Illustrative image: Capitec is using AI to open doors to the traditional banking system rather than slam them shut. (Image: Midjourney AI) 
South African banks are going deep on artificial intelligence (AI). Fraud detection, customer service bots and predictive analytics are the new battlegrounds for the Big Four. But look past the hype and you’ll see that Capitec is bending AI into shapes its American and European inventors never imagined, because South Africa’s risk profile is unlike anything you’d see anywhere else.
“A lot of these models come in from overseas, and they’re based on things that actually don’t work here,” Capitec chief technology officer Andrew Baker told Daily Maverick during a media roundtable at the AWS (Amazon Web Services) Summit in Joburg last week.
“In the South African theme market, traditional creditworthiness models mean most people will fail. We can’t just copy-paste those assumptions.”
Instead, Capitec has been tinkering at the edges of micro-lending, using what Baker calls “micro-decision lending”. Something as small as a R10 airtime advance becomes a test of trust: if clients repay promptly, the bank system knows it can scale up the offering.
It’s a deliberate rejection of the blunt-force credit scoring tools imported from global financial hubs, and it shows how AI can be used to open doors rather than slam them shut.
This approach matters in a country where millions remain “thin file clients” – people with no formal credit history who would usually be invisible to conventional lenders. This is what it takes to bank the unbanked, even if it means starting with tiny loans to street vendors or first-time earners.
“AI lets us democratise credit,” Baker said. “It’s about applying logic to lend responsibly where others see only risk.”
Fighting fraud with algorithms
If micro-lending is Capitec’s offensive play, fraud detection is its defensive wall and, here too, AI is directing the backline.
Now that Capitec runs on AWS, it can do things that would have been science fiction a few years ago. “We can literally pull payments out of the air,” Baker explained. “If we see something’s wrong with a payment, we’ll pluck it off the network and stick it somewhere.”
Every transaction from Capitec’s 25 million clients passes through what Baker calls a “beneficiary firewall”. If a recipient has been flagged in scams elsewhere, the system throws up a warning before the client even hits send.
And then there are the 30 malware probes that live inside the Capitec app, scanning your phone like a digital sniffer dog.
“If I get a signal from one of those probes, I’m not letting your payments through.”
The app can even act like a malware removal tool, rooting out ransomware that has hijacked your device.
Behavioural banking
Unlike traditional binary fraud systems, which block or allow transactions outright, they take a probability approach. If the system isn’t sure, it slows things down, quarantining the payment while machine learning models grind away in the background.
The system also measures intent by cross-referencing a global beneficiary database. If you suddenly send money to a never-before-seen account with suspicious velocity or amounts, the app might pop up a chat: “Hey, we saw this. Why are you doing this?”
Even services like PayShap, which can be exploited by shifting endpoints, are checked against Capitec’s databases to sniff out fraud markers.
The upshot is a multilayered AI defence grid that stops fraud and nudges safer behaviours.
AI integration isn’t just about keeping money safe. It’s also about mining insights that feed back into product development.
Shifting to AWS has given the bank back-to-back months with “zero-hour downtime” and the ability to roll out features faster. “We’ve had projects landing within two months that would have taken a year before,” Baker said.
Capitec now even deploys conversational AI assistants for its analysts, using tools that can dive into oceans of transaction data and return insights in plain language or visual dashboards. Instead of wasting weeks on manual reports, analysts can probe problems and innovate new products.
One tangible example: data showed young clients struggling to build credit. That insight led directly to the launch of a low-limit starter credit card for youth, a small but significant step in building financial literacy and trust.
“Data-driven product innovation sounds like jargon, but for us, it literally means creating new products that solve real customer problems,” Baker said.
Yet even the most resilient system can wobble. Barely two hours after Baker’s conversation with Daily Maverick, Capitec suffered an outage that left millions of clients unable to transact.
The fragility of uptime
For Baker, the irony was unavoidable. He had just finished explaining his obsession with N+1 resilience (always having a backup system ready to fail over), and his intolerance for lag.
“I want five-millisecond latency or less,” he said. “If we can’t meet that, we’re failing our clients.”
The outage demonstrated the knife’s edge on which digital banking operates. Customers don’t care whether the cause was AWS, an upstream provider, or a routing error. They expect services to work. And Capitec’s shaky reputation for reliability means the margin for error is brutally thin.
Microsoft in the back office
While Capitec is customising AI for township traders and first-jobbers, Microsoft is cementing its hold on the quieter, but equally lucrative, back office.
Enterprise Resource Planning and Customer Relationship Management (CRM) systems are the plumbing of modern banking, and in South Africa Microsoft’s Dynamics suite, combined with its Copilot and Power Platform, is becoming the default.
Nedbank and FNB are among the banks integrating Copilot into customer engagement, even supporting conversations across 13 local languages. At FNB, 94% of bankers reportedly use Copilot in daily workflows.
For Ravi Bhat, chief solutions and AI transformation officer at Microsoft South Africa, the pitch is simple: don’t waste millions building your own AI, just plug into ours. Copilot works across platforms, even non-Microsoft CRMs such as Amazon Connect. For data-heavy industries like banking, the allure of off-the-shelf AI is obvious.
But the company’s real advantage may be compliance. South African banks wrestle with strict data governance and regulatory frameworks. Microsoft’s hybrid cloud, a blend of localised Azure servers with global standards, allows banks to stay compliant while still harnessing generative AI.
“Banks don’t get stuck on the technology,” Bhat tells Daily Maverick. “They get stuck on governance and security.” Microsoft has quietly built itself into the solution to that problem.
A leap of faith
Where Capitec goes, it’s digital-first rival TymeBank is not far behind. And it’s right there next to the Techno Park company, staking its future on AI.
But rather than caution, TymeBank’s rhetoric is bold, even evangelical: “At TymeBank, we’re not afraid of AI and its limitless potential. We’re investing in it because we don’t see AI as a threat but rather as a tool.
“Smarter systems, faster answers, better service… That’s what innovation should do: enable people, not replace them,” TymeBank chief commercial officer Cheslyn Jacobs wrote on LinkedIn.
Where Capitec talks about probes, firewalls and probability fraud models, TymeBank speaks in the language of empathy and efficiency. For it, AI is about freeing staff from admin to “build real relationships”.
That’s not just branding spin. In a highly competitive digital banking market, tone matters. TymeBank is positioning AI not as the cold hand of automation but as the enabler of warmth.
For all the banks, the endgame is the same: trust. Fraud detection, credit scoring, customer service; AI is becoming the unseen referee. DM
Lindsey Schutters is a Business Maverick journalist.
This story first appeared in our weekly Daily Maverick 168 newspaper, which is available countrywide for R35.
