This Strange Machinery: How African Banking Cannot See Its Own Clients

African banking systems frequently fail to serve the majority of the continent's population because they were designed around formal employment records, fixed addresses, and credit histories that most Africans simply do not have. Transplanted from Western financial contexts, these frameworks function exactly as intended, yet their core assumptions render hundreds of millions of potential customers effectively invisible to the institutions meant to serve them.

Ride’s “Chrome Waves” is a 1992 B-side that most people will not know, and one of its lines has lodged itself in my thinking about African banking: “this strange machinery, is keeping you from seeing me.” The lyric was written about something entirely personal, a distance between two people that the ordinary machinery of life was somehow perpetuating. That is not what this piece is about. But the image transfers more cleanly than it has any right to, because the machinery this piece is about also runs correctly, also produces outputs that look reasonable to the people who built it, and in doing so fails systematically to see the majority of the people it was built to serve, not through malice but because it was engineered for a different world and then transplanted wholesale into this one.

1. How the Lending Machine Decides

When you apply for credit at a South African bank, the lender does not simply look at your bank balance but queries one or more credit bureaus, organisations that hold a consolidated record of every credit facility you have ever opened, how consistently you have repaid each one, whether you have ever defaulted or been handed over for collection, how many times you have applied for new credit in the recent past, and how much of your available credit you are currently using. From that history the bureau produces a score, a single compressed number representing its model’s prediction of how likely you are to repay the new facility being considered. The lender combines that score with an affordability assessment, comparing your declared income against your known expenses and existing debt obligations, and from those two inputs a credit decision emerges.

The model works reasonably well for the population it was designed to serve: if you have held a credit card for five years, maintained a vehicle finance agreement and paid your store accounts on time, the bureau has enough signal to produce a reliable prediction about your future behaviour, and the lender has enough data to price the risk with some confidence.

A thin file client is someone for whom the bureaus have very little or nothing at all to report, and this is a critically important distinction: thin file does not mean bad credit but absent credit, which means the bureau cannot tell the lender that this person is risky; it can only tell the lender that it does not know, and the lending machinery treats not knowing as a reason to decline or to apply conservative pricing. A thin file client might be a young adult applying for their first facility, a homemaker whose partner has always managed the household finances, or an informal earner who has never engaged with the formal credit system. From the bureau’s perspective all three are the same unknown quantity, and unknown quantities attract premium pricing, outright rejection, or both.

2. The R50 Problem

The total cost to originate a loan, covering bureau checks, affordability calculations, fraud screening and administrative overhead, sits somewhere between R50 and R150 depending on channel and complexity. That cost structure makes reasonable sense when you are lending R400 000 for a vehicle or R2 million for a home, because the fixed origination overhead becomes a rounding error against a transaction large enough to carry it.

Now consider what that same origination cost looks like when the loan is R50 for an electricity voucher. The cost of evaluating the credit exceeds the value of the credit itself, because the full machinery, every check and calculation and system call in that origination flow, spins at full cost to decide whether to extend a single unit of value, and every mechanism in that flow was designed for transactions three orders of magnitude larger. This is a structural mismatch that closes the door on an enormous proportion of everyday financial need before a single credit decision is even made, and raising minimum loan amounts in response, which is the lending industry’s standard reaction to this cost structure, is rational for the lender and entirely the wrong answer for the client. The people who most need small, fast access to credit are precisely the people the system is least designed to serve.

3. Born with a Credit Deficit

My wife carries no credit bureau history, managing our household, raising our family and contributing economically in ways that are substantial and entirely invisible to the credit system. Because I run the finances, she would be rejected by almost every lender who ran her profile today, not because she is a bad credit risk but because the system has no evidence either way and reads the absence of evidence as evidence of risk.

My children carry the same profile. Every young adult entering the credit system for the first time arrives there with what I can only call a credit deficit, not because they have done anything wrong but simply because they have not yet borrowed money, which in the logic of the current system is treated almost as a form of original financial sin. When they do borrow for the first time, they will absorb a fiscal loading, meaning worse rates and harder terms than someone with identical income and behaviour but a longer borrowing history, paying a premium not for the risk they represent but for the data the system does not yet have about them.

This inverts the relationship between information and opportunity in a way that should trouble everyone who thinks seriously about financial inclusion. The absence of a credit history is not a risk signal; it is a blank page, and a blank page should be read as an opportunity to build a new relationship from a clean start rather than as grounds for closing a door or inflating a price. We have constructed a system that penalises people for not having previously needed the system, and we treat that construction as though it were a natural law rather than a deliberate design choice we could revisit at any time. Why are we all born with a credit deficit? Why does the absence of history become a weakness rather than an opportunity?

4. The Western Money Flow Assumption

The credit machinery was designed around a specific and fairly narrow model of how people receive and spend money: a monthly salary deposited by an employer into a bank account, a stable set of recurring debit orders covering rent, utilities and repayments, and a credit history built through hire purchase and revolving facilities. This model describes perhaps thirty percent of the South African working population with any accuracy, and a considerably smaller fraction of the broader African working population.

The other seventy percent earn differently, with money arriving daily, weekly or irregularly through informal trade, piece work, cash businesses, remittances and multiple small income streams that aggregate to a meaningful total but never appear as a single predictable deposit. These earners may move substantial amounts through their accounts across a month while the pattern looks, to a system calibrated for salaries, like nothing coherent at all.

This failure is not confined to thin file clients; it extends to anyone whose income does not arrive in the pattern the system was trained to recognise, regardless of how much money is actually moving through their account. An Uber driver I spoke to recently illustrated this precisely. Most banks had turned him away because their systems were searching for a salary deposit tied to a recognisable company name on a predictable date each month, which is not how gig income arrives. One bank had looked past that and read the actual pattern of his earnings, which were consistent and substantial, and extended credit on that basis. The contrast was not between a creditworthy person and an uncreditworthy one; it was between a system fixated on a salary drop on the twenty-fifth of the month and a system willing to read money movement for what it actually is. The data to make a sound lending decision was present in both cases; only one institution had built the models to see it. Money flow means nothing in most current credit models, and that is not a data science limitation; it is a choice about what to measure.

5. The Machinery of Collection

Even where lenders are willing to extend credit to clients outside the salaried mainstream, the repayment infrastructure creates its own layer of distortion, beginning with the debit order system, which was designed for a specific purpose and serves that purpose well. For a home loan, a vehicle finance agreement or a salary backed personal loan, a debit order pulling a fixed amount on a predictable date from a known account is a reasonable and efficient mechanism, because the client has agreed to it, the amount is stable, and the timing corresponds to when income reliably arrives.

For everyday lending the debit order is the wrong tool, because it assumes account access that many clients do not have reliably, income timing that informal earners cannot guarantee, and it attaches repayment infrastructure with real teeth, meaning the lender acquires direct access to the client’s account, and that access creates leverage wildly disproportionate to the transaction it nominally serves. For a R50 electricity advance, the collection architecture is structurally heavier than the credit itself.

The superior model for this segment is the advance and the payment reminder, where a lender extends a small amount, the client receives a reminder when repayment falls due, and the client chooses to repay, with the resulting history captured and fed into a model actually built for the client’s context rather than imported from a different world. This is not a naive or unenforceable model; the fastest growing lending platforms on the continent by volume are built on exactly this architecture, and they work because they align the collection mechanism to the actual financial reality of the client rather than demanding the client conform to the mechanism. Debit orders and collection systems that reach into client accounts make sense for large structured purchases, but for everyday needs a visible payment reminder attached to a small advance is architecturally superior in almost every dimension.

6. Seeing the Client You Actually Have

The deeper problem is epistemological before it is operational. African banking has largely not built the measurement instruments it would need to see the clients it actually has. The bureau system captures what it captures because that is what was worth capturing in the world it was designed for, and affordability models are calibrated on income patterns that represent a minority of the population. Risk pricing is derived from default data drawn from a customer base that was already filtered through the machinery before that data was collected, which makes the whole process circular: the system prices risk based on the behaviour of the clients it was willing to accept, and that tells you very little about the clients it has never seen.

The data that would allow a lender to see an informal earner accurately does exist, living in mobile money transaction histories, airtime purchase patterns, utility payment records, the rhythm of cash deposits across a month, and the consistency of small transactions sustained over years. Some of this sits with mobile network operators and some is embedded in USSD and app transaction logs, yet very little of it is currently being used to build the credit models that would allow the system to see past its own structural blind spots. Building those models requires a prior decision that the client the system currently cannot see is worth developing new instruments to see, and that decision is as much a strategic and ethical one as it is a technical one; the technology exists, but the question is whether the will does.

7. What the Machinery Needs to Become

The credit infrastructure appropriate for Africa’s actual economy looks different from what most banks currently operate, and the distinction matters because the instinct in large institutions is to take the existing system and add an inclusion layer on top of it, which tends to reproduce all the original assumptions in a slightly cheaper wrapper rather than addressing the underlying architectural mismatch.

Loan origination needs to become dramatically cheaper for small transactions, which means removing bureau checks that cost more than the loan value and replacing them with behavioural signals that are both cheaper to acquire and more predictive for this client segment. Affordability assessment needs to move from payslip analysis toward money flow analysis that can read irregular income accurately and give it the weight it deserves, and risk pricing needs to shed the fiscal loading applied to thin file clients and replace it with a model that treats the first transaction as the opening of a data relationship rather than as a penalty event.

Repayment infrastructure for everyday lending needs to be redesigned around the advance and reminder model, preserving client agency while building the repayment history that makes subsequent lending cheaper and more accessible over time.

The bureau system needs to be supplemented with data sources that reflect how informal earners actually move money, which will require partnerships with mobile network operators, utility providers and informal sector platforms that most banks have not yet chosen to prioritise. None of this is architecturally simple, but all of it is necessary and the entire body of work is achievable with technology the industry already has.

8. The Lyric That Keeps Coming Back

That image earns its place more fully at the end of this argument than it could at the beginning.

The machinery is not broken, which is precisely the problem, because it runs exactly as designed and that precision is what keeps the bank from seeing the client. Africa’s economic majority does not move money the way the machinery expects, their creditworthiness does not appear in the data sources the machinery reads, and their repayment behaviour does not map to the collection architecture the machinery uses, so the system reports that they are not present, or too risky to serve, or that the absence of evidence is itself evidence of failure, and the door closes. Changing this requires more than a new product or a financial inclusion campaign; it requires a willingness to look at the machinery itself and ask honestly whether it was built to serve the continent’s actual population or a projection of a different population onto this one. That question, answered honestly, leads somewhere uncomfortable, and discomfort is a more productive starting point than the comfortable fiction that the machinery is working fine and the clients are simply the wrong shape.

References

1. National Credit Act 34 of 2005 — The Banking Association South Africa — The primary legislation governing consumer credit in South Africa, establishing mandatory affordability assessments and the reckless lending prohibition that drives the origination cost structure described in section 2.
2. NCA Affordability Assessment Regulations — Polity.org.za — The 2015 amendment regulations under section 81 of the NCA that formalised the pre-agreement affordability assessment duty for all credit providers.
3. TransUnion South Africa — Cross-Border Payments and Financial Inclusion — TransUnion Consumer Pulse data cited here indicates approximately 16 million South African adults currently lack active credit bureau profiles.
4. Informal Sector Employment Q4 2024 — Statistics South Africa — Stats SA QLFS data showing informal sector employment at 19.5% of total South African employment in Q4 2024, with 1.9 million non-VAT-registered businesses operating in the informal economy.
5. Work and Jobs in Africa — ISS African Futures — ILO estimates cited here place approximately 85% of total African employment in the informal sector, the continental figure underlying the argument in section 4.
6. Financial Inclusion in Sub-Saharan Africa — FinDev Gateway — Global Findex 2025 data showing Sub-Saharan Africa account ownership at 58% of adults in 2024, alongside the continent’s mobile money leadership at 40% penetration.
7. Access to Credit in Sub-Saharan Africa — Ecofin Agency — Analysis of World Bank Global Findex 2025 data showing only 7% of Sub-Saharan African adults borrowed via mobile money in 2024, despite the region hosting 52% of the world’s mobile money accounts.
8. The Global Findex Database 2025 — World Bank — The World Bank’s primary dataset on global financial inclusion, covering account ownership, borrowing behaviour, and digital payment adoption across 141 economies.
9. AI Credit Scoring: How Mobile Money Is Lending to the Unbanked — Covers the emerging architecture of behavioural and mobile data credit scoring, including the Lulalend case study showing a 30% increase in approval rates for informal retailers when regional payment cycles were incorporated into credit algorithms.
10. South Africa Alternative Lending Market — ResearchAndMarkets via BusinessWire — South Africa’s alternative lending sector growing at 29.5% annually to reach $297.2 million in 2024, with a projected CAGR of 20.2% through 2028, supporting the claim in section 5 that the advance and reminder model is gaining commercial ground.

Andrew Baker is Group CIO at Capitec Bank. He writes at andrewbaker.ninja and on Substack at @futureherman.