Market Sizing in a Venture Interview: The Bottom-Up Method

Every founder's pitch deck contains a TAM slide, and a striking number of those figures trace back to the same handful of market-research reports. That is not a reason to trust the number. It is the reason venture interviewers ask candidates to build one from scratch, out loud, with no deck and no report to lean on.

The instinct most candidates bring to a market-sizing question is to recall a figure: a headline market size read somewhere, delivered with confidence and no visible reasoning behind it. That instinct is backwards. A number with no visible construction cannot be argued with, which makes it useless to an interviewer whose job in that moment is to find the weakest link in the reasoning and press on it.

A bottom-up estimate does the opposite. Built from a population, a penetration rate and a price, it exposes every assumption behind the final figure, which is the actual point of the exercise. The number is not the deliverable. The assumptions are, and what follows is about building, defending and sanity-checking them the way a fund actually wants to see it done.

Why the Question Is Asked at All: The Power Law

A venture fund does not need most of its portfolio to succeed. It needs at least one outcome large enough, on its own, to return the entire fund to its investors, because venture returns are dominated by a small number of very large outcomes rather than an even spread of moderate ones.

Fred Wilson of Union Square Ventures has written about this "power law" repeatedly on his own blog, describing fund returns as concentrated in a handful of the largest exits in a portfolio rather than in its median company.

That structure is why a market-sizing question is really an underwriting question wearing a different costume. If the total addressable market for a company is a few hundred million dollars, no amount of flawless execution turns it into a fund-returning outcome, because the ceiling on any possible result is set by the market underneath it. A brilliantly run company in a market too small to matter is still, mechanically, the wrong answer for a venture fund.

Bessemer Venture Partners makes this explicit in its own published guidance to founders: "The number one reason we pass on vertical software companies is that we lack confidence in the total addressable market." Not the product. Not the team. The market ceiling itself, which is exactly the judgement a candidate is being tested on when asked to size one from nothing.

TAM, SAM and SOM, Defined Precisely

The three acronyms are worth defining precisely, because the definitions do real work rather than functioning as jargon. Total addressable market (TAM) is the revenue that would exist if every possible buyer in a category bought the product, an upper bound rather than a forecast. Serviceable addressable market (SAM) narrows that to the slice a specific company, with its specific product, geography and customer definition, could realistically reach.

Serviceable obtainable market (SOM) narrows further still, to what that company could realistically capture within a stated period, given competition and its own go-to-market capacity.

Those definitions sound tidy. In practice they are argued about even by professional investors. Pear VC surveyed 30 investors on exactly this question and found real disagreement over what each term actually means, concluding that founders should state their own methodology explicitly rather than assume the acronyms carry one shared, precise definition. The lesson generalizes directly to an interview: naming the tier being sized, and how it is being calculated, matters more than getting the acronym technically right.

A real company's own numbers make the distinction concrete. Toast, Inc., the restaurant-technology company, filed its IPO registration statement with the SEC in August 2021, and it disclosed figures at all three tiers using two different methods side by side.

SOM (actual ARR)

$494M

SAM (near-term addressable)

$15bn

TAM (industry ceiling)

$55bn

Three tiers of the same market, from what Toast had actually captured to the outer ceiling above it.

The Bottom-Up Method: Population, Penetration and Price

A bottom-up estimate reduces almost every market-sizing question to a small number of moving parts, each of which can be stated, sourced where possible, and argued about individually:

• Population or account base — how many potential buyers exist, defined precisely: not "everyone who could ever want this," but a countable group, ideally anchored to a public figure.
• Penetration — what share of that population would realistically buy the product within the period being sized, stated as an assumption when no public figure exists.
• Frequency and price — how often the purchase happens and what it is worth each time, which for a subscription business collapses into a single annual price.
• Revenue — population multiplied by penetration multiplied by price, arriving at the number, with the least certain input named out loud.

Bessemer Venture Partners' own published framework for healthcare markets, which it calls VIPR, follows the identical shape: it defines an "Immediate TAM" as the number of relevant buyers multiplied by how often the need occurs multiplied by what is paid per buyer, and a broader "Big Hairy TAM" as the same arithmetic run over a wider buyer base. Different vocabulary, same construction.

The discipline that makes any of this defensible is naming the population precisely. "The market for enterprise software" is not a population; "companies in the Fortune 2000 with more than 500 engineers" is. A number built on a vague population cannot be pressure-tested, because there is no countable group underneath it to check.

A Worked Build: Sizing Restaurant Software From Scratch

Toast's own 2021 filing gives a full, real set of inputs, which makes it possible to build a bottom-up estimate independently and then check it against what the company actually disclosed, the exact sequence an interviewer wants to see.

1. Population. Toast estimated approximately 860,000 restaurant locations in the United States in its 2021 filing, spanning single-location independents through large chains.
2. Penetration. Not every location is a realistic near-term buyer of a paid, cloud-based point-of-sale platform. Assume, as a candidate would say out loud, that roughly half, 430,000 locations, represent realistic buyers now rather than five or ten years out. State this plainly as an assumption: it is the number in this build most worth challenging.
3. Price. Toast's own filing discloses both its annualized recurring revenue, $494 million, and its location count, approximately 48,000, both as of June 30, 2021. Dividing one by the other gives roughly $10,290 in average annual revenue per location, a real, disclosed figure rather than a guess.
4. Revenue. 430,000 addressable locations multiplied by roughly $10,290 per location comes to approximately $4.4 billion.

Sanity-Checking the Number Three Ways

A bottom-up estimate built entirely from assumptions is still just a guess with better bookkeeping unless it survives being checked against something real. Three checks catch most bad numbers fast, and this example can run all three.

Against a known comparable, the $4.4 billion built above sits well inside Toast's own disclosed $15 billion SAM, using a more conservative penetration assumption than Toast's own methodology implies.

Against a national statistic, the National Restaurant Association reports that eating and drinking places directly contributed $1.4 trillion in output to the US economy, in 2024 dollars, based on 2022 data. Applying Toast's own disclosed technology-spend share, under 3% of sales in 2019 and rising toward roughly 5%, produces a technology-spend ceiling of roughly $42 billion to $70 billion, comfortably bracketing Toast's own $55 billion TAM built from a different base entirely.

Against total category spend, a software-only estimate should always sit below that whole-market ceiling, since software is one line item inside a broader category that also includes hardware, payments processing and financing.

None of the three checks confirms the number is correct. What they confirm is that it is not wrong by an order of magnitude, which is the actual bar an interviewer is applying.

Where Top-Down Is Legitimate: A Cross-Check, Never the Answer

Top-down market sizing is not illegitimate. It is misapplied, almost always in the same way: presented as the entire answer rather than as a check on a bottom-up build that came first.

Toast's own filing is a clean illustration of the correct order. Its SAM, the number closest to what it could sell today, is built bottom-up from a real population and a real price. Its TAM, the outer boundary of what the category could ever be worth, is built top-down, from an industry-wide technology-spend estimate projected forward. Both numbers appear in the same filing, doing different jobs, and neither substitutes for the other.

An interviewer who hears a candidate reach for an industry report before building anything has already learned the candidate's default is recitation rather than construction. An interviewer who hears the same report cited after a bottom-up number, as a check on it, has learned the opposite.

When a Market Is Moving: The Uber Debate

The clearest public dispute over top-down sizing happened in June 2014, when Aswath Damodaran, a finance professor at NYU Stern, sized Uber using a top-down TAM: the global taxi and limousine market at $100 billion, a plausible market-share ceiling applied on top, arriving at a valuation of roughly $5.9 billion, days after investors had priced Uber's round at $17 billion.

Bill Gurley, a Benchmark partner and Uber board member, responded on his own blog the following month. His objection was not to the arithmetic. It was to the input: sizing Uber against the historical taxi market "is making an implicit assumption that the future will look quite like the past."

Uber's own product, in Gurley's account, was creating new price points, new use cases and new occasions for a car ride that the historical taxi market had never counted, because they had never existed. He quoted the company's own founder on the point directly: "it's not about the market that exists, it's about the market we're creating."

Both men were sizing markets carefully. The dispute was entirely about which market was the right one to size: a static historical category, or a category being actively reshaped by the product itself. That distinction, the market that exists versus the one a company is creating, is one of the most common ways a market-sizing answer goes wrong, and it comes back below.

Stating Assumptions So They Can Be Challenged

A defensible assumption is stated in a specific form: the number, where it comes from or why it is reasonable, and an explicit flag on how confident the candidate actually is in it. "Fifty percent" is not something an interviewer can engage with. "Fifty percent, because roughly half of independent restaurants still run on a cash register or an older terminal rather than a modern cloud platform, and that is the number I am least sure of" is.

That last clause matters more than it sounds like it should. Naming the input you are least confident in tells the interviewer exactly where to press, which sounds like a risk and is actually the opposite: it signals that every other number in the build was reasoned through carefully enough to rank against it.

• State the number before anyone asks for it, not defensively after a challenge.
• Attach a reason: a public figure, a comparable, or plain logic, in that order of preference.
• Name the assumption you would revise first if new information arrived.
• Never present a rounded, over-precise number as more certain than the reasoning behind it.

What to Do When the Interviewer Pushes Back

An interviewer who challenges a specific assumption is not signalling that the answer is wrong. In most documented accounts of this exercise, the challenge is the actual test, arriving regardless of which number was given first.

Two responses hold up. The first is updating visibly: taking the new information, redoing the arithmetic out loud, and stating the revised number without defensiveness. The second is defending the original assumption with a specific reason, holding the position if the reason is genuinely sound. Both demonstrate that the number was reasoned rather than recited.

One response does not hold up: freezing, repeating the original number without engaging the challenge, or quietly changing the subject to a part of the answer that feels safer. That reaction tells an interviewer the number was never really owned, only produced.

A Menu of Fast Sanity Checks

Beyond the three checks run against the worked example above, a short menu of general-purpose sanity checks catches most bad numbers within seconds, before any detailed arithmetic is needed:

• Compare to a known company's revenue. If an estimate for a market implies a company larger than the biggest real company anyone can name in that space, the estimate is probably too high.
• Compare to a national statistic. Total consumer spending, total employment or total industry output in a well-measured category puts a hard ceiling on anything built underneath it.
• Compare to total category spend. A specific product's addressable market should always be smaller than the broader category it sits inside.
• Check the implied per-unit number. Divide the total back down to a price or quantity per person, household or company, and ask whether that number is plausible on its own.

None of these checks proves a number correct. Each is fast enough to run inside an interview, and each catches the class of error that costs the most points: being wrong by an order of magnitude rather than being imprecise at the margins.

The Same Prompt, a Weak Answer and a Strong One

Two candidates get an identical practice prompt: size the market for a subscription dog-walking service in a mid-sized city.

The weak answer

• The national pet-care market is worth well over a hundred billion dollars, so even a small share here is a large opportunity.
• Moves straight to an unnamed percentage of the national figure, with no local reasoning underneath it.
• So a few percent of that national market would already be a great outcome.

The strong answer

• Assume the city has roughly 500,000 households. Call dog ownership about a third of them, so around 165,000 dog-owning households.
• Of those, maybe one in ten would pay for scheduled walks rather than a neighbor or doing it themselves: roughly 16,500 households.
• At about $40 a month, that is roughly $7.9 million a year in this city. The number I am least sure of is that one-in-ten adoption rate.

The weak answer never leaves the national figure it opened with. The strong answer never touches it, building instead from a population that can be counted, an adoption rate that is named as an assumption, and a price stated plainly enough to be argued with. Both candidates said a number out loud; only one said something an interviewer could actually engage with.

The Classic Failures

Most weak market-sizing answers fail in one of a small number of recognizable ways.

• Citing a research-firm TAM as the answer itself. A number with an invisible construction cannot be interrogated, and an interviewer who cannot interrogate a number will discount it immediately.
• Sizing the wrong unit. Sizing "healthcare" instead of "self-funded employer health plans," or "restaurants" instead of "independent full-service restaurants," produces a number too broad to say anything about the specific business in front of you.
• Forgetting the fraction. A startup captures a slice of a market, never the whole of it; presenting TAM as if it were an achievable revenue figure skips the step that actually matters.
• False precision. "$847 million" sounds more rigorous than "roughly $800 million to $900 million," but it is not, since the underlying inputs were never that precise to begin with.
• Sizing the market that exists rather than the one being created. The Uber dispute above is the clearest public example: a historical, static market is the wrong anchor for a product designed to expand or replace it.

When the Market Does Not Exist Yet

Every technique above assumes a countable population already exists to build from. Sometimes it does not, and treating that case the same way as an existing market is its own failure.

Steve Blank's market-type framework, developed across his own writing on customer development, separates an existing market, where customers, competitors and demand are already observable, from a new market, where a product creates demand with no prior category to measure. Sizing an existing market is a counting problem. Sizing a new one is closer to a bet: there is no established base, adoption curve or competitor set to anchor the arithmetic to.

The honest approach in that situation is not to force a bottom-up number where no population exists yet. It is to anchor to the nearest adjacent existing market as a floor, and state plainly that anything larger depends on demand that has to be created rather than measured.

Uber, again, is the clean example: the company's own founder framed it as not capturing an existing taxi market but creating a new one, which is exactly this distinction, made by a founder rather than an investor. A candidate who can name which situation, existing or new, applies to the market in front of them has already answered the harder half of the question.

The practical move is to say both numbers out loud rather than pick one and hide the other. Anchor to the taxi market as a floor, then name the adjacent behavior, occasional rides, business travel, drunk nights out, a cheaper product could pull in.

That second, larger number depends on adoption nobody can yet measure, and saying so plainly is a stronger answer than picking either Damodaran's side or Gurley's, because it shows the reasoning that produced their disagreement rather than just choosing a winner.

How to Prepare, in Order

1. Learn the four-part shape cold: population, penetration, price, revenue, so it comes out under pressure rather than being derived from scratch each time.
2. Practice defining the population precisely before touching any arithmetic; a vague population is the single most common reason a bottom-up build falls apart under questioning.
3. Build two or three estimates in categories you know nothing about, so the method is rehearsed independently of any memorized numbers.
4. Rehearse naming your weakest assumption out loud, before anyone challenges it, until it stops feeling like a concession.
5. Practice the two acceptable responses to pushback: updating the arithmetic visibly, and defending an assumption with a stated reason. Freezing is the only response that actually costs points.
6. Run all three sanity checks on every practice estimate: a known comparable, a national statistic, and total category spend, until it becomes reflexive rather than an afterthought.

The Bottom Line

A market-sizing question is not really about the market. It is about whether the reasoning behind a number is visible enough to survive someone else pushing on it, which is exactly what a top-down figure, however large or well cited, cannot offer.

Bottom-up beats top-down for one reason, not several: it is falsifiable. Every input, the population, the penetration, the price, can be named, sourced where possible, and argued about individually, which is the entire point of asking the question out loud instead of over email. A number that cannot be taken apart this way is not a stronger answer. It is one nobody can actually test, which is precisely why it earns no credit.

The candidates who struggle with this exercise are rarely the ones who know the fewest facts about a given industry. They are the ones who reach for a fact instead of building an argument, mistaking a number they can recite for one they can defend.