Your market number is not being weighed. It is being marked down before the sentence finishes. An experienced investor cannot check a billions-of-dollars figure you lifted from a report, so they treat it the way any priced actor treats a claim they cannot verify: they move it to the low end and work from something they can rebuild instead. The size of that markdown is not a mystery you have to guess at. You can compute it.
Here is the decision this piece is for: which sizing method goes in the memo, and which number you delete before the call.
Your market is not too small to fund. Your number is too big to check. Those are different problems with different fixes, and founders keep solving the first when they have the second. The top-down slide reads well and checks badly. It arrives as one large claim, some billions of addressable dollars and a wedge you will own, and it hands the reader nothing to test. So the reader tests nothing, discounts the claim, and restarts the sizing conversation from their own back-of-envelope number. You were in the room for that conversation and you added nothing to it.
Why these three models
Market sizing under thin data is a disclosure problem before it is an arithmetic problem. Three lenses carry it, and each fails somewhere the other two do not, which is the whole reason to run them together instead of trusting one.
The first asks what a number is worth to a reader who cannot open it. The second asks why a figure everyone in your category cites tells an investor nothing about you. The third asks how a number built from several weak public counts becomes both sturdier and checkable. They span three outcome types: an equilibrium of who discloses what, a complex system where a convention spreads by copying, and a random-error argument about combining independent estimates. Their mistakes do not point the same way.
1. The disclosure: why a number you cannot rebuild is priced at the floor
Verifiable disclosure is the economics of provable claims. The core result, surveyed by Dranove and Jin, is unraveling: when a claim can in principle be proven and proving it is cheap, silence is read as the worst case, so good types keep disclosing to separate themselves until non-disclosure marks you as the bottom.3 An investor runs this on your TAM without naming it. A number they cannot verify is treated as a number you chose not to verify, which is treated as the number that would not survive verifying.
The top-down figure is unverifiable by construction. It is the output of someone else’s proprietary model. Even the good ones are closed boxes: the GSMA builds its mobile money estimates from a mixed bottom-up and top-down model, a spreadsheet of modelling assumptions per country that no reader of your deck can reconstruct.5 So the borrowed figure inherits the floor. And the floor tells you the discount’s size directly. It is the distance between the number you claimed and the number you can document.
Assumes: the reader treats an unverifiable claim as concealment and prices it at the skeptical low end.
Fits because: a consultancy TAM is unrebuildable from your inputs, so it reads as a number you would not stand behind under a check.
Breaks when: the receiver is insufficiently skeptical. In experiments, people under-infer from non-disclosure and only converge on full unraveling once they get feedback, so a naive investor may not discount at all in the room.4
Counteracts: the instinct to reach for the biggest defensible figure, which is exactly the figure that invites the check.
May reinforce: under-claiming, if you only ever cite what is trivially provable.
2. The convention: why a figure everyone cites says nothing about you
Top-down TAMs spread by imitation. One founder cites the report, the next cites the same report, and the number becomes the focal figure every deck in the category lands on. An investor who has seen that exact figure in ten decks this quarter reads it as the category’s wallpaper, not as information about your business. This is an emergent property, not a fact about any one slide: the more decks converge on the number, the less any single citation means. The top-down figure is the category’s wallpaper. It carries no signal that separates you from the ten founders who pasted the same slide.
Damodaran calls the market-level version the big market delusion: a whole cluster of founders and their financiers converge on the same large-market story, each convinced it will be the winner, and collectively overprice until the correction arrives.2 The same shared number can be attacked from the opposite side too. When Damodaran valued Uber near 5.9 billion dollars off a 100 billion dollar global taxi market and a 10 percent share cap, Bill Gurley argued the addressable market could be larger by a factor of 25.1 Notice what that disagreement proves. A top-down TAM comfortably supports two serious numbers a factor of 25 apart. A reader who knows that treats the figure as non-informative and prices off what can be built.
Assumes: your top-down number is one many other founders in the category also cite, so it is common knowledge, not private information.
Fits because: a single widely-quoted report becomes a convention, and a convention cannot distinguish one founder from another.
Breaks when: the figure is genuinely yours: primary research you commissioned that no competing deck holds. Then the number does carry founder-specific signal.
Counteracts: the comfort of citing the same authority everyone cites, which feels safe and reads as noise.
May reinforce: contrarian overreach, if you reject a sound consensus number just to look different.
3. The triangulation: why several weak public counts beat one strong report
Crowd aggregation is the reason a built number is sturdy where a borrowed one is brittle. Francis Galton’s fairground result is the origin: 787 independent guesses at the dressed weight of an ox produced a median of 1207 pounds against an actual weight of 1198, closer than almost any single expert.6 Scott Page later gave the mechanism a clean form, the diversity prediction theorem: the collective error equals the average individual error minus the diversity of the estimates.7 Independent, differently-biased estimates cancel. Correlated ones do not.
Apply it to sizing. Instead of one consultancy figure, triangulate from several public instruments that measure your market through different lenses: a communications regulator’s active-line count, a revenue authority’s register of filing enterprises, a licensing board’s count of outlets. Each is individually weak. Because different institutions produced them by different methods, their errors are independent, so the triangulated number is both sturdier and, the part that beats the discount, checkable. The reader can open every source. The World Bank’s Global Findex is the model of this: a public dataset built from probability samples of about 128,000 adults, with the microdata released, so a stranger can reconstruct the number rather than take it on faith.8 A number a stranger can rebuild beats a number no one can open, even when it lands smaller.
Assumes: your counts come from genuinely independent producers, so their errors are uncorrelated.
Fits because: a regulator’s register, a tax roll and a household survey measure the same market through unrelated instruments a reader can each verify.
Breaks when: the sources are not independent. Three reports all reselling one government feed are the same error three times, and agreement between them reads as accuracy when it is only correlation.
Counteracts: single-source risk, staking a whole slide on one PDF.
May reinforce: false precision, if you dress a triangulated estimate as a hard number.
How to read the size of your own discount
The three models converge on one measurement you can take this afternoon. Build the number from checkable public counts, then set your headline TAM next to it. The gap is the discount. If your slide says two billion and the most you can rebuild from public instruments is eighty million, the investor’s working number sits near the eighty, and the entire distance to your two billion is what they are silently subtracting. You do not have to imagine the haircut. You compute the floor and read off the gap.
The Uber disagreement sets your expectation for how large that gap can be: serious analysts differed on the same market by a factor of 25.1 That is why a disciplined reader will not lean on your top-down figure in either direction. Your market may well be large. The top-down number simply cannot prove it, so they set it aside and price off what can be rebuilt.
The levers, cheapest and most reversible first
None of this needs a data budget, and every step is reversible.
- Compute your discount today. Build the number from two or three public counts, write your headline TAM beside it, and mark the gap. That gap is the number the reader was already subtracting. An hour of work tells you how exposed the slide is.
- Move from asserted to rebuildable. Replace the consultancy citation with source links a reader can open, and show the arithmetic beneath each count. You are converting a claim into a derivation.
- State the one input that moves the total most. Name your largest sensitivity and its vintage. A sized market that shows its own error bar reads as competence, not doubt.
- Delete the borrowed billions. Cut the top-down headline, or demote it to a single labelled line clearly marked as a magnitude check, never as the number.
What to move before the round opens, and what to hold
Sequence it so the reversible, dominant moves happen now and the expensive, irreversible one waits for a trigger.
- Do now, wins in every scenario. By T+3, replace the headline TAM with a rebuildable build from a few verifiable counts, each with a live source link, and put the method in the memo. This helps whether your market turns out large or small, because it is the version that survives a check.
- Hedge, cheap cover against the generalist. By T+14, keep one top-down figure in a footnote, labelled as an order-of-magnitude sanity check, for the investor who wants the big headline and may not discount it. You give them the number without staking your credibility on it.
- Defer and trigger, pre-commit the signal. Do not buy primary market research yet. Write the trigger down: the moment an investor asks for segment depth your public counts cannot reach, and only then commission it. By T+28 with no such ask, you have saved the money and lost nothing.
What usually happens after the slide lands
Match this to the right reference class. Not startups in general, but founders walking into diligence with a market slide in a category official statistics barely cover. In that class the polished top-down number is the common case, and it reliably underperforms: it gets discounted, then quietly set aside, and the investor re-sizes from their own estimate while you watch.
Your present state shifts the odds in your favour on two counts. You have public registers a reader can open, which most sizing debates lack, and you hold your own transaction data, which turns a public count into a revenue estimate with a defensible attach rate. Subtract the counterfactual honestly. A glossy top-down slide would not have won you the term sheet, so building the rebuildable version costs a morning and forfeits nothing you actually had.
One matrix-break flag. If your investor is a mandate-driven generalist who rewards the big headline and never checks, the unraveling logic weakens and the large number can help you in the room.4 Do not mistake that for safety. The check has not been cancelled. It has been moved to diligence, where a number you cannot rebuild fails more expensively and later.
Where this ensemble goes blind
Every one of these lenses can only see what someone has counted. Together they are sharp on the formal, registered, licensed market and blind on the informal share that no register captures, which in many African categories is the larger half. So the rebuildable number is a floor, not the truth. It can understate your real market badly, and the models cannot tell you by how much, because the missing part is missing on purpose. They also do not know your true attach rate. That number is yours to defend, and the public count only sets the ceiling.
So do not present the built figure as the whole market. Here is the action that survives the ignorance. This week, build the floor from the public counts you can link, put it in the memo with its sources, and delete the borrowed billions. Then name the uncounted segment as a separate gap you are pricing, and say plainly what you cannot yet size. The number a stranger can rebuild will out-argue the number nobody can, in every room where the decision actually gets made.
Sources and notes
Ensemble members are named in prose and appear in the Wire Model outcome-type map: verifiable disclosure and unraveling (equilibrium), agent-based emergent convention (complex), crowd aggregation (random). The governance and behavioral layers the method usually appends are folded into the cards, the buyer under-inference in model 1 and the copying convention in model 2, rather than shipped as extra members, because neither adds a lever the three cards do not already give the reader. This piece is the discount side of market sizing. Its sibling, on how to construct the bottom-up number row by row, is a separate build.
- Bill Gurley, “How to Miss By a Mile: An Alternative Look at Uber’s Potential Market Size,” Above the Crowd, 11 July 2014. Damodaran’s base case used a 100 billion dollar global taxi and car-service market and a 10 percent share cap to value Uber near 5.9 billion dollars; Gurley argues the addressable market could be larger by a factor of roughly 25. abovethecrowd.com
- Bradford Cornell and Aswath Damodaran, “The Big Market Delusion: Valuation and Investment Implications,” Financial Analysts Journal 76:2 (2020). Big markets attract clusters of founders and funders who collectively overprice, then correct. Readable author version at Aswath Damodaran, Musings on Markets, December 2019. aswathdamodaran.blogspot.com
- David Dranove and Ginger Zhe Jin, “Quality Disclosure and Certification: Theory and Practice,” NBER Working Paper 15644 (2010), Journal of Economic Literature 48:4, on why sellers do not voluntarily disclose and how non-disclosure unravels toward the worst case. Cited for the theorem, not a figure. nber.org
- Ginger Zhe Jin, Michael Luca and Daniel Martin, “Is No News (Perceived As) Bad News? An Experimental Investigation of Information Disclosure,” NBER Working Paper 21099 (2015), published in American Economic Journal: Microeconomics (2021). Receivers are insufficiently skeptical about non-disclosure and only converge on full unraveling once given feedback. HBS working-paper copy: hbs.edu
- GSMA, “Introducing new mobile money estimates to capture the rapid transformation of the global industry,” Mobile for Development. The estimates rest on a mixed bottom-up and top-down proprietary model, a per-country spreadsheet of modelling assumptions. gsma.com
- Francis Galton, “Vox Populi,” Nature 75 (1907), as re-analysed in Kenneth F. Wallis, “Revisiting Francis Galton’s Forecasting Competition” (787 estimates; median 1207 lb, mean 1197 lb; actual dressed weight 1198 lb). Galton’s original page is an image-only scan, so the figures are cited from Wallis’s machine-readable re-analysis. hummedia.manchester.ac.uk
- Scott E. Page, “Making the Difference: Applying a Logic of Diversity,” Academy of Management Perspectives (2007). The diversity prediction theorem: collective error equals average individual error minus predictive diversity. engineering.pitt.edu
- World Bank, “The Global Findex Database 2021: Survey Methodology.” Individual responses from about 128,000 adults, drawn by probability-based, nationally representative samples, with the microdata released publicly so the number can be reconstructed. thedocs.worldbank.org