Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use Wearable Payment Data for Credit Scoring
Transcript
- Lucas: I want to start today with something I saw on my own wrist this morning. My Apple Watch buzzed with a notification from Nubank — I'm a beta tester for their new credit scoring feature — and it said, essentially, that my payment consistency score had gone up because I'd used tap to pay at the same coffee shop three days in a row. Luna: Wait — your watch is telling you your credit score is improving because you bought coffee? That is a wild leap from the traditional FICO model of 'did you pay your mortgage on time.' Lucas: Exactly. And that's what we're digging into today: how fintechs are starting to use data from wearables — smartwatches, fitness bands, even smart rings — to build alternative credit scores. We're talking about a world where the frequency and consistency of your tap to pay transactions, the time of day you spend, maybe even your step count relative to spending, becomes part of your financial profile. Luna: And that's not science fiction. Nubank has been running this pilot since April 2026 in Brazil, and they just expanded it to about 200,000 users. So let's talk about what they're actually seeing. Lucas: Before we go deep — if this kind of conversation is useful for what you're building or running, these episodes stay ad-free because listeners choose to support the show. The way to keep that going is buy me a coffee dot com slash fexingo. It genuinely makes a difference, and it lets us keep covering these frontier topics without sponsors calling the shots. So thanks for that. Luna: Yeah, it's a small thing that adds up. Back to the wearable credit score — what metrics are they actually using? Lucas: So Nubank's pilot pulls three main data points from the wearable's payment history. First, transaction frequency — how many tap to pay transactions per week. Second, consistency — do you spend at roughly the same merchants, or are you hopping around? Third, time-of-day patterns — are you making purchases at 2 AM regularly, which might signal instability or just a night-shift worker. Luna: That third one feels dangerous. A night-shift nurse would get penalized for buying groceries at 3 AM. And that's exactly the kind of bias alternative data can introduce if you're not careful. Lucas: Nubank says they've trained their model to recognize shift-worker patterns — they actually published a white paper last month showing that after controlling for occupation proxies, the time-of-day signal loses most of its predictive power. The stronger signals are frequency and consistency. Their data shows that users who complete at least 12 tap to pay transactions per week with no gaps longer than 48 hours have a default rate about 30 percent lower than the general unbanked population in Brazil. Luna: Thirty percent lower — that's significant. And it makes intuitive sense: if you're using a wearable to pay, you're probably more organized, more digitally literate, maybe more stable. But the sample is still self-selecting. Lucas: Right. You have to own a wearable and be willing to share that data. So this isn't a replacement for traditional scoring — it's a supplement for the underbanked who do have a smartwatch but no credit history. In Brazil, about 45 million adults are unbanked or underbanked, but smartphone penetration is over 80 percent, and wearable adoption is growing fast — about 12 percent of adults now use a smartwatch or fitness band. Luna: That's 5.4 million potential new creditworthy customers who have no FICO score but have a data trail from their wrist. So the addressable market is real. What about privacy? Are users opting in, or is it buried in terms of service? Lucas: Nubank uses an explicit opt-in screen in the app — you have to grant permission for the wearable data to be used. They also let you see exactly which data points are being pulled, and you can revoke access at any time. But the bigger question is: once the genie is out of the bottle, will other fintechs follow the same standards? There's no regulation yet specific to wearable credit scoring. Luna: The Consumer Financial Protection Bureau in the US hasn't issued guidance on this. And in Brazil, the central bank's open banking framework technically covers payment initiation data, but wearables sit in a gray zone — is it payment data or biometric data? Lucas: Exactly. And the classification matters for what consumer protections apply. If it's payment data, it falls under existing financial privacy rules. If it's biometric, you might have different consent requirements. Some legal scholars argue it could be both. Nubank is treating it as payment data, which means they have to comply with the Brazilian data protection law, the LGPD. Luna: Let's talk about the technology side. How does the data actually flow from the watch to Nubank's credit model? Is it through Apple Pay's API, or something else? Lucas: It's through the Nubank app itself. When you enable the feature, the app uses the HealthKit and Wallet APIs on iOS — with user permission — to read transaction metadata from the watch. Not the full merchant name, just category, amount, timestamp, and device ID. That metadata is then hashed and sent to Nubank's model, which runs a machine learning algorithm trained on historical repayment data from their existing credit card portfolio. Luna: So it's not just 'you use tap to pay, you get a good score.' The model is comparing your patterns to patterns of people who either did or didn't default. That's a fairer approach than a simple threshold. Lucas: Right. And they've shared some of the feature importance weights. The most predictive feature is what they call 'merchant diversity' — how many different types of merchants you visit in a week. More diversity correlates with lower default risk. Think about it: someone who only goes to one gas station and one grocery store may be cash-constrained. Someone who visits seven different categories — coffee shops, restaurants, transit, retail — tends to have more financial flexibility. Luna: That's counterintuitive. I would have thought the person who sticks to a few merchants is more disciplined. But the data says the opposite — interesting. Lucas: And they also track something called 'transaction regularity' — are your purchases spread evenly throughout the week, or do they cluster on payday? Clustering on payday actually predicts higher default risk, which makes sense if you're living paycheck to paycheck. Luna: So the wearable is capturing behavioral signals that a credit bureau never sees. But what about the hardware limitation? Not everyone has an Apple Watch. Android wearables have different APIs, and many budget fitness trackers don't support payments at all. Lucas: That's a real equity issue. Currently the pilot is iOS-only because the Wallet API is more mature. Nubank says they'll expand to Wear OS by the end of 2026, but that still leaves out users with cheaper fitness bands from Xiaomi or Amazfit, which don't support NFC payments. So the people most likely to benefit from alternative scoring — the unbanked with low incomes — are least likely to own a device that generates the data. Luna: Which means this could actually widen the gap if it becomes mainstream. The people who get scored are the ones who already have a certain level of financial inclusion. Nubank is aware of this — they've partnered with a Brazilian retailer to offer a subsidized smartwatch for users who open a savings account, but it's early. Lucas: Let's look at another player. In the US, a startup called CrediFit — I think we mentioned them in episode 89 — has been testing a model that uses Fitbit data, including step count and sleep patterns, correlated with spending data from linked bank accounts. Their thesis is that consistent sleep and exercise habits predict lower credit risk. They published a paper in February 2026 showing a 15 percent improvement in default prediction accuracy over traditional models. Luna: But that's a huge privacy leap. Now you're not just sharing payment metadata, you're sharing your sleep and activity data. That feels different from 'I bought coffee.' Lucas: And CrediFit has been criticized for it. They defend it by saying the data is aggregated and anonymized, but the correlation is strong enough that critics argue it's effectively a health-based underwriting criterion, which could violate the Americans with Disabilities Act in some interpretations. The EEOC hasn't ruled on it, but it's a live legal question. Luna: So we have two approaches: Nubank's narrower, transaction-only model, and CrediFit's broader, health-inclusive model. Which one is likely to become the standard? Lucas: I think the market will push toward Nubank's approach because it's less legally risky and more defensible. Transaction metadata is clearly financial behavior. Sleep data is not. Regulators in both the US and EU are watching CrediFit closely, and I suspect they'll face restrictions before they can scale. Nubank's model is already live and hasn't drawn a regulatory challenge yet. Luna: There's also the question of accuracy over time. The Nubank pilot has only been running for three months. We don't know how well the model performs through a full economic cycle — say, a recession where spending patterns change dramatically. Lucas: Exactly. The model was trained on data from a period of relatively stable growth in Brazil. If unemployment spikes, someone who previously had high transaction frequency might suddenly stop spending. The model might flag that as a risk increase, but it could also overreact. Nubank told me they're planning to retrain the model quarterly and include macroeconomic features, but it's an open question whether wearable data is robust through a downturn. Luna: Let's zoom out. What does this mean for the broader fintech landscape? Are we heading toward a future where your watch decides your loan terms? Lucas: I think it's more nuanced. Wearable data will become one input among many in a multi-layered scoring system. We're already seeing credit card issuers use bank transaction data, rent payments, even utility bills. Adding wearable data is the next logical step. But I don't think it replaces traditional credit scores — it fills gaps. The early adopters will be fintechs targeting the underbanked, where every additional signal helps. Luna: And the incumbents? Are JPMorgan or Wells Fargo looking at this? Lucas: They are. Chase has a small team exploring wearable data for pre-approved credit line increases, but they're moving slowly due to compliance concerns. The big banks have more to lose if a model goes wrong. Fintechs can move faster because they're smaller and often have a more tolerant customer base. But once the model is proven, the incumbents will adopt it. Luna: So what's the one concrete thing our listeners should take away from this episode? Lucas: That your spending patterns, captured by a device on your wrist, are becoming a financial identity signal. If you own a wearable and use tap to pay, you're generating data that could help you get credit — or hurt you if the model is biased. The key is to understand what data you're sharing and with whom. It's a tool, not a magic bullet. Luna: And for builders and operators listening: the privacy-first, opt-in, transparent approach Nubank is taking is probably the right blueprint. The next big opportunity is figuring out how to bring in users who don't have the high-end hardware. That's where the real financial inclusion impact lies. Lucas: Alright, that's a wrap for this episode. We'll be back tomorrow with a deep dive into regulatory sandboxes for ai driven lending. See you then.