Informal Proof Is Still Proof

Read the discount correctly. When an investor marks down your WhatsApp order book, they are pricing the cost of checking it, and you pay that price whether or not you earned it.

The evidence exists. Ninety days of mobile money settlements. Delivery notes signed at the gate. A local purchase order from a distributor whose name the reader already knows. What the evidence lacks is an issuer, and a way for a stranger to confirm it without phoning you.

That is an engineering problem. It has a cost, a sequence and a delivery date.

Why these three models

This piece runs the Wire Model: score the features of the decision, route to a small ensemble of formal models, then force the ensemble to produce dated actions. The scores that mattered:

• Asymmetric information with costly verification (0.9). The reader cannot separate a real revenue claim from a staged one without spending time they will not spend.
• Falsification asymmetry (0.8). Some artifacts require only you to lie. Others require a telecom operator, a bank or a revenue authority to lie with you.
• Independent-source availability (0.75). Most African transactions now touch at least two parties who keep their own records.
• Regime-break risk (0.8). Mandatory electronic invoicing and instant payment rails are converting informal trade into machine-readable records on a government timetable.
• Cognitive distortion (0.45). Real, and scored too low to earn a card of its own.

That routes to adverse selection and certification (the discount), independent-source aggregation (the stack), and regime-switching (the rails). Against the outcome-type map those are equilibrium, random and cycle-regime: three distinct types, so the errors point in different directions and partly cancel.

Behavioral and governance layers are folded rather than shipped as cards. The behavioral effect here is an effort discount, and the fix for an effort discount is identical to the fix for an information discount, which is to lower the cost of checking. Folding it costs the reader no lever. The governance layer sits in the blind spot, where it belongs.

The framework: the proof is real, the address is missing

1. The discount: you are priced as the average of everyone who claimed what you claimed

Akerlof’s result is the base case and it is unsentimental. When a buyer cannot observe quality before purchase, they pay a price set by the average quality on offer. Sellers holding above-average goods withdraw, which lowers the average, which lowers the price again. His paper closes on what he called counteracting institutions: guarantees, brand names, licensing, chains. Institutions that let a good seller separate.¹

Run that on a Series A pipeline. A partner in Nairobi has seen twelve decks this quarter claiming forty to eighty thousand dollars a month in revenue, each supported by a screenshot. Three are real. The partner cannot tell which three inside the twenty minutes they will spend, so they price all twelve as one. The discount is the market price of checking, charged to everyone in the queue regardless of who is telling the truth.

Spence gives the escape route. A signal separates types only when it costs the weak sender more than the strong one.² Translate that into a test you can run on every document in your folder: who else would have to lie for this to be false?

• A WhatsApp screenshot: only you.
• A spreadsheet of orders: only you.
• An M-Pesa or MTN MoMo statement pulled from the operator: the operator.
• A signed LPO from a listed distributor: that distributor’s procurement officer.
• A fiscal receipt carrying a revenue authority’s verification code: the revenue authority.

Rank your evidence by that count. How impressive a document looks is not on the scale.

Now the part founders miss. The informal record carries strong information in a wrapper nobody has addressed. Studying more than 250,000 observations, Berg, Burg, Gombović and Puri found that simple digital footprint variables, the residue people leave just by using a website, equal or exceed the information content of a credit bureau score, and work on customers the bureau cannot score at all.³ Björkegren and Grissen matched mobile phone records to repayment outcomes for credit extended by a South American telecom and found that borrowers in the highest risk quintile were 2.8 times more likely to default than those in the lowest, with their method outperforming bureau models on thin-file individuals.⁴

Lenders already underwrite people who look exactly like your customers, using exhaust that looks exactly like your evidence. The information problem is solved. The addressing problem is not.

Adverse selection and certification, the discount lens

Assumes: the reader cannot verify cheaply, and knows it, so they pool you with your claim class.

Fits because: asymmetric information with costly verification scored 0.9 and falsification asymmetry scored 0.8.

Breaks when: the reader has private information about you already, from a prior deal or a shared customer. Then they are pricing you directly, and better documents change little.

Counteracts: reading rejection as a judgment of the business.

May reinforce: document theatre, and the belief that thicker folders beat better issuers.

2. The stack: one artifact is a claim, four independent artifacts are a fact

Aggregation works because errors from independent sources partly cancel. The active ingredient is independence. Volume on its own does nothing, and ten screenshots from one phone is one source photographed ten times.

The cleanest demonstration of the principle is a customs method. Fisman and Wei compared what China reported importing from Hong Kong against what Hong Kong reported exporting to China. Two independent records of the same shipments. The gap between them, which they called the evasion gap, tracked tariff and value-added tax rates so tightly that a one percent rise in the tax rate produced roughly a three percent rise in evasion.⁵ Nobody confessed. The truth fell out of the reconciliation.

Build that on yourself. Take one month, not twelve. Assemble five records of the same revenue number:

1. Your own order log or till book. Issuer: you.
2. The settlement statement pulled from the mobile money operator or the merchant acquirer. Issuer: the operator.
3. Delivery notes with the buyer’s stamp and the date received. Issuer: the buyer.
4. A written confirmation of volumes from your two largest buyers, with a phone number the reader may call. Issuer: the buyer.
5. The fiscal invoice or e-receipt for the same period. Issuer: the state.

Five records. Four issuers. One number. Then write a single page that ties them together and states the variance.

A reconciliation that ties to the shilling reads as staged. One that ties to 97 percent and explains the other 3 percent reads as a business. Variance is the tell that a human ran the numbers against reality rather than backwards from a target.

Finance has an institutional version of the same move. Klapper’s work on factoring shows why it reaches suppliers ordinary lending will not touch: credit is linked to the value of the receivable and the credit quality of the buyer, not the overall creditworthiness of the supplier. Reverse factoring programmes, her example being Nafin in Mexico, exist specifically to work around a small supplier’s informational opacity.⁶ A signed LPO from a large buyer is a promise about them, which you happen to hold.

Stop asking readers to trust you. Give them four parties who have something to lose.

Independent-source aggregation, the corroboration lens

Assumes: sources are genuinely independent, and each keeps its own record for its own reasons.

Fits because: independent-source availability scored 0.75, and every transaction on a payment rail leaves at least two records.

Breaks when: the sources are correlated. Your buyer’s confirmation and your delivery note are one source if your own staff wrote both.

Counteracts: volume as a substitute for verifiability.

May reinforce: reconciliation as a hobby, and months spent tidying records that no reader ever asked for.

3. The rails: your proof state is changing whether or not you act

The structure this argument sits on is moving, and moving in your favour.

Kenya now requires all persons carrying on business, explicitly including the informal sector and small businesses, to generate and transmit invoices electronically to the revenue authority. The simplified route runs over USSD on *222# and a web platform on eCitizen, and there is a reverse-invoicing option under which a buyer issues the invoice on the seller’s behalf with consent.⁷ Uganda’s EFRIS transmits transaction details to the revenue authority in real time and returns a fiscal document number, a verification code and a QR code on every receipt.⁸

Read what that actually is. A government, for its own reasons, has issued you a free, timestamped, third-party revenue record that a stranger can verify from a phone camera. Founders treat enrolment as a tax event. It is also the cheapest certification instrument available to you.

The scale underneath is large. The International Labour Organization counts two billion workers, 61.2 percent of global employment, in informal employment.⁹ The Global Findex reports 79 percent of adults worldwide now hold an account, and 84 percent of adults in low- and middle-income economies own a mobile phone.¹⁰ An enormous informal base is now instrumented. The records are being generated already. Almost nobody is collecting them.

The capital that reads this format is the capital that is growing. In 2025 African tech ventures raised just over US$4 billion, of which debt was US$1.64 billion, up 63 percent year on year across 108 transactions, a 40 percent rise in deal count. The number of unique debt investors rose 10 percent while unique equity investors fell 7 percent.¹¹ Debt underwriters read settlement files, receivables and repayment histories. They already speak the language your business writes in.

The day you enrol on a fiscal rail, your pitch numbers and your filed numbers become the same document. Decide now which one was true.

Regime-switching, the rails lens

Assumes: the rules governing what counts as a record are shifting on a published timetable you can read.

Fits because: regime-break risk scored 0.8, with mandates already in force in Kenya and Uganda.

Breaks when: enforcement lags the gazette. A mandate nobody polices produces a record nobody trusts, and you have bought a liability without buying a certificate.

Counteracts: planning your evidence around the market you last raised in.

May reinforce: compliance as an identity, and the assumption that a state-issued record is automatically a persuasive one.

GEER: the levers, cheapest first

Four channels carry the discount: verification cost, source independence, issuer authority, reconcilability. Pull the cheap and reversible ones first.

1. Export, do not screenshot. Pull the operator statement or the acquirer report in its own format, with its own header. Hits verification cost. Costs an afternoon.
2. Label every artifact with its issuer. Who made this record, on what date, and how does a stranger reach them. Hits issuer authority. Costs an hour.
3. Write the one-page reconciliation. One number, every source, the variance, the reason for the variance. Hits reconcilability.
4. Get two buyers to confirm volumes in writing, with a named person and a working number. Hits source independence. Costs a week of asking.
5. Enrol on the fiscal rail and start the clock. Hits issuer authority at the top of the ladder. Costs a day, plus the tax position you should have taken anyway.
6. Move one revenue line onto a rail that timestamps it. A till number, a merchant account, a payment link, an invoice carrying a fiscal code, instead of a bank transfer with a first name in the reference field. Hits everything. Costs months, because the record accrues forward and cannot be backfilled.

No-lever flag: if every artifact you hold was issued by you, no formatting fixes it. You are in a source-acquisition state, and that state costs calendar time rather than effort. Say that to yourself in writing and re-plan the quarter around it.

RADAR: the portfolio, dated

DO NOW, by T+3 days. Reversible, and dominant across every scenario.

1. Pick the most recent complete month. One month, not a year.
2. List every artifact that touches that month’s revenue. Beside each, write the issuer and a yes or no: can a stranger verify this without contacting me.
3. Export the operator and bank statements yourself, in their native format, unedited.
4. Write the one-page reconciliation, variance included, and hand it to someone who has never seen your business.

HEDGE, by T+14. Cheap insurance against a long verification cycle.

1. Enrol on the fiscal invoicing rail in your market, so the next quarter accrues with a verification code attached.
2. Collect two written buyer confirmations naming volumes and dates.
3. Open a dedicated merchant or till account, so revenue and personal float stop sharing a ledger.

DEFER AND TRIGGER. Irreversible or expensive, so wait, and pre-commit the trigger now.

1. Defer: a statutory audit, an ERP migration, a full data room.
2. Trigger to spend: two separate readers ask the same verification question, or a lender requests six months of statements. That is demand confirmed. Commission the accountant’s review that week.
3. Counter-trigger: by T+28, if no third party can issue any record of your revenue, stop raising. The gap is in the payment rail, not the deck. Fix the rail first.

If you are the one reading. DO NOW: publish the evidence you accept, ranked by issuer rather than by format. HEDGE: accept operator statements pulled under founder authorisation, and price the difference between self-issued and third-party issued explicitly. DEFER: changing your underwriting model until you have measured failure rates by evidence type across two cohorts.

CHAIN: what usually happens next

Match the reference class on structure. Thin-file consumer lending, receivables factoring and customs mirror statistics all run the same machinery as investor diligence: the subject’s own claim is discounted, and a record issued by someone else clears it.^{4, 6, 5} The base rate across all three is consistent. Where an independent issuer exists, capital flows. Where it does not, it does not, regardless of the underlying quality.

Present-state modifiers push the same way. Mandated e-invoicing is creating issuers where none existed.^{7, 8} Debt capital, which reads settlement data natively, is the fastest-growing pool on the continent.¹¹

Subtract the counterfactual before you credit legibility with everything. Part of the informal discount is real risk rather than poor addressing. A business that cannot produce a third-party record often cannot because the revenue is lumpy, partly cash, and partly already spent. Fixing the paperwork does not fix that, and readers who have been burned know the difference.

Matrix-break flag. Fabricating a convincing PDF statement now costs minutes. Documents are dying as evidence, including good ones, which means the honest founder loses the cheap channel first. What survives is a record the reader pulls themselves: a read-only bank link, an operator API, a QR code that resolves on a revenue authority’s server. Short run, third-party documents still work. Medium run, only pull-based verification does. Build for the medium run.

What this ensemble cannot see

Four things, and they are large.

Cash. A meaningful share of real African trade leaves no record with anyone. These models cannot price what nobody logged, and the businesses least able to produce an issuer are often the ones most deserving of capital.

Whether the discount is legibility or taste. Some readers will not fund your category, your city or your stage at any resolution. Perfect evidence does not move them, and they will cite evidence anyway because it is the polite reason.

How much any single read is worth. Early-stage picking sits close to the luck end of the luck-to-skill continuum. Treat one rejection as one draw from a noisy distribution.

Survivorship. Every study cited here counts businesses that produced enough of a record to enter a dataset. The ones that never did leave no trace, and they are the majority.

So act on the part that is engineering and price the rest as noise. Pick last month. One revenue number, five records, four issuers. Reconcile it, publish the variance, enrol on the fiscal rail by T+14. Then the next rejection tells you something, because you will have removed the only reason for it you controlled.

Sources and notes

1. Akerlof, G. A. “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism.” Quarterly Journal of Economics 84(3), 1970, 488-500. The published article is behind the publisher’s paywall; bibliographic record and section structure, including section IV on counteracting institutions, at EconPapers. The model is set out in the Nobel Committee’s scientific background paper, Markets with Asymmetric Information (2001).
2. Spence, M. “Job Market Signaling.” Quarterly Journal of Economics 87(3), 1973, 355-374. Full text.
3. Berg, T., Burg, V., Gombović, A., and Puri, M. “On the Rise of FinTechs: Credit Scoring Using Digital Footprints.” NBER Working Paper 24551, 2018; Review of Financial Studies 33(7), 2020. Over 250,000 observations; digital footprint variables equal or exceed credit bureau score information content, including for unscorable customers. Abstract and working paper.
4. Björkegren, D., and Grissen, D. “Behavior Revealed in Mobile Phone Usage Predicts Credit Repayment.” World Bank Policy Research Working Paper 9074, 2019. Call records matched to repayment outcomes for credit extended by a South American telecom; highest-risk quintile 2.8 times more likely to default than the lowest, and the method outperforms credit bureau models on thin-file individuals. Full text PDF; earlier version as arXiv preprint 1712.05840.
5. Fisman, R., and Wei, S. “Tax Rates and Tax Evasion: Evidence from ‘Missing Imports’ in China.” NBER Working Paper 8551, 2001; Journal of Political Economy 112(2), 2004. The evasion gap compares China’s reported imports from Hong Kong with Hong Kong’s reported exports to China; a one percent rise in the tax rate is associated with roughly a three percent rise in evasion. Abstract and working paper.
6. Klapper, L. “The Role of Factoring for Financing Small and Medium Enterprises.” World Bank Policy Research Working Paper 3593, 2005. Credit linked to the value of receivables and the buyer’s credit quality rather than the supplier’s; reverse factoring and the Nafin programme in Mexico. Full text PDF and document record.
7. Kenya Revenue Authority, “Simplified eTIMS Solutions for Informal Sector and Small Businesses.” All persons carrying on business, including the informal sector and small businesses, must generate and transmit invoices electronically; eTIMS Lite runs on USSD *222# and on eCitizen, with a reverse-invoicing option under which the buyer issues on the seller’s behalf. Press release. See also the public notice on electronic tax invoicing for non-VAT registered persons.
8. Uganda Revenue Authority, Electronic Fiscal Receipting and Invoicing Solution (EFRIS). Transaction details are transmitted to URA in real time, generating e-receipts and e-invoices carrying a fiscal document number, verification code and QR code; mandatory for VAT-registered taxpayers under section 73A of the Tax Procedures Code Act 2014. Official page.
9. International Labour Organization, “Women and Men in the Informal Economy: A Statistical Picture” (third edition, 2018). Two billion workers, 61.2 per cent of the world’s employed population, are in informal employment. Publication record.
10. World Bank, Global Findex 2025. 79 per cent of adults globally hold an account; 84 per cent of adults in low- and middle-income economies own a mobile phone. Report page.
11. Partech, 2025 Africa Tech Venture Capital Report. Just over US$4B raised in 2025; debt US$1.64B, up 63 per cent year on year across 108 transactions, a 40 per cent rise in deal count; unique debt investors up 10 per cent while unique equity investors fell 7 per cent. Report page.