Zambia's smallholders are not too risky to bank. They are invisible.

By DNC · Published 8 July 2026 · 7 min read

On the best available surveys, only about 14 percent of Zambia's roughly 1.5 million smallholder farmers have access to formal credit.¹ The families who grow most of the country's food are, financially, almost unseen by the institutions meant to serve them. The common explanation is that farming is simply too risky to lend into: rain fails, prices swing, harvests are small and scattered. That explanation is not wrong. It is just incomplete, and the part it leaves out is the part worth writing about.

Ask a bank why it does not lend to a smallholder and the answers are consistent across the continent, not only in Zambia. The farmer has no collateral a bank can seize, and often no bankable land title to offer as one. The cost of reaching a single remote client for a small loan is high, and the demand is dispersed. Returns lag far behind the money that must go in. Harvests, and the prices they fetch, swing with the weather and with policy. And these risks move together: when the rain fails, it fails for everyone at once, so a lender cannot diversify the weather away. Banks also carry real institutional reasons to stay out of agriculture, from portfolio concentration limits to regulatory capital, weak contract enforcement, and the ever-present risk of political intervention in food markets. None of this is imaginary, and it would be a false choice to pretend a single fix sweeps it away.

But underneath many of these barriers sits a quieter problem they all depend on. The bank cannot tell a reliable borrower from an unreliable one, and the farmer has no way to prove which he is. Economists call this information asymmetry, and in African agriculture it runs deep. As one review of the literature puts it, farmers are unable to send out the right signals of themselves as trustworthy customers.² A lender who cannot see who is good tends to treat everyone as if they were the worst, and prices most of them out. Information asymmetry does not replace the other constraints. It sits beneath them, and it makes each one harder to price.

The scale of what this leaves unfinanced is not marginal, and here the numbers widen from Zambia to the region. ISF Advisors estimates that more than 70 percent of smallholder demand for finance goes unmet worldwide, on the order of 170 billion dollars a year.³ The IFC puts the unmet financing need across agri-SMEs and smallholders in Sub-Saharan Africa at around 117 billion dollars. Zambia's 14 percent is one country's reading of a continental pattern, not an outlier.

Here is the detail that reframes the problem. Where smallholders in Sub-Saharan Africa do get working capital, it is mostly not from banks. Agribusinesses, the traders, processors and input suppliers the farmer actually deals with, provide around 25 percent of smallholder working capital. The formal financial sector provides about 6 percent.⁵ The people financing the African smallholder are, overwhelmingly, the people who buy from him and sell to him, not the people with banking licences.

Why can the trader lend where the bank cannot? Because the trader can see the farmer. He knows what this farmer delivered last season, whether he honoured his word, what his land tends to yield. He holds, informally and in his head, much of the information the bank lacks. Proximity does not make the trader fair or his records complete: he may favour relatives, misjudge, keep poor accounts, or use his position against the farmer. Proximity does not cure the blindness. But it shrinks it, and that alone is enough to move capital the banking system will not.

This is also where the enthusiasm for algorithms meets its most useful finding. A great deal of money has gone into scoring smallholders from alternative data: satellite images of their fields, weather patterns, mobile phone behaviour. Some of it helps. But among the signals researchers at CGAP and others have tested, the one that tends to predict repayment best is not the view from space or the pattern of a phone. It is the farmer's own commercial record: his history of sales and repayments.⁶ What a farmer actually does in the market tells a lender more than what can be inferred about him from the outside.

Put those two findings together and the real problem comes into focus. The single most useful piece of credit information about a smallholder is a verified record of what he sells and what he repays. And for most smallholders that record does not exist, because they sell informally, for cash, to middlemen who keep no account the farmer can ever use. The farmer is not invisible because nothing about him is knowable. He is invisible because no one writes the knowable down. His creditworthiness is generated every season and then immediately thrown away.

That is the sentence worth remembering. Smallholder farmers are not unbankable because they are poor. They are unbankable because they are invisible, and the one thing that would make them visible, a trusted record of their trade, is a by-product of exactly the kind of formal market they are excluded from.

It is worth meeting the strongest objection head on. Banks already know that transaction histories help, a sceptic will say, and they still do not lend, so the data cannot be the missing ingredient. The answer is in the word verified. The transaction traces banks currently see for smallholders are informal, incomplete and unstandardised, precisely because the sales happen off the books. A verified, standardised, season-on-season trading record is a different object from a middleman's memory or a scatter of mobile-money receipts, and it is exactly what does not yet exist for these farmers. Nor is it the same as a national identity document, which most Zambians can obtain: an ID proves who you are, not whether you pay. The missing asset is a financial track record, not an identity.

If that is right, then the order of operations most interventions assume is backwards. You cannot credit-score your way out of a data desert; a clever model laid over missing information mostly produces confident errors. The record has to come first. And it is not built by asking farmers to fill in forms or by buying a satellite feed. It is built by trading with them in a way that captures each transaction: what they sold, when, for how much, and whether the credit advanced to them was repaid. Do that for a few seasons and a farmer who was a blank to every lender becomes, at least, legible.

That legibility is a precondition, not a guarantee. A verified record does not make a bank lend; whether lenders use it depends on their own regulation, underwriting policy and appetite for agriculture, and none of that changes overnight. This is the honest limit of the argument. The evidence shows that a farmer's commercial record improves the information a lender could underwrite on, not that lending and repayment outcomes automatically follow. That remains a strong hypothesis rather than a settled result. But without the record, the decision cannot even begin. The institution that ends up holding it does not have to become the lender. It becomes something rarer: the precondition that lenders, and the farmer himself, can finally build on.

None of this is a promise that data solves everything, and the failure modes are real. Alternative-data lending, done carelessly, has pushed borrowers into over-indebtedness, and practitioners at established banks call some of it very risky.⁷ Models fail when the data is thin, inaccurate or inferred, and borrowers learn to game whatever signal they know is watched. Regulation meant to protect farmers from predatory digital lending lags well behind the technology. And a record of a person's livelihood is not a neutral thing to hold: it raises hard questions about who owns the data, how consent is given, who may see it, and whether the farmer can take it with him. A trading record the farmer does not control is not inclusion; it is another form of capture. So the case for building the record is also a case for building it well: real verified transactions rather than proxies, owned by the farmer and shared with his consent, and lending that follows the data patiently rather than racing ahead of it. The value of a genuine trading record is precisely that it is hard to fake, because it is a history of things that actually happened.

This is the reasoning behind how DNC works. We buy maize from smallholder farmers in Zambia's Eastern Province and pay them fairly, and we record every transaction as we go. We did not set out to build a credit-scoring company. We set out to be a fair and reliable buyer, and the verified record of each sale and each repaid input advance is the by-product of that relationship, not a product we have to sell. If the evidence is right that a farmer's own commercial history is the most honest signal available of his creditworthiness, then the most useful thing an organisation can do for rural finance may not be to invent a cleverer score. It may be to give the farmer a trade worth recording, and to make sure the record belongs to him. Farmer data, governed well and controlled by the farmer, need not become surveillance. It can become the beginning of financial infrastructure, and it may be the first asset a farmer has ever held that a bank could one day recognise.

Zambia's smallholders have been called unbankable for so long that the label has come to feel like a fact about them. It is not. It is a fact about a market that never bothered to see them. Change what gets written down, and you change what they are able, one day, to borrow.