Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use Social Media Data for Credit Scoring
Transcript
- Lucas: So Luna, when you think about credit scoring, what comes to mind? FICO score? Maybe your debt to income ratio? Luna: Yeah, pretty much. Payment history, credit utilization — the usual suspects. But I know you're about to tell me there's a whole different world out there. Lucas: Exactly. There are fintechs underwriting loans based on stuff like your Facebook friends, how many texts you send, even the time of day you charge your phone. And they're doing it in real time. Luna: That sounds both fascinating and a little creepy. But I bet it's reaching people who'd otherwise have no credit profile at all. Lucas: Right. And that's exactly the kind of insight we try to bring you on this show — no ads, no sponsors, just the story behind the technology. If you find these episodes useful, you can support us at buy me a coffee dot com slash fexingo. Luna: It's a small way to keep this going. Alright, so who's actually doing this at scale? Lucas: The poster child is Tala. They operate mainly in Kenya and the Philippines. A user downloads the app, gives permission to access their smartphone data, and within minutes Tala can approve a small loan — sometimes as little as ten dollars, sometimes a few hundred. Luna: And they're using social media specifically? Like, what are they looking at? Lucas: It's not just social media. They look at over 10,000 data points. But social signals are a big piece. For example, the size and structure of your social network — how many Facebook friends you have, how frequently you interact, whether those connections seem real or like bots. Luna: So if you have a lot of strong ties, that's a positive signal for repayment? Lucas: Exactly. Tala found that users with more diverse social networks — friends from different cities, different professions — tend to default less. It's a proxy for stability and community embeddedness. Luna: Huh. What about call logs or text messages? I've heard some fintechs use those too. Lucas: Yes. Tala looks at call frequency, duration, even the time of day calls happen. Someone who makes calls late at night might be judged as less reliable — it's a pattern that correlates with higher default rates. They also look at SMS: do you receive regular messages from the same contacts? Is there evidence of a job or a landlord relationship? Luna: That's incredibly granular. But how accurate is it compared to traditional credit scores? Lucas: In markets where traditional scores barely exist, Tala's model has a default rate comparable to or better than microfinance institutions. They publish a repayment rate above 90 percent. And because they can re-evaluate in real time, they can adjust loan offers based on recent behavior — if you suddenly start calling new numbers at odd hours, your credit limit might drop. Luna: So it's dynamic underwriting. But there's a huge privacy question here. Are users really aware that their social graph is being analyzed for credit? Lucas: That's the tension. Tala and similar fintechs require explicit consent during app installation. But how many people read the fine print? In emerging markets, where the need for credit is urgent, users often accept terms without full understanding. Regulators are starting to take notice. Luna: What's the regulator stance in Kenya or the Philippines? Have they put guardrails in place? Lucas: Kenya's Data Protection Act, passed in 2019, requires clear consent and data minimization. The Philippines has similar rules under the Data Privacy Act. But enforcement is still catching up. Some fintechs have been criticized for using data in ways users didn't expect — like accessing photos or contacts without clear purpose. Luna: Is there a risk that this kind of scoring could introduce bias? If your social network is small because you're an immigrant or a low-income worker, you might be penalized. Lucas: Absolutely. That's a real concern. Tala says they audit their models for disparate impact, but it's hard to eliminate bias entirely. The good news is, because the data is so rich, they can find alternative signals that don't rely on network size. For example, consistency in app usage — someone who checks their phone at regular intervals might be more organized. Luna: Let's talk about the technology stack. How do they process 10,000 data points per user in real time? Lucas: Tala built a proprietary machine learning platform they call the 'credit engine.' It runs on-device initially — the phone's processor extracts features like app usage patterns, contacts, GPS location — then sends a compressed feature vector to the cloud for scoring. The whole thing takes under 60 seconds. Luna: That's impressive. And they're doing this for millions of users? I read Tala has over 6 million customers. Lucas: Yes, 6 million and growing. They've disbursed over three billion dollars in loans since launch. The model improves continuously because every loan outcome feeds back into the algorithm. Luna: So it's a classic data flywheel. The more loans they make, the better their predictions get. Lucas: Right. And they're not alone. Competitors like Branch and PayJoy use similar approaches. Payjoy, for example, uses the phone's health metrics — like battery drain rate — to determine if a loan is risky. If your battery is often low, you might be less reliable. Luna: That's wild. Next thing you know, lenders will look at your fitness tracker data to see if you're disciplined enough to pay back a loan. Lucas: Some startups are already experimenting with that. But let's bring it back to the broader picture. The trend is clear: any digital exhaust can be credit data. And it's not just emerging markets. In the US, companies like Upstart use education and job history. A few are starting to incorporate social data too, though with more regulatory scrutiny. Luna: Do you think the FICO score will eventually be replaced by a more dynamic, data-rich profile? Lucas: I think we'll see a hybrid. FICO is deeply embedded in the lending infrastructure. But for thin-file or no-file consumers — which is about 26 million Americans — alternative data is the only way in. So fintechs will continue to innovate, and incumbents will either acquire or partner. Luna: Speaking of partnerships, last year Mastercard launched a 'social media scoring' pilot in some markets. Are you seeing big banks get interested? Lucas: Yes, cautiously. The big banks are watching. They don't want to be seen as intrusive, but they also want to expand their customer base. I suspect we'll see a few pilot programs in 2027, especially for credit cards aimed at younger consumers who live on social platforms. Luna: Right, because Gen Z might actually prefer being scored on their TikTok activity than on a missed student loan payment. Lucas: That's probably true. And ironically, they might be more comfortable with it — as long as it's transparent. The key is consent, control, and the ability to challenge the data. Tala lets users see which factors influenced their decision, which builds trust. Luna: Let's circle back to Tala for a second. How do they handle data security? If your phone is compromised, someone could steal your identity and apply for loans. Lucas: Good point. Tala uses encryption both at rest and in transit. They also do device fingerprinting — linking the loan to the specific phone. So even if someone steals your data, they'd need your physical device to get approved. Plus, they monitor for unusual application patterns. Luna: Okay, so the technology is robust. But the bigger question: does social media data actually predict creditworthiness better than traditional metrics? Lucas: In contexts where traditional data is scarce, yes. But in mature markets, research shows that basic transactional data — like bank account cash flow — outperforms social signals. So social data is best used as a supplement, not a replacement. Luna: That makes sense. So where do you see this going in the next five years? Lucas: I think we'll see a convergence. Fintechs will combine social, behavioral, and transactional data into a single 'digital identity score' that's portable across lenders. Regulation will force transparency, and consumers will have more control over their data — maybe even get paid for sharing it. Luna: Imagine getting a better interest rate just because you post consistently on LinkedIn. Lucas: Exactly. Or a discount on your insurance premium if you keep a regular sleep schedule. The possibilities are endless — and a little unnerving. Luna: Well, that's a great place to leave it. Thanks, Lucas. Lucas: Thanks, Luna. And thanks to our listeners. If this episode got you thinking, share it with someone who works in credit or fintech. We'll be back next time with another deep dive.