|
|
From her home outside Soweto, a South African woman who has never had a bank account opens a smartphone application. She scans her face. Within a mere four minutes, she receives her first microloan.
There was no branch visit, no need to print numerous bank statements, get a certified copy of her ID and no credit check required. The algorithm behind the scenes assessed hundreds of alternative data points and decides she is a good risk. The bank that underwrote the loan has never met her and never will. This is not a pilot program. It is already happening across South Africa, Nigeria and Kenya.
AI is doing what decades of traditional banking — and even crypto — could not, in Africa.
Access versus inclusion
The World Bank’s 2025 Global Findex suggests that around 42% of Sub-Saharan Africa’s population remains unbanked.
In Kenya, only about 10% of the population isn’t financially included, thanks to the rapid adoption of mobile money systems like M-PESA. In Nigeria, however, that figure rises to 17%, with those left unbanked often informally employed, living in a rural location, or running a cash first micro business.
South Africa, meanwhile, has the continent’s most sophisticated banking system. One in five people, however, remain unbanked. These are mostly people in townships and informal settlements, where minimum balances, documentation requirements and the distance required to travel to a branch makes banking inaccessible.
All of these people are invisible to a financial system that was never designed to see them.
AI changes who gets assessed for financial products
AI changes who is able to apply for credit, explains Michael Temitope Collins, founder and CEO of Nigerian AI and finance startup Periculum Technologies.
“The biggest thing AI does is make invisible people visible to the financial system,” he says.
His company builds credit infrastructure for financial institutions across Nigeria, Kenya, and Canada. Its AI systems can assess merchants with no credit score, no payslip and no bank account by reading and interpreting patterns in the data that merchants generate simply by living their lives. That type of data is found in people’s mobile airtime top-ups, mobile money transfers, utility payments, inventory purchases, and seasonal income flows.
Gaurav Sharma, CEO of decentralized computing network io.net, says that “AI can build a financial picture from scratch,” noting that previous systems could only digitize processes that already existed.
Catherine Lückhoff, of 20fifty, describes the people AI can now reach as “invisible primes.” These people are borrowers who appear high-risk to traditional credit scoring. In reality, however, they are prime candidates. They are truly low-risk and creditworthy when assessed by AI systems using the alternative behavioral data.
M-Pesa in Kenya
AI is being integrated into the M-PESA mobile money platform that has become widely adopted across Kenya since its launch in 2007.
M-PESA enables anyone with a basic mobile phone to store, send, and receive money in a process akin to sending a text message. No smartphone or bank account is required.
M-PESA spread rapidly because it solved the urgent problem of moving money safely, in a country where most people had no formal financial identity and bank branches were concentrated in cities.
The platform’s AI integration was launched in September 2025 and enables users to build up a credit record and to apply for credit.
The repayment data those products generate feeds back into increasingly sophisticated credit models, building a picture of borrower behavior that no formal credit bureau could have constructed from Kenya’s predominantly informal economy.
Women-led small and medium enterprises have been among the beneficiaries. Women-owned micro-businesses are often locked out of traditional credit programs for lacking collateral, formal employment records, or banking relationships.
“AI is adding to this boost in financial inclusion by providing a means to prove that someone is a safe borrower,” says Ariana Issaias, a director at pan-African law firm Bowmans in Kenya.
Kenya’s National AI Strategy 2025–2030 laid out a roadmap for the country to become a leader in AI adoption and governance across Africa. Some of its proposals, however, raise concerns about regulatory overreach, increased compliance burdens and the potential effect on investment and innovation Issaias say.
There are also concerns that mobile platforms using different AI systems to independently assess credit risk could result in users taking on too much debt. The AI systems that can make credit accessible to the unbanked, can make a loan available at 3am to someone who has already borrowed from three other platforms that week.
TymeBank in South Africa
TymeBank operates entirely without physical branches, which makes it more accessible for the many South Africans who live far away from traditional bank branches. Tymebank verifies identity documents, assesses creditworthiness from transaction behavior, and approves accounts in under five minutes through kiosks placed in Pick n Pay and Boxer supermarkets.
OM Bank, Old Mutual’s digital banking subsidiary, found the cost of serving low-income customers through traditional infrastructure was prohibitive, but AI-driven automation makes it viable.
No matter how sophisticated the technology, fintechs in general face an uphill battle attracting South African customers who live in rural or township communities. A large majority of South Africa’s people were excluded from formal finance for generations. Instead, they built their own informal financial systems including stokvels, burial societies and rotating credit associations that run on social trust and community accountability.
Persuading these communities that an algorithm is more trustworthy than a neighbor who has known them for twenty years is not a data science problem. It’s a trust problem, that no fintech or traditional financial product has convincingly solved it yet.
What AI could do next
Collins predicts that there will be three waves of AI adoption in African finance.
The first-wave of AI will sit alongside existing financial systems and improve them. It focuses on fraud prevention, credit decisions onboarding and compliance. It’s already happening.
The second-wave will occur when AI becomes embedded infrastructure, woven into payments, identity verification, cross-border remittances and insurance underwriting, rather than layered on top of them.
The third wave is the most ambitious.
“The biggest milestone will be when Africa stops just consuming AI and starts building localized AI systems trained on African data,” he says.
No such system exists at scale yet in African finance, but there are signs that things could be headed in that direction. Kasi Cloud‘s $250 million hyperscale data centre campus in Lekki, Lagos is Nigeria’s first AI-focused facility of its kind, designed to reach 100 megawatts at full build-out.
In Kigali, Horus Labs is deploying modular, renewable-powered data centres that accept mobile money payments, removing the barrier of international payment rails for African startups trying to access computing power.
Problems will persist
Improving local infrastructure will bring down costs, says Sharma:
“Many African AI companies are paying a latency and cost penalty purely because of geography. Their workloads run on servers thousands of miles away.”
African startups paying Western cloud prices on emerging-market revenue face a structural disadvantage that no amount of clever product design can fully overcome.
AI-powered financial products that work well for customers transacting $500 a month fall apart economically at the $30 many African customers might transact. The cost of computation is a significant part of why.
Lückhoff concluded our exchange with a warning: AI systems have the potential to exclude the very groups they are meant to include if they negatively assess users with thinner data trails or who are less digitally connected. She says models need to face audits, and platforms need to have clear recourse and appeals mechanisms.
“The customer must not become a data point with no right of reply,” she says. “AI is not a magic inclusion layer. It is a risk-intelligence layer. Used well, it can help banks understand and serve people whose economic lives are real but poorly captured by formal systems. Used badly, it can simply automate exclusion at greater speed.”
