The Collateral Gap — Part OnePart One of Four

Start with the fact that the continent's own development bank is not debating whether this gap exists. The African Development Bank's Affirmative Finance Action for Women in Africa exists specifically to close a documented $42 billion financing gap facing women entrepreneurs, and has approved over $1 billion in lending toward it to date. Its central design feature is reducing collateral requirements. That is not incidental; it is a diagnosis embedded in a product.

The market-by-market deployments are recent and dated. A $61 million AfDB facility approved for the Development Bank of Nigeria in May 2026, with more than 95% earmarked for women-owned MSMEs. A $332 million AfDB investment into Standard Bank Group in South Africa in August 2026, paired with a $1 million technical-assistance grant whose explicit purpose is helping women build verifiable credit histories through digital payment tools. Read that grant line carefully: a development bank is separately funding the creation of the evidence base, because the evidence base does not exist.

Beneath the continental figure sit larger structural numbers. The IFC estimates a $5.2 trillion global financing gap for formal MSMEs and $2.9 trillion for informal ones. East Africa alone carries an estimated $65 billion financing gap for agricultural SMEs in the missing middle — businesses too large for microfinance and too small, or too collateral-poor, for commercial bank credit.

The $42 billion is therefore not a standalone number. It is the gendered slice of a much larger structural gap, and it is the slice with the clearest single causal mechanism attached to it.

World Bank research finds that just under 13% of African women aged 20–49 claim sole ownership of land, compared with 36% of men. Counting joint ownership, only 38% of women report owning any land at all, against 51% of men. In Nigeria the gap is sharper: 8.2% of women hold sole title against 34.2% of men, driven partly by inheritance customs that in practice still favour sons over daughters across large parts of the country.

The World Bank's own analysis is explicit about why this matters for credit specifically: gender gaps in land and property ownership reinforce broader inequalities precisely because of land's dual function as collateral. The property gap and the credit gap are the same gap, observed at two points in the chain.

In parts of Southern Africa the mechanism was legally explicit rather than merely customary. Under historical marital-power provisions, married women could not apply for credit in their own right without a husband's written consent to borrow, mortgage or sell property. Those provisions have been repealed; the balance-sheet consequences of three decades of them have not.

This is the finding that reframes the whole sector. It is not a lending industry accidentally overlooking women. It is a credit system built around an asset class women were structurally prevented from holding, and then scoring — accurately, on its own terms — on what it can see.

Here is the finding that should complicate any comfortable version of this story. The AfDB's 2025 Trade Finance Report found that many African SMEs continued repaying loans consistently through periods of currency volatility and tightening global credit — suggesting the sector's real risk profile is more stable than lenders have priced it.

Perceived risk, not demonstrated repayment behaviour, has driven the stricter interest rates, shorter tenors and tougher collateral demands documented across the continent's formal trade finance system. That is a pricing error, not a prudence decision, and pricing errors of that size are ordinarily arbitraged away. They have not been here, because the repayment behaviour is real but unverifiable at the point of underwriting.

That gap between perceived and actual risk is exactly where alternative-data scoring is pitched as the fix. This publication has already found reason for caution. A ten-algorithm audit of African fintech credit-scoring systems found women-led SMEs facing an estimated 37% underfunding penalty from proxy variables and biased training data.

Two findings that do not yet agree: demonstrated repayment says the risk is overstated; algorithmic outcomes say the new tools are re-encoding the old bias in a faster wrapper. The old system overpriced women's risk because it could not see their real repayment behaviour. The new one, built carelessly, finds a more efficient way to overprice it anyway.

The standard remedy is to substitute behavioural data for property: mobile money velocity, till receipts, supplier ledgers, utility and airtime payment regularity. The logic is sound and the data genuinely exists at scale across East and West Africa. The failure is in what happens next.

First, most alternative-data models are trained on portfolios whose historical approvals were themselves collateral-gated. A model learning from twenty years of decisions in which women were structurally under-approved will reproduce that distribution and call it a prediction. This is the mechanism behind the 37% penalty finding — not malice, and not even bad engineering, but a training set that encodes the constraint as if it were the outcome.

Second, alternative data is currently non-portable. A borrower's twenty-four-month mobile-money repayment record sits inside one operator's or one lender's system. Move to another institution and the record does not travel; the applicant re-enters as a thin file. The behavioural evidence is generated, observed, and then discarded at the institutional boundary — exactly the memory failure this series has documented elsewhere.

Third, none of it is legally usable as security. A strong behavioural score improves the probability of approval; it does not give the lender a recoverable claim if the loan sours. Until movable-asset registries, warehouse receipts, invoice assignment and equipment liens are digitised and enforceable, the lender's downside is still uncollateralised, and the price still reflects that.

Treat the $42 billion as the stock of unmet demand rather than the revenue opportunity, and the layer becomes measurable. If a verified, portable non-property collateral rail unlocks 12–22% of that stock over five years, that is $5.0BN–$9.2BN of incremental credit originated — lending that is currently declined at the collateral test, not at the affordability test.

The originator economics follow from spread and volume. At blended net interest margins of 6–9% typical of African SME lending, incremental annual interest income across participating lenders sits at $300M–$830M once the book seasons. Against that, the infrastructure operator prices for a slice: an origination and verification fee of 35–80 basis points on facilitated volume produces $17M–$74M of annual platform revenue at scale — a substantial business built entirely from friction that lenders currently absorb as declined applications.

The cost side is where the case sharpens. Manual collateral assessment, valuation and registry search in these markets runs $180–$650 per facility, plus 9–26 days of elapsed time. Digitised movable-asset verification with a portable behavioural file compresses that to $40–$120 and 2–5 days. Across 350,000–900,000 facilities a year in the addressable band, that is $60M–$400M of annual origination cost removed from the system.

The second-order pool is loss reduction. Portable repayment history reduces adverse selection: lenders currently price the whole thin-file cohort at the risk of its worst quartile because they cannot separate it. Even a 90–180 basis point reduction in blended expected loss across the segment, on a $5BN–$9BN incremental book, returns $45M–$165M annually. The composite is a $120M–$640M annual value pool, against a build cost in the low tens of millions.

Year 1 is a single-market proof, most plausibly Kenya, where mobile-money penetration, a functioning movable-property security registry and a concentrated lender population coincide. Three to five lender partners, 8,000–20,000 facilities scored against a portable file, and the deliverable is a published default-comparison between collateral-gated and behaviour-verified cohorts. Revenue $0.6M–$1.5M against a $3M–$5M build.

Years 2 and 3 extend to Nigeria and Ghana, where the registry infrastructure is weaker and the partnership burden shifts from banks to regulators. Volume reaches 90,000–250,000 facilities a year; the corpus becomes the moat, because eighteen months of matched repayment outcomes across gender, sector and geography is not purchasable. Revenue $6M–$16M at gross margins above 65%, with the bias-audit function priced separately to DFI and guarantee-fund buyers.

Years 4 and 5 monetise the rail rather than the score. Guarantee funds, AFAWA-style facilities and commercial lenders underwrite directly off the portable file; the operator earns on verification volume, portfolio analytics and risk-sharing structuring. Revenue $22M–$55M, with cumulative additional credit facilitated of $4BN–$11BN across the period.

The governing sensitivity is guarantee availability, not technology adoption. Every dollar of first-loss cover from a development finance institution converts roughly seven to twelve dollars of commercial lending in this segment. Without it, the layer proves the risk is mispriced and still cannot get the first cohort funded.

A scoring rail aimed at a historically excluded cohort must be auditable by construction, not by promise. Three requirements are non-negotiable. Publish approval and default rates disaggregated by gender and market. Prohibit property ownership and its close proxies — registered business premises, formal payslip history, years since incorporation — from carrying decisive weight. And run continuous counterfactual testing: would this file have been approved with an identical behavioural record and a different gender marker?

Three things would break the thesis. First, a state-mandated national credit registry with behavioural data included, collapsing the addressable market to integration work — a real possibility in Kenya and Ghana within the window. Second, guarantee capital contracting as DFI budgets tighten, removing the mechanism that converts proven risk into funded risk. Third, and most likely, the layer succeeds commercially while the bias problem is quietly left unaudited, at which point it is simply a faster version of the system it replaced.

The honest reading of Part One is therefore conditional rather than promotional. The gap is documented, the mechanism is named, the repayment behaviour suggests the risk is overpriced, and the current algorithmic fixes have measurable bias in them. Whether that resolves into an infrastructure opportunity or an automated repetition of the same exclusion depends on decisions being taken now, by a small number of identifiable institutions.

Part Two — 'The Audit: Scoring Five Markets, Not Averaging Them' — widens the frame to the UK as a central claim rather than an afterthought: the collateral gap is one problem with two regional expressions. The Rose Review quantified the UK's version at over £250 billion in lost economic value, running on a parallel mechanism — women-led businesses statistically less likely to hold formal payslip history or registered years-in-business in their own name. Kenya, Nigeria, Ghana, South Africa and the UK are audited individually against the same four-part scoring test. Never averaged.