AI in Hiring and Lending: Promise and Peril for Africa

A friend of mine in Dar es Salaam applied for a small loan through one of those lending apps last year. She had run a vegetable stall for six years, paid her suppliers on time, and topped up her mobile money line almost every week. The app asked for permission to read her contacts and her SMS messages. She tapped "allow," because that is what you do when you need the money. A few seconds later, the screen said no. No reason. No human to call. No way to ask what she had done wrong. She still does not know.

Around the same time, a young graduate I mentor submitted his CV to a multinational hiring through an online portal. He never heard back. Months later he learned the company used an automated screening tool that filtered candidates before any person looked at a single application. He may have been a good fit. He may not have been. The point is that nobody ever decided — a model did, silently, in a fraction of a second.

These two stories are becoming ordinary across the continent. Algorithms are increasingly deciding who gets a job and who gets a loan in Africa. I want to look at this honestly, because the truth is neither the glossy pitch the fintech founders give nor the doom some critics warn of. It is both promising and dangerous at the same time, and we have to hold both ideas in our heads.

The genuine promise

Let me start with what is real and good, because it is easy to be cynical and miss it.

For generations, the formal banking system in much of Africa worked like a locked room. To get a loan you needed a credit history. To have a credit history you needed prior loans from formal institutions. To get those you needed collateral, payslips, a salaried job — the very things most people in the informal economy do not have. Millions of hardworking traders, farmers, and hustlers were simply invisible to the system. Not because they were bad borrowers, but because the system had no way to see them.

AI-driven credit scoring genuinely cracks open that locked room. Instead of asking "do you have a formal credit history," these models ask "what does your actual behaviour tell us?" They look at alternative data: how regularly you top up airtime, the rhythm of your mobile-money transactions, how consistently you repay small amounts. A market trader who moves money every day, pays suppliers reliably, and keeps her phone active is showing real signals of creditworthiness — signals the old system threw away.

When this works well, it is a quiet revolution. Someone who could never walk into a bank branch and be taken seriously can now receive a small loan in minutes, grow a stall into a shop, smooth out a bad month, or pay school fees on time. On the hiring side, AI can sift through thousands of applications quickly, which matters in markets where a single advertised post draws hundreds of candidates. In principle, that speed could let smaller companies hire more fairly and reach people they would never have found.

I do not want to dismiss any of that. The promise is not marketing fluff. It is a real chance to extend dignity and opportunity to people the old institutions ignored.

The peril, just as real

But the same tools that open the door can also slam it — and do it invisibly, at scale, with the false authority of a machine.

The first problem is bias. A model learns from data, and our data carries our history. If past lending favoured men, or city dwellers, or people from certain regions, a model trained on that record will quietly learn to keep favouring them. It does not need a column labelled "tribe" or "gender" to discriminate — it can infer those things from your location, your name patterns, the people in your contacts, the way you use your phone. The discrimination becomes harder to see precisely because it hides inside a system we are told is objective. I have written more about this in AI Bias and Ethics, Explained Simply, and it is worth understanding before you trust any "data-driven" verdict on yourself.

The second problem is thin data. Alternative data sounds clever until you remember that a person's airtime habits are not the same as their character. Two equally honest people can have very different phone behaviour. When a model treats every signal as meaningful, it can punish people for being poor, rural, or simply private.

The third problem is privacy, and it troubles me deeply. Think again about my friend's loan app demanding access to her contacts and messages. Why does a lender need to read your private conversations or know everyone you call? In too many cases that data is harvested, stored loosely, and sometimes weaponised. We have seen predatory digital-lending apps that, when a borrower falls behind, message that person's entire contact list to shame them publicly — calling their pastor, their mother, their boss. That is not credit. That is harassment dressed up as fintech.

The fourth problem is the black box. When a human loan officer says no, you can at least ask why and sometimes argue your case. When a model says no, there is often no explanation and no appeal. This is the part that struck me most in both my stories: the silence. Being rejected is hard. Being rejected with no reason, no person, and no path to challenge it strips away something human. People deserve to know why a decision was made about their lives.

And finally there is the debt trap. The same speed that makes digital credit empowering also makes it dangerous. Loans you can get in thirty seconds, repaid on terms you barely read, with rates that compound brutally, have pushed real people into spirals — borrowing from one app to pay another. The technology removed the friction that used to make us pause and think.

What fairness and accountability should look like

So how do we keep the promise without the peril? I do not think the answer is to reject the technology. I think it is to demand that it be built and governed with a conscience. A few principles matter to me.

Transparency. A person should be able to know, in plain language, the main reasons a decision was made. "Your application was declined mainly because of X and Y" is not too much to ask. Opacity should not be a business model.

A right to explanation and appeal. Every automated decision that affects someone's livelihood should come with a real channel to a human being who can review it. Machines make mistakes. People should be able to say "you got this wrong" and be heard.

Human review for high-stakes calls. Speed is wonderful for convenience and dangerous for justice. The bigger the decision, the more a human should stay in the loop.

Regulation with teeth. Several African regulators have begun cracking down on predatory lenders, capping abusive practices, and licensing operators. This is the right direction. Data-protection laws are spreading across the continent, and they matter — but laws only help if they are enforced.

Fairness testing. Builders should actively test their models for discriminatory patterns across regions, genders, and groups, and fix what they find. Bias is not a rumour you wait to hear about. It is something you go looking for, on purpose, before you ship.

What you can do to protect yourself

If you are an applicant or a borrower, you are not powerless. A few habits help.

Guard your data like it is money, because it is. Only grant app permissions that genuinely make sense, and uninstall anything that overreaches. Borrow only from licensed, reputable lenders, and read the cost of the loan — the real annual cost, not the friendly little number on the screen. If you can build any formal financial footprint, do it; it gives you more options and more leverage. Keeping your mobile money secure is part of this too, and I have laid out practical steps in How to Keep Your Mobile Money Safe. And if you are starting to think beyond surviving month to month, How to Start Investing in Africa is a calmer path to building something that does not depend on an app's verdict.

On the hiring side, keep your CV clear and keyword-honest, since a machine may read it first — but never lie to a model. Where you can, find a human contact inside an organisation. People still open doors that algorithms keep shut.

What African builders and businesses should do

This part is for us — the founders, engineers, and managers building these systems. We carry more responsibility than we sometimes admit.

Build for the long term, not for the squeeze. A lending model that traps people is not a business; it is extraction, and it will eventually meet a regulator or a reckoning. Collect the minimum data you need and protect it seriously. Make your decisions explainable, even when the law does not yet force you to. Keep a human in the loop for the decisions that change lives. Test for bias before you launch and keep testing after. And remember who is on the other side of the screen — not a "user," but a market trader, a graduate, a mother trying to pay school fees.

We are early enough in this story that the culture is still being set. The choices African builders make now — whether we treat people as data to be mined or as customers to be served — will shape the continent's relationship with this technology for a long time.

I keep coming back to my friend and her vegetable stall. The machine that judged her in a second knew her airtime habits but not her six years of honest work. That gap — between what the data sees and who a person actually is — is the whole problem, and the whole opportunity. We can build systems that close it, or systems that pretend it does not exist. I would rather we chose the harder, more human path.

Frequently asked questions

How does AI credit scoring actually decide if I get a loan?

Instead of relying on a formal credit history, many lenders feed a model alternative data — your mobile-money transaction patterns, how regularly you buy airtime, repayment behaviour on past small loans, and sometimes phone-usage signals. The model turns those patterns into a score. The upside is that it can include people the banks ignored; the downside is that it can punish you for habits that have nothing to do with your honesty.

Why was my loan or job application rejected with no explanation?

Because an automated system likely made the call, and many of them are 'black boxes' that give no reasons. This is exactly the problem with opaque AI decisions. You have every right to ask the lender or employer for an explanation and a human review, and growing data-protection rules in several African countries are starting to back that right.

Are digital lending apps safe to use?

Some are genuinely useful; others are predatory. Warning signs include demands to access your contacts and messages, hidden or sky-high interest rates, no clear licensing, and no real customer service. Stick to lenders licensed by your country's regulator, read the true cost of the loan, and never grant permissions a loan does not actually require.

Can these AI systems be biased against me?

Yes. Models learn from historical data, and if that data reflects past discrimination — by region, gender, or background — the model can quietly repeat it, even without being told those details directly. It can infer them from your location, name, or contacts. This is why fairness testing and human oversight matter so much.

What can I do to improve my chances fairly?

Protect your data and only use reputable, licensed lenders. Build any formal financial footprint you can, since it gives you more options. Keep your mobile money secure. For jobs, keep your CV clear and honest, and try to reach a real person inside the organisation — humans still override machines more often than you think.

Further reading on this site

• AI Bias and Ethics, Explained Simply
• How to Keep Your Mobile Money Safe
• How to Start Investing in Africa
• Browse Business

If this gave you something to think about, subscribe to the newsletter for more honest writing on technology, money, and life in Africa.