Size the Market From the Bottom, One Verifiable Row at a Time

You do not have a reliable market number, and no report will hand you one. So stop borrowing a top-down figure and build the number from the bottom, one count a stranger can independently verify. The built number lands smaller than the borrowed one, and it survives the meeting the borrowed one dies in.

Here is the decision this piece is for: which four public counts you anchor your market slide on this week, and which figure you delete before the call.

A top-down slide reads well and checks badly. It arrives as one big claim: the market is worth some billions, and you own a wedge of it. The investor cannot rebuild that claim from your inputs, so they discount it toward zero and move on to the number they can test. A bottom-up build starts from counts that already exist in public: registered mobile money agents, licensed pharmacies, tax-filing enterprises, active SIM connections. Each count is a row. Each row has a source a reader can open. The number is not missing. You have not built it yet.

Why these three models

Market sizing under bad data is a disclosure problem before it is an arithmetic problem. Three lenses carry it, and they fail in different places, which is the point of running them together rather than trusting one.

The first asks what a claim is worth when the reader cannot check it. The second asks how to build a robust number out of several weak ones. The third asks which method actually survives the room, judged on the record rather than on how the slide looks. They span three outcome types: an equilibrium of who reveals what, a random-error argument about combining independent counts, and a reference-class bet about what usually clears diligence. No two of them make the same mistake.

1. The disclosure: why a checkable number beats an impressive one

Verifiable disclosure is the economics of provable claims. Paul Milgrom’s result is the anchor: when a sender can prove any true statement, silence is read as the worst case, and the good types keep separating themselves by disclosing until the market correctly infers the rest.¹⁰ Investors run this on your TAM without naming it. A number they cannot verify is treated as a number you would not verify, which is treated as the low end.

The move that beats the discount is to disclose in a form the reader can reconstruct. A bottom-up build is exactly that. You show the count, the source, the multiplier, and the arithmetic. The investor can verify it in one government page instead of taking your word for it. Verification is the tax every investor was always going to charge. Build the number they can check, and you have already paid it.

Assumes: the reader can, in principle, check your inputs, and reads unverifiable claims as concealment.

Fits because: public registries make your rows provable, so disclosure separates you from the founder who cannot show the work.

Breaks when: the binding input is genuinely unmeasured (an informal segment no registry counts), so there is nothing to disclose and silence is honest.

Counteracts: the reflex to inflate, because an inflated row invites the check that exposes it.

May reinforce: under-claiming, if you only count what is easy to count and leave real market on the table.

2. The triangulation: three weak counts beat one strong report

Independent-source aggregation is the reason a built number is sturdy. Combine several estimates whose errors are independent, and the errors partly cancel; the average is tighter than any single input. One consultancy report is a single draw with an unknown bias. Three public counts from different institutions, measuring the market from different angles, are three draws whose mistakes do not point the same way.

Take Kenya. The number of mobile money agents is a real, published count: the agent network grew 26.8% to 501,399 in the 2025 figures, against 51.4 million mobile subscriptions.³ That count sits in the Central Bank of Kenya’s monthly payments statistics, which anyone can pull.⁴ Now cross it against a second, independent source: the Global Findex, which puts Sub-Saharan African account ownership at 55% in 2021, up from 23% a decade earlier, with all eleven of the world’s mobile-money-majority economies sitting in the region.² And a third: the GSMA, which counts over two billion registered mobile money accounts globally, more than half of them driven by Sub-Saharan Africa.^{1, 2} Three institutions. Three methods. If a founder’s agent-led market size is consistent with all three, it is hard to wave away.

Assumes: your counts come from genuinely independent producers, so their errors are uncorrelated.

Fits because: a regulator’s agent register, a household survey, and an industry census measure the same market through different instruments.

Breaks when: the sources are not independent (three reports all reselling one government feed), so the errors are the same error three times.

Counteracts: single-source risk, the quiet failure of leaning a whole slide on one PDF.

May reinforce: false confidence, if three correlated sources agree and you read agreement as accuracy.

3. The base rate: which method actually clears the room

Reference-class thinking judges the two methods on their record, not their polish. Line up the memos that reached a term sheet against the ones that stalled at the market slide, and ask which sizing method sat in each. The borrowed top-down number has a poor record for one structural reason: in this region even the denominator moves.

When Nigeria rebased its GDP in 2014, the estimate jumped 89% overnight, from 42.4 trillion naira to 80.2 trillion, on nothing but a change of base year from 1990 to 2010.⁷ The African Development Bank recorded the dollar figure moving from 285.56 billion to 509.9 billion, an 89.2% increase, and the country going from Africa’s second economy to its largest in a single announcement.⁶ Ghana had done the same in 2010, revising GDP up 60% and crossing from low to middle income without producing a shilling more of output.⁸ Morten Jerven’s audit of African statistics found the problem is not one bad year: rank African economies by income across the three most cited datasets and roughly a fifth of them jump more than ten places depending on which source you pick.⁹

If the denominator can move 89% in a single day, any market you sized as a slice of it inherits that swing. A diligence analyst knows this. That is why the top-down slide loses the reference-class bet, and why a build from stable, source-level rows wins it.

Assumes: the past pattern of what survives diligence still holds for your raise.

Fits because: the instability of African macro aggregates is documented and repeated, so the base rate against top-down sizing is earned, not guessed.

Breaks when: your investor is a generalist who wants the big headline number and treats a small built figure as a small opportunity.

Counteracts: anchoring on the first impressive figure you were shown.

May reinforce: excessive caution, if the base rate talks you out of a market that is real but young.

Here is the whole method in one worked build, the kind you can finish this afternoon. Say you sell software to pharmacies in Kenya. Start with the regulator’s count of licensed outlets, a public number that already exists. Multiply by the share you can realistically reach in your launch counties, a fraction you defend with your own pipeline rather than a hope. Multiply by your annual price. That product is your serviceable revenue, and every term in it points either at a named source or at your own data. Then add a second row from an independent count, agents or registered enterprises, and check that the order of magnitude agrees. Your slide is now four lines of arithmetic a reader can redo on the back of a napkin, not a large rectangle they have to accept on faith. It will read smaller than the borrowed figure. It will also still be standing at the end of the meeting, which the borrowed figure will not.

The levers, ordered by how little they cost you

Cheapest and most reversible first. None of this needs a data budget.

• Pull one count today. Open the registry closest to your revenue: mobile money agents for an agent business, the SMEDAN and National Bureau of Statistics MSME survey for a B2B tool (it counted 39,654,385 enterprises in Nigeria in 2021), the pharmacy or transport regulator for a vertical.⁵ Write your arithmetic chain directly under it. This is an hour of work and it replaces your weakest slide.
• Add a second, independent count. Not a second article about the first count. A different institution measuring the market a different way. If your build survives both, you have triangulation, not a guess.
• State the vintage and the one sensitivity that matters. Name the year of each row and the single number that would move your total most. A sized market with its own error bar reads as competence.
• Delete the borrowed figure. The consultancy billions add nothing a diligence analyst will keep. Cut it, or demote it to a one-line magnitude check clearly labelled as a check.

What to build before the round opens, and what to hold

Sequence the work so the reversible, dominant moves happen now and the expensive, irreversible ones wait for a trigger.

• Do now (reversible, wins in every scenario). By T+3, replace the headline TAM with a bottom-up build from three or four verifiable rows, each with a live source link, and publish the arithmetic in the memo. This helps you whether the market turns out large or small, because it is the version that survives a check.
• Hedge (cheap insurance against looking naive). By T+7, keep one top-down cross-check in a footnote, labelled as a sanity check on order of magnitude, never as the number. If your build and a rough top-down land within the same power of ten, say so. If they diverge by an order of magnitude, say that too, and explain why the build is the honest one.
• Defer and trigger (irreversible, so pre-commit the signal). Do not commission paid primary research or buy a private dataset yet. Write the trigger down instead: the moment an investor asks for segment depth your public rows cannot reach, and only then. By T+28, if that ask has not come, you have saved the money and lost nothing.

What usually happens next

Match this to the right reference class. The comparison is not other African startups in general; it is founders who walked into diligence with a market slide, in a market with thin official data. In that class, the pretty top-down slide is the common case and it reliably underperforms: it gets discounted, then quietly ignored, and the sizing conversation restarts from the investor’s own back-of-envelope.

Your present state changes the odds in your favour on two counts. Your market has public registries a reader can open, which most sizing debates lack. And you hold your own transaction data, which lets you turn a public count into a revenue estimate with a defensible attach rate. Subtract the counterfactual honestly: a polished top-down slide would not have won you the term sheet, so building the bottom-up version costs you a morning and loses you nothing you had.

One flag. If a regulator stops publishing the registry you leaned on, or a rebasing lands in the middle of your raise, the rules under your number change and you re-anchor on the rows that did not move. Watch for it. It is rare, but it rewrites the arithmetic when it happens.

Where this method goes blind

The ensemble can only see what someone has counted. It is 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. It cannot vouch for the quality of a registry, and a stale or padded government count will pass straight into your build wearing the authority of its source. It does not know your true attach rate; that number is yours to defend, and the public row only sets the ceiling.

So do not pretend the built number is the whole market. Name the uncounted segment as a gap and size it separately, or say plainly that you cannot size it yet. That admission runs on the same disclosure logic that made the rest of the number credible, and a diligence analyst reads it as competence rather than doubt.

Here is the action that survives the ignorance: pick your four rows this week and write the arithmetic beneath each one. If you can only verify three, ship three and label the fourth as the gap you are closing. Three rows a stranger can check will out-argue a billion-dollar figure nobody can, in every room that matters.

Sources and notes

Ensemble members are named in prose and appear in the Wire Model outcome-type map: verifiable disclosure and unraveling (equilibrium), independent-source aggregation (random), base rates and reference class (cycle-regime). The behavioral layer (anchoring on the first big number) is folded into model 3 rather than shipped as a fourth card, because it adds no lever the reference-class read does not already supply.

1. GSMA, “Mobile Money Surpasses Two Billion Registered Accounts and Over Half a Billion Monthly Active Users Globally,” press release on the State of the Industry Report on Mobile Money 2025. gsma.com
2. CGAP, “Findex 2021 Insights: Boosting Financial Inclusion in Africa” (Sub-Saharan African account ownership 23% in 2011 to 55% in 2021; all eleven mobile-money-majority economies in the region), drawing on the World Bank Global Findex Database 2021. cgap.org
3. The Kenya Times, reporting the Kenya National Bureau of Statistics Economic Survey 2026: mobile money agents grew 26.8% to 501,399; subscriptions rose 21.4% to 51.4 million in 2025. thekenyatimes.com
4. Central Bank of Kenya, National Payments System, mobile payments statistics (the public monthly registry a founder can pull directly). centralbank.go.ke
5. SMEDAN and the National Bureau of Statistics, National Survey of Micro, Small and Medium Enterprises 2021: 39,654,385 MSMEs operating in Nigeria as of December 2021. Figure carried at moniepoint.com/blog/nigeria-small-business-statistics
6. African Development Bank, “Nigeria becomes largest economy in Africa with $509.9 billion GDP” (up from $285.56 billion, an 89.2% increase, on rebasing from 1990 to 2010 constant prices), 10 April 2014. afdb.org
7. “2014 Nigeria GDP rebasing” (GDP estimate rose 89%, from 42.4 trillion to 80.2 trillion naira, on a base-year change from 1990 to 2010), citing The Economist, “Step change,” 12 April 2014. en.wikipedia.org
8. “2010 Ghana GDP rebasing” (60% upward revision on a base-year change from 1993 to 2006), citing Jerven and Duncan. en.wikipedia.org
9. Gerardo Serra, review of Morten Jerven, “Poor Numbers: How We Are Misled by African Development Statistics and What to Do About It,” Africa at LSE (roughly a fifth of African economies jump more than ten places in income rank depending on the source used). blogs.lse.ac.uk
10. Paul Milgrom, “Good News and Bad News: Representation Theorems and Applications,” Bell Journal of Economics 12 (1981), on verifiable disclosure and unraveling. Author copy: milgrom.people.stanford.edu. Cited for the theorem, not a figure.