Can AI promise Egypt’s lenders more borrowers and fewer bad loans?

Regulators are starting to ease the path for more machine decision-making in lending. From April 2027, consumer finance companies will have to consult a score from the country’s only licensed private credit bureau (I-Score) before approving financing, under a new rule from the Financial Regulatory Authority. The score rests on behavioral analysis of alternative data, the kind of input that feeds directly into AI-backed risk prediction and automated lending decisions.

On the plus side: AI models are making it easier to assess the creditworthiness of small businesses and individual borrowers, especially newcomers to the financial system and lower-income people without bank accounts. These people often have little or no conventional credit history, but an AI model can use non-conventional data to make a decision. For people wanting personal loans, that can mean combing through their phones to look at how they pay their phone and utility bills, Ahmed Abaza LinkedIn, co-founder and CEO of Synapse Analytics, tells EnterpriseAM.

Hand-reviewing each loan eats the margin: Making a decision on every personal or small-business loan by hand takes so much staff time that it can wipe out much of what the lender earns on the facility, a World Bank Group SME note (pdf) has found. Letting software make those decisions can cut that cost significantly, McKinsey says, making those segments, as well as microfinance, more attractive to banks and non-bank lenders alike.

Why it matters: Fewer than half of Egyptian adults have a bank or mobile-money account, according to the World Bank (pdf), which makes it hard to judge their creditworthiness on traditional data alone.

The promise

Advocates say there is an open-and-shut case for turning lending decisions over to AI agents, arguing they are smarter, faster, and more capable than humans. Think of it as hiring “the best risk agent you can ever hire, that never sleeps, never eats, never takes time off,” and who’s always “exploring how to accept more customers with less risk,” Synapse’s Abaza tells EnterpriseAM.

The results prove it, Abaza says. Synapse, whose software is used by more than 50 institutions, says its clients are approving up to five times more customers with 40% fewer bad loans compared with before they signed up with Synapse. Abaza says his clients are not lowering the bar; they’re just getting through far more applications. Banks that once approved around 10k people a month may end up approving 10k a day, he adds.

Fintechs are already deploying: MNT-Halan uses its in-house tool, Apollo, to tell staff which overdue loans to chase each month, as we’ve previously reported. It will roll the tool out across “many departments” within three months, CEO Mounir Nakhla said at our Egypt Forum, weeks before the company’s Egyptian unit debuted on the EGX. Another fintech, Klivvr, expects to soon let AI agents make underwriting decisions without any human approval, as we reported last week.

More are joining: Tanmeyah, EFG Finance’s microfinance arm, is rolling out its own credit-scoring model through I-Score’s Decision-as-a-Service platform, built on Synapse Analytics’ Konan engine and using bureau data to size up micro- and small-business borrowers. In September, Raya Holdings spun up Raya Nexus, an AI shop with EGP 250 mn in issued capital that sells credit decisioning, collections, and customer management tools to banks and fintechs.

A less rosy version

On the other hand, AI is trained on existing data with existing biases. If women, informal workers, and young people barely show up among a lender’s past borrowers, a model can end up excluding them. So while AI may be able to dramatically increase efficiency, it can simultaneously remove the human elements of any business transaction. It’s a trade-off that’s especially troubling when it comes to personal, home, and business loans, which can be life-changing for many borrowers.

Executives want a human in the loop. “I use [AI] to increase human efficiency, not to replace people. A human must guide it and assess the output, because AI can often be inaccurate,” Hossam Heiba, managing director of Orbis Capital and former head of the General Authority for Investment and Freezones (Gafi), tells EnterpriseAM.

There’s also anxiety about moving too fast toward AI, especially at more traditional banks. Executives there worry about giving power to AI without ensuring that its decisions are properly vetted and that there’s a human taking responsibility. “If you have poor governance and no accountability, it simply leads to making bad decisions faster,” HSBC Egypt CEO Todd Wilcox LinkedIn said at our Egypt Forum last week.

Corporate loans are a harder sell: Some banks already use AI to help judge whether a company can handle a loan, Bassem Fayek LinkedIn, managing director and partner at Boston Consulting Group (BCG), tells EnterpriseAM. But handing the final call to a machine is tough when a bank’s reputation is on the line. “I answer to the depositors and the shareholders. But the buck stops with me and my team,” Wilcox said at the forum. “It’d be difficult to defend to depositors that the bank failed because of our AI model and [its] judgment.”

Most banks are keeping AI away from their core infrastructure, for now. They’re using it with customers, “but not in core banking operations yet due to high confidentiality and data protection rules,” Heiba says.

Egyptian banks aren’t spending much on AI, either. Only “a couple” of Egypt’s 30-plus banks are well ahead of the curve, BCG’s Fayek says. The nation’s two biggest state banks trail the region too: on the Evident AI Index of 25 banks across the Middle East and Africa, Banque Misr ranked 19th, and the National Bank of Egypt came last.

Messy data is another drag: AI is only as good as the data it is fed, and most companies’ data is a mess, Hazem Gebili, chief AI and digital officer at Orange Egypt, tells us. Records sit scattered in old systems across a company, much of it inaccessible to AI, and just plugging AI into the way things already work would be “the biggest failure,” he says. “You need to reimagine the journey.”

The wall is computing power: Central bank rules effectively prevent Egyptian lenders from sending customer data to cloud-based AI services like ChatGPT, Claude, or Gemini, two bank CEOs confirmed to EnterpriseAM. Non-bank lenders have told us they feel similarly constrained. The result is that those experimenting with AI today are running smaller, open-source models on their own servers instead, Abaza notes. That means buying GPUs, the super-expensive advanced computer chips that run most AI. “In Egypt, sometimes it’s very hard to get GPUs,” he says.