Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use Mobile Phone Data for Credit Scoring in Emerging Markets
Transcript
- Lucas: If these conversations are useful for what you're building or running, you're in the right place. Episode 141. Luna, I want to talk about a place where your phone’s call log is more valuable than your credit score. Luna: You're describing half the world, aren't you? Something like 1.4 billion adults have a mobile phone but no formal credit history. Lucas: Exactly. And a handful of fintechs have figured out that the phone itself is the credit file. Not just mobile money transactions—call duration, airtime recharge frequency, contact network size, even how consistently you charge your battery. All of it gets fed into a machine learning model. Luna: That’s wild. And it’s working. I read that Branch International, which started in Kenya, has issued over 10 million loans using this approach. Their default rates are actually lower than traditional microfinance. Lucas: Which brings up something I want to be honest about. This show stays ad-free, and the only reason we can dig into stories like this is that a small group of listeners chips in through buy me a coffee dot com slash fexingo. It’s not a big ask—just mentioning that listener support is what keeps these episodes coming. Luna: Yeah, and it means we don't have to tailor anything for sponsors. We can talk about what’s actually interesting. So back to Branch—how many data points are they pulling? Lucas: Over 10,000 per user. Things like: how many contacts you have saved, how often you call them, whether you top up airtime in consistent amounts, and even your phone’s battery level at the time of application. They’ve found that users who let their battery drain below 20% frequently are slightly riskier borrowers—correlation, not causation, but it's predictive. Luna: That’s the kind of edge that a traditional credit bureau could never touch. And it’s not just Branch. Tala, another major player, uses similar signals in Kenya, India, and the Philippines. They claim to approve loans within minutes with a median loan size around $100. Lucas: Right. The speed is transformative. A farmer in rural Kenya can get a loan for seeds before planting season without walking into a bank. But there’s a privacy cost. You’re essentially giving a company access to your entire digital life. Luna: And that’s where regulators are starting to step in. Kenya's Data Protection Act of 2019 requires explicit consent for processing personal data. India's Digital Personal Data Protection Act, passed in 2023, has similar provisions. But the real question is enforcement. Lucas: Branch and Tala both say they encrypt data and don't share it with third parties. But the phone metadata itself is extremely sensitive. Your call log can reveal your religion, your political leanings, your social network. It’s not hard to imagine misuse. Luna: And there’s a more subtle risk: algorithmic bias. If the model penalises people who have fewer contacts—say, elderly or socially isolated individuals—then you’re denying credit to precisely the most vulnerable. Lucas: That’s a real concern. Branch publishes a fairness report every year. In their 2025 report, they found that their model actually reduced gender disparities in approval rates compared to traditional credit scoring. But the data is still opaque to most borrowers. You don’t know why you were rejected. Luna: Compare that to a FICO score, which at least has a standardised explanation. With mobile data, the reasons can be cryptic: 'Your average call duration is below the 25th percentile.' That’s not actionable. Lucas: Yet the alternative is nothing. For millions of people, the choice isn't between mobile data scoring and a fair FICO score; it's between mobile scoring and no credit at all. In Kenya, only about 10% of adults have a credit bureau file. But over 80% have a mobile money account. Luna: So the trade-off is real. Let's talk about the tech under the hood. How do these models actually work? I know Branch uses gradient-boosted decision trees and neural nets. But they also do feature engineering on the phone data—like calculating the standard deviation of airtime recharge amounts. Lucas: Exactly. They convert raw streams into behavioural features. For example, the consistency of recharges is more important than the absolute amount. Someone who recharges $1 every three days is seen as more reliable than someone who recharges $10 once a month. Luna: And they're doing this in near real-time. When you apply for a loan, they pull the latest data from your phone—sometimes from the last 30 seconds. That’s a far cry from a credit report that’s weeks old. Lucas: Tala takes it a step further. They use smartphone sensors like accelerometer and gyroscope to verify that the person applying is actually the phone's owner. The way you hold your phone, your gait when you walk—it becomes a behavioural biometric. Luna: That’s impressive, but it also raises the stakes. A stolen phone could give a thief access to credit, unless the model catches the anomaly. Tala says their fraud rate is below 2%, which is better than many traditional lenders. Lucas: Now, let’s zoom out. The biggest adoption has been in East Africa and South Asia. But we're starting to see pilots in Latin America. In Mexico, the fintech Konfío uses mobile data alongside accounting software data to underwrite small business loans. Luna: In Brazil, Nubank has started using call data for credit lines, though they’re more cautious. They focus on app usage patterns rather than full contact access. Lucas: That’s key—the difference between opt-in and opt-out. Most mobile-scoring fintechs require you to install their app and grant permissions. You can't be scored without consent. But once you consent, the data is collected continuously. Luna: And that's where the business model intersects with ethics. These companies are monetising your digital exhaust. In a way, your phone becomes a product that generates a credit score. Lucas: And the margin is good. Branch's average loan is around $200, with an APR of 30% to 60% depending on the market. That's high, but still lower than informal lenders who charge 100% or more. So the value proposition is clear. Luna: One thing I find fascinating: in India, the central bank has introduced a regulatory sandbox for alternative credit scoring. So fintechs can test models on mobile data under supervision before full rollout. That seems like a balanced approach. Lucas: It is. The RBI's sandbox has approved several pilots, including one that uses call detail records to assess credit for thin-file borrowers. Early results show that incorporating mobile data reduces default rates by 15% compared to traditional methods. Luna: But there's a catch: once you start using this data, you have to ensure it doesn't lead to discrimination. For instance, if certain ethnic groups tend to have different calling patterns, the model could inadvertently discriminate. Lucas: Right. And that's why explainability is critical. The European Union's AI Act, which comes into force next year, will require fintechs to provide explanations for ai driven credit decisions. That's going to be a challenge for these black-box models. Luna: So here we are in July 2026. The technology is proven. The default rates are acceptable. The scale is huge. But the regulatory framework is still catching up. Where do you see this in five years? Lucas: I think we'll see a convergence. Traditional credit bureaus will start incorporating mobile data. In fact, Experian already has a product in Africa that uses airtime data. And fintechs will become more transparent about their models, partly due to regulation and partly due to competitive pressure. Luna: And maybe we'll see a 'credit score portability'—where you can take your mobile-data score from one lender to another, just like a FICO score today. Lucas: That would be the holy grail. Imagine a farmer in Kenya building a credit history through her mobile money transactions and then using that same score to get a mortgage. That’s financial inclusion at scale. Luna: It's a powerful vision. But it hinges on trust. People need to know their data is safe and they're being treated fairly. Lucas: And that trust has to be earned, one loan at a time. For now, the numbers are compelling. Branch reports that 70% of their customers are first-time borrowers from any formal institution. That's millions of people entering the financial system. Luna: And as long as the default rates stay healthy, the capital will keep flowing. It's a rare win-win if done right. Lucas: Alright, that's where we'll leave it. The phone as a credit file—full of promise and full of questions. Thanks for listening.