top-down build only. Before you quote it, pair it with an independent bottom-up (outlet count × average daily footfall × ticket size × days), as section 9 says, and only then treat it as a finished estimate. Worked Examples 1–3 all do this.

P4 (guided). Bottom-up only, US $. The US has roughly 130 million households, of which about 62% are single-family homes. Professional lawn-care service penetration among single-family homes is roughly 20%, at an average annual contract value of $1,500. Size the market bottom-up.

Solution. Single-family homes = 130,000,000 × 62% = 80,600,000; customers = 80,600,000 × 20% = 16,120,000; market = 16,120,000 × $1,500 = ≈$24.2 billion. As in P3, this single-method build is a first pass. The complete exercise pairs it with a top-down check (total US home-services or landscaping-industry spend × lawn-care's share) before it counts as a finished guesstimate.

P5 (guided). Spot the non-MECE tree. A learner segments the Indian personal-loan market into: salaried borrowers, self-employed borrowers, women borrowers, and digital-first (app-only) borrowers. Is this MECE? If not, name the fix.

Solution. Not MECE. It overlaps badly. A salaried woman who borrows through an app belongs in three of the four buckets at once, so summing "market size" across them would count her loan up to three times. The fix is to segment along one dimension at a time. Employment type (salaried vs self-employed) is a clean, mutually exclusive cut on its own. "Women borrowers" and "digital-first borrowers" are cross-cutting attributes. Each could be laid across that same segmentation as a further breakdown, but never mixed into the same flat list of buckets as if it were a peer of employment type.

P6 (independent). Full sizing, India ₹. Size the annual Indian market for two-wheeler helmets (₹), top-down and bottom-up, and triangulate. Two-wheeler installed base (parc) ≈ 25 crore; new two-wheeler sales ≈ 1.8 crore/year; assume 1.3 helmets bought per new vehicle (rider plus occasional pillion) and a 3-year helmet replacement cycle across the parc; average helmet price ≈ ₹900. Separately, India's organized two-wheeler accessories-and-safety-gear market is ≈ ₹90,000 crore/year, of which helmets are about 10.6%.

Solution. Bottom-up: new-vehicle helmets = 1,80,00,000 × 1.3 = 2.34 crore units; replacement helmets = 25,00,00,000 ÷ 3 = 8.33 crore units (one helmet per parc unit per 3-year cycle); total ≈ 10.67 crore units × ₹900 = ₹9,606 crore. Top-down: ₹90,000 crore × 10.6% = ₹9,540 crore. Gap <1%, a tight, credible ₹9,500–9,600 crore band, roughly 10.6–10.7 crore units. One chain starts from vehicle counts and a replacement cycle, the other from a category-share ratio. They share no assumption, so the convergence is real evidence and not luck.

P7 (independent). Full sizing, US $. Size the current annual US market for pickleball paddles and gear, top-down and bottom-up, and triangulate. US adults ≈ 260 million, of whom ≈4.5% have played pickleball; of those players, ≈30% buy their own gear (rather than rent/borrow) this year, at an average spend of $120. Separately, the total US sporting-goods equipment retail market is ≈$60 billion/year, of which pickleball gear is about 0.7%.

Solution. Bottom-up: players = 260,000,000 × 4.5% = 11.7 million; buyers = 11.7m × 30% = 3.51 million; market = 3.51m × $120 = ≈$0.421 billion. Top-down: $60bn × 0.7% = $0.420 billion. Gap <1%, a tight band around $420 million. A fast-growing niche category is where triangulation earns its keep. Neither number alone would inspire much confidence, but two independently built routes landing within a rounding error of each other does.

P8 (independent). Building a range from stated uncertainty. A bottom-up build multiplies three inputs: households (a range of 120–140 million, base 130 million), category penetration (a range of 8–12%, base 10%), and average price (a range of $200–260, base $230). Compute the base case and the full best/base/worst range (using each input's own low or high at the same time, not mixed).

Solution. Base case = 130,000,000 × 10% × $230 = $2.99 billion. Worst case (all three at their low end) = 120,000,000 × 8% × $200 = $1.92 billion. Best case (all three at their high end) = 140,000,000 × 12% × $260 = $4.37 billion. Stated range: $1.9–4.4 billion, centered near $3.0 billion. That's a wide band, because all three inputs were allowed to move together in the same direction. For deciding what to research further, the single-highest-leverage method from section 10 (next problem) usually tells you more than this all-inputs-at-once band.

P9 (independent). Highest-leverage assumption. Using P8's base case, the three inputs' individually plausible ranges are: households ±8%, penetration ±25% (relative), price ±15%. Which single input most deserves a follow-up before this number is finalized, and why, given that all three are multiplied together?

Solution. Penetration. In a pure product of independent factors every input has identical mathematical leverage, since a 1% move in any one factor moves the final answer by exactly 1%. So "leverage" here is entirely about which input's plausible range is widest. Households' range (±8%) and price's (±15%) are both narrower than penetration's (±25%). Penetration is simply the number the team is least sure of. That makes it the one worth a data source or an expert call before the estimate is finalized, because its uncertainty dominates the other two.

P10 (independent). Rewrite an indefensible assumption. A colleague defends a sizing input with: "I'll just say about 10% of people probably buy this: seems about right." Rewrite this as a defensible assumption, naming one of the four anchor types from section 8.

Solution. Any of the four anchor types would fix it. Here's one using a benchmark: "I'll assume roughly 10% adoption, in line with reported penetration of comparable mid-priced discretionary categories among the same income segment, a figure I'd confirm against a sector report before finalizing." The rewrite does two things the original didn't. It names something a listener could check independently, and it flags the number as provisional until verified. "Seems about right", which only asserts confidence, becomes an anchored claim someone can argue with.

P11 (independent). Divergence diagnosis. Sizing India's organized/branded preschool (nursery) market: top-down gives ₹18,000 crore (from ₹1,20,000 crore total private K-12 spend × 15% preschool share). A bottom-up gives ₹54,000 crore (6 crore children aged 3–5, 60% enrolled in some preschool, ₹15,000/year average fee), a 3x gap. Diagnose and take one correction pass.

Solution. The likely broken assumption is a definitional/scope mismatch (section 9). The top-down chain filtered specifically to "organized/branded" preschools. The bottom-up chain's ₹15,000/year fee, though, is a blended average across organized/branded AND informal/unorganized preschools. Informal preschools charge far less, and community or government-run options charge nothing. That blended fee was applied to all 3.6 crore enrolled children (60% of 6 crore), whichever type they attend. Now correct for scope. If only about 25% of enrolled children attend an organized/branded preschool (0.9 crore), at a branded-segment average fee of ₹36,000/year, the corrected bottom-up = 90,00,000 × ₹36,000 = ₹32,400 crore. The gap narrows from 3.0x to 1.8x. Report that as real progress, and don't present it as a finished reconciliation. The remaining 1.8x gap means at least one more round of checking is still needed. Is 25% organized attendance too high? Is the 15% top-down share too low? Real divergence-hunting is iterative, and an honest report of a partial narrowing beats forcing a false final match.

P12 (independent). Supply-side cross-check. India has roughly 4,00,000 practicing chartered accountants, of whom about 35% do audit-related engagement work, averaging 12 engagements/year at ₹2,50,000 average fee. Compute the implied capacity-side market size, and compare it to an independent demand-side (top-down) estimate of India's statutory-and-tax-audit fee market of ₹38,000 crore.

Solution. CAs doing audit work = 4,00,000 × 35% = 1,40,000; engagements = 1,40,000 × 12 = 16,80,000; capacity-side market = 16,80,000 × ₹2,50,000 = ₹42,000 crore. Against the demand-side ₹38,000 crore, the gap is ~10.5%. That's the same order of magnitude and a workable band (₹38,000–42,000 crore). The two routes are structurally independent. One counts professionals and their realistic workload while the other estimates client-side demand directly, so their agreement is meaningful corroboration and not coincidence.

P13 (timed, 6 min). Full sizing cold, India ₹. Size India's annual market for professional wedding photography and videography services. (No inputs given, state your own defensible assumptions, build both ways, and triangulate.)

Solution. Bottom-up: India has about 80 lakh weddings a year (approx, check, a widely-cited illustrative figure). About 55% use professional photography/videography, which gives 44 lakh weddings. At an average package price of about ₹85,000, 44,00,000 × ₹85,000 ≈ ₹37,400 crore. Top-down: total Indian wedding-services spend (venues, catering, photography, apparel, and more combined) is about ₹4,50,000 crore/year (approx.). Photography/videography's share is about 7.5%, which gives ₹33,750 crore. Gap ≈10%, a workable band of ₹33,750–37,400 crore. That's wider than several of the other examples here, but still the same order of magnitude. The softest input is almost certainly the photography-adoption share (55%), and it's worth a vendor-association data point before this range is finalized.

P14 (timed, 6 min). Full sizing cold, US $. Size the current-year US market for home Level-2 EV charger installations. (State your own assumptions, build both ways, and triangulate.)

Solution. Bottom-up: new EV sales this year come to about 1.4 million (approx.). About 60% of buyers install a home Level-2 charger the same year, so 8,40,000 installations. At an average install price (hardware + labor) of about $1,800, that's ≈$1.51 billion. Top-down: the total US EV-charging-infrastructure market (home plus public) is about $6.0 billion/year (approx.), and home installations' share is about 21%, giving ≈$1.26 billion. Gap ≈17%, wider than most of the other triangulations here. It's still the same order of magnitude (a ~$1.3–1.5 billion band), but wide enough to flag instead of accepting quietly. The attach-rate assumption (60% of new buyers installing the same year) is the softest number in the bottom-up chain. It's the first one worth checking against an actual installer or utility data source before this range leaves the room.

P15 (timed, 6 min). Analogy sizing, US $. A wearable AI note-taking device (an even newer category than Worked Example 4's smart ring) has no sales history at all. Using the same enthusiast base as Worked Example 4 (78 million adults who already own a fitness wearable) and a lower, more conservative borrowed penetration rate of 0.8% (since the analogy is one step further removed), at an average price of $200, estimate this year's US market.

Solution. Buyers = 78,000,000 × 0.8% = 6,24,000; market = 6,24,000 × $200 = ≈$125 million. The biggest caveat here has nothing to do with the arithmetic. An analogy-based estimate for a category one step further from any precedent than Worked Example 4's already-soft one deserves an even wider stated range: illustratively ±75–100%, against the ±20–30% a well-anchored bottom-up would earn. The thing most likely to be wrong is the analogy itself, more than any input. A truly novel product can adopt far faster or far slower than any precedent suggests, and no amount of arithmetic precision downstream fixes a bad comparison upstream.

P16 (timed, 6 min). The answer-first close. Using Worked Example 6's standing-desk range ($0.7–1.6 billion, base $1.08 billion, adoption-share the highest-leverage input), write the three-sentence answer-first "so what" a consultant would actually say to a client, without building a slide.

Solution. "The home-office standing-desk market is $0.7 to 1.6 billion this year, centered near $1.1 billion, and two independently-built routes (a worker-adoption funnel and a furniture-market share check) agree exactly at the base case. The estimate is most sensitive to the adoption-share assumption, not to the worker-base count or price, so that is the one number worth confirming with a retailer or a channel-check before you commit to it. If you need a tighter range than $0.7–1.6 billion, that confirmation is where the next hour of work should go, not into re-deriving the other two inputs, which are already well anchored." Answer first (the range), the reason it's credible (two independent routes converging), and the one thing that would tighten it. That's the shape section 11 and the answer-first logic of the so-what lessons both ask for. No chart or deck is involved.

Applied mini-project: size two markets, cold, and defend the weak link

Budget ~2 hours. It's the largest hands-on block in the ~10 hours you'll spend on this topic.

Part A, full sizing, India ₹ (45 minutes). Pick a market you haven't built in the worked examples or practice set here, e.g., the annual Indian market for solar water heaters, organized pet food, private cord-blood banking, or licensed sports merchandise. Run the full six-step routine from section 11. Clarify stock/flow, and TAM/SAM/SOM if relevant. Build top-down. Build bottom-up. Triangulate by reconciling or diagnosing. State a best/base/worst range. Then name the single highest-leverage assumption.

Part B, full sizing, US $ (45 minutes). Do the same for a US market you haven't already built, e.g., the annual US market for home water-softener systems, meal-kit subscriptions, or residential backup generators.

Part C, the answer-first close and self-score (30 minutes). For each market, write the three-sentence answer-first close in the style of Worked Example 6 / Practice P16. Then self-score both builds against the rubric below.

Scoring rubric (self-grade honestly; 0/1/2 each; 10 points per market, 20 total):

Pass standard (self-check): ≥8/10 per market, with no dimension at 0, and the answer-first close actually leads with the range (not with the spreadsheet). If either market scores below 8, the fix is targeted: rebuild the weakest-scoring dimension on a fresh market before moving on, instead of re-running a market you've already seen the answer to.

Guesstimate drill. The mapped gate component here is the app's guesstimate drill set (Practice → Consulting tab). It's a bank of structured estimation trees, India and US alternating, auto-checked on two things: (a) whether each assumption you enter falls inside a defensible reference band, and (b) whether your own combined estimate lands inside a wide reference range built the same two-method way you've learned here. Run it to the drill set's own pass mark alongside the mastery quiz below. You need both to complete this topic.

Reading & resources

Use the sources below to learn technique from and to check magnitudes against. Never quote a client-facing figure from them directly. Every illustrative number here is flagged "approx: verify" for that reason.

On sizing and casing specifically:

• Marc Cosentino, Case in Point: the standard case-format reference; its market-sizing chapter is the closest thing to a canonical treatment of the topic. [Paid] [Beginner]
• Victor Cheng, Case Interview Secrets, plus the free videos and the "Look Over My Shoulder" audio series at caseinterview.com, the clearest treatment of thinking out loud through a guesstimate under time pressure. [Free/Paid] [Beginner–Intermediate]
• Free official practice cases from the McKinsey, BCG, and Bain career sites, and the free MBA consulting-club casebooks (Wharton, Kellogg, Fuqua, Darden, INSEAD, LBS, Ross, Booth, Columbia), the deepest free bank of real sizing prompts with model structures. [Free] [Intermediate]
• RocketBlocks, PrepLounge, Management Consulted, CaseCoach (free tiers), timed drill platforms with sizing-specific practice sets. [Free/Paid] [Beginner–Intermediate]

For real anchor data, when a real engagement (not a practice guesstimate) needs one:

• Aswath Damodaran's data library (NYU Stern, freely published), industry-level aggregates, margins, and multiples useful as a top-down anchor's starting point; always dated and sourced, never a guesstimate substitute once real data exists. [Free] [Intermediate]
• India: MoSPI (Ministry of Statistics), RBI, and sector-regulator publications for population, household, and category-level statistics; company annual reports and investor presentations for a specific firm's own disclosed figures. [Free] [Beginner–Advanced]
• US: the US Census Bureau and Bureau of Labor Statistics for population, household, and spending data; SEC EDGAR for any public company's own disclosed segment or category figures. [Free] [Beginner–Advanced]

On the structuring discipline underneath sizing:

• Charles Conn & Robert McLean, Bulletproof Problem Solving, issue trees and structured estimation as part of a general problem-solving loop, beyond the interview context. [Paid] [Intermediate]
• Barbara Minto, The Pyramid Principle: the answer-first, MECE logic that sections 7 and 11 here both borrow directly; read in full ahead of CN1.03. [Paid] [Advanced]

The taster this builds on: E11.03 (The Consulting Case-Interview Lab), section 4.4, and the app's CL-11.3 on-ramp case. It's already in your hands if you ran it. Its vocabulary is assumed here, and the material builds past it, not around it.

Do this, not just read. Pick a category you walk past every day (a coffee chain, a pharmacy aisle, a parking lot) and size it cold, out loud, both ways, before you finish reading this section. Ten minutes of building a real number from nothing teaches more than another chapter of worked examples.

Teach it back & journal

Feynman prompt. Explain to a friend who has never taken a case interview why "how big is this market?" has no single right answer, and why that's a feature of the exercise and not a flaw in it. Use the sentence "a market size is a number you construct, and you are not finished until you have built it at least two independent ways." Then, without prompting, have your friend poke holes in a guesstimate you build out loud (any category: a coffee shop, a gym, a parking garage) by asking "how do you know that number?" at every step. If you can answer every challenge by naming an anchor (a benchmark, a bound, an analogy, or a flagged illustrative figure) instead of shrugging, the habit has stuck.

Journal prompt. Pick three "how big is X" questions from your own daily life this week (a category you buy from, a service you use, an industry in the news) and size each one cold, both ways, before you look anything up. For each, note whether your two routes converged or diverged, and by how much. If they diverged, which of the four broken-assumption families (population-base, stock/flow, definitional/scope, unit/scale) explained the gap once you found it? Then look across all three for the single input you were consistently least sure of. It may point to a kind of number you tend to under-anchor (penetration rates, prices, population counts), and that's worth watching for next time. Write one paragraph naming that pattern and one concrete habit you'll change because of it.

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