Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use Real-Time Travel Data for Dynamic Card Offers
Transcript
- Lucas: If these conversations are useful for what you're building or running, I want to start today with a number that made me stop: Brex disclosed in their 2025 year-end review that their cardholders spent four point two billion dollars on travel alone. Luna: Four point two billion. That's a lot of flight and hotel transactions flowing through one card program. Lucas: Exactly. And what caught my attention isn't just the volume — it's what Brex does with that data in real time. Every time a cardholder books a flight or checks into a hotel, that transaction stream feeds into a model that can adjust their card offers, their credit limit, even their rewards rate, all while they're still at the gate. Luna: So the card is learning your itinerary as you travel, and reshaping itself around what you're about to spend. Lucas: Right. And it's not just Brex. Ramp does something similar with their travel booking product. When an employee books a trip through Ramp's portal, the system sees the hotel name, the check-in date, the room rate — and it can dynamically set a per-diem limit or offer a cash-back bonus on that specific hotel chain, all before the employee swipes the card. Luna: That's a big shift from the old model where you get a generic two percent back on everything and call it a day. Lucas: Huge shift. And the technical pipeline is actually pretty straightforward once you look under the hood. The card networks — Visa, Mastercard — they already transmit merchant category codes, or MCCs, with every transaction. An MCC for 'airline' is 4511, for 'hotel' it's 7011. The fintechs just layer a rules engine on top of that stream. Luna: So the second the transaction hits the processor, the fintech knows you're at an airport, not a grocery store. Lucas: Exactly. And then they can take action. Say you're a Brex cardholder and you just bought a round-trip on Delta. The system sees MCC 4511, looks up your historical spend pattern, and predicts you'll probably need a rental car and a hotel in the next 48 hours. So it might preemptively raise your credit limit by 20 percent for those categories, or offer a five percent rebate on Hertz bookings made through the Brex travel portal. Luna: And all of that happens before you even land. The machine learning model is basically forecasting your next two purchases based on the first one. Lucas: That's the core insight. Brex's CTO mentioned in a fintech conference last year that their model can predict the next merchant category a traveler will spend in with about 85 percent accuracy within a 15-minute window after the first transaction. Luna: Eighty-five percent is surprisingly high. I would have guessed closer to 60 or 70. Lucas: It's high because travel has such a predictable sequence — flight first, then ground transport, then hotel, then meals. The model learns that pattern from millions of trips. And once it's confident, it can dynamically adjust things like the APR on a carry balance, or waive foreign transaction fees for that specific trip. Luna: Let me push back a little. If the card is adjusting limits and offers based on where I am and what I'm about to buy, am I being surveilled every time I travel? Lucas: It's a fair question. The data is technically transaction-level, not location-level — the fintech doesn't see your GPS coordinates, just the MCC and the merchant name. But in practice, if you buy a coffee at a specific airport Starbucks, the merchant name plus the timestamp gives a pretty good picture of where you are. Luna: So the line between transaction surveillance and location surveillance starts to blur. Lucas: It does. And that's something regulators are starting to look at. The CFPB has been asking questions about how real-time spending data is used for dynamic underwriting, especially when it triggers a credit limit increase that the cardholder didn't explicitly request. Luna: If today was actually useful to you, the way these stay ad-free is listener support — buy me a coffee dot com slash fexingo. Lucas: Yeah, and it's a small thing that keeps us from having to run ads or chase sponsors. If you've gotten value from the deep dives, that's the link. Luna: Alright. Back to the mechanics — you mentioned the model predicting a rental car after a flight. How fast does the system actually update the cardholder's available credit? Lucas: In Brex's case, it's sub-second. The authorization request hits their processor, the rules engine checks the MCC against the traveler's profile, and if it decides to adjust the limit, the new limit is reflected before the point of sale terminal even prints the receipt. Luna: So the traveler never sees the adjustment happening. They just swipe and it works. Lucas: Exactly. And that's the user experience goal — frictionless but personalized. Ramp takes a slightly different approach. Instead of adjusting limits automatically, they surface an offer in the app. So when an employee books a hotel, the Ramp app might show a notification: 'Book this car rental through our portal and get three percent cash back.' The employee has to opt in. Luna: That feels more transparent. But it also means Ramp is relying on the employee to take action, which probably reduces the conversion rate. Lucas: It does. Brex's automatic adjustments likely drive higher spend on their platform because the cardholder doesn't have to think. But Ramp's opt-in model is more compliant with current regulations around unsolicited credit line increases. Luna: Speaking of compliance, the travel data also feeds into fraud detection in a really interesting way. If my card is used for a flight from New York to London, and then 20 minutes later there's a transaction at a coffee shop in New York, the model knows that's impossible and flags it. Lucas: Right. That's a classic use case — geolocation via transaction sequencing. But the real innovation is on the rewards side. Some fintechs are starting to use real-time travel data to dynamically adjust the rewards rate based on supply and demand. Luna: Like surge pricing for cash back? Lucas: Exactly. If the system sees that a lot of cardholders are booking flights to Miami this weekend, the model might lower the rewards rate on Miami hotels because demand is already high. But if there's a hotel in Cleveland with low occupancy, it might boost the cash-back offer to steer travelers there. Luna: So the fintech is basically acting like a revenue management system for travel spend, using their own card data as the demand signal. Lucas: That's a perfect way to put it. And the data advantage is enormous. Traditional travel companies like Expedia or Booking.com see booking data, but they don't see the actual card transaction. Fintechs that issue cards see both — the booking if it's through their portal, and the settlement when the charge posts. Luna: That double view gives them a closed-loop data set that no one else has. Lucas: And that closed loop lets them do things like offer a guaranteed best rate on a hotel room because they can see exactly what you paid last time. Or they can offer to price-match a competitor's rate in real time, using their own transaction history as proof. Luna: I want to go back to the four point two billion figure for a second. Brex's travel spend grew 40 percent year-over-year in 2025. What's driving that growth? Lucas: A lot of it is the shift from managed travel programs to card-based spend. Companies used to have travel agencies book everything. Now employees just book on their corporate card and the fintech handles the policy compliance in the background. The real-time data makes it possible to enforce spending rules without pre-approval. Luna: So the card becomes the travel policy. Lucas: Exactly. If your company's policy says no first-class tickets over $500, the card can decline a $600 first-class fare at the point of sale, based on the merchant category code and the transaction amount. The employee gets an instant decline and a message explaining why. Luna: That's a much better experience than submitting an expense report and getting rejected two weeks later. Lucas: Much better. And it's not just about enforcement — it's about offering alternatives. Some fintechs are experimenting with dynamic policy exceptions. If the system sees that all economy seats on a particular flight are sold out, it might temporarily raise the first-class limit to $700 so the employee can still book that flight. Luna: That's a smart use of real-time inventory data. The card is basically saying 'I see the market conditions, and I'll adjust the rule for this one transaction.' Lucas: And that kind of flexibility builds trust with the user. The employee feels like the card is on their side, not just enforcing arbitrary rules. Luna: Let me ask about the data privacy angle again, because I think it's important. If a fintech is tracking every travel transaction in real time and using it to predict behavior, what happens to that data after the trip? Lucas: Most fintechs keep the transaction data for compliance and underwriting purposes, but they aggregate it for modeling. Brex, for example, says they use de-identified travel spend data to train their credit models, but they don't sell individual transaction data to third parties. That said, the data is still stored and could be subpoenaed or used in ways the cardholder didn't anticipate. Luna: So the trade-off is clear: you get a card that adapts to your trip in real time, but you're generating a permanent digital trail of everywhere you go. Lucas: That's the trade-off. And I think as more fintechs move into travel, we'll see regulators push for clearer disclosure. Right now, most cardholder agreements mention data usage in broad terms, but they don't say 'we will use your flight purchase to predict your next hotel booking and adjust your credit limit accordingly.' Luna: That level of specificity would probably scare some users, but it would also build trust. Lucas: It would. And the fintechs that get ahead of that disclosure — that explain the value proposition clearly — are the ones that will win long-term. Because the technology is undeniably useful. Getting a higher credit limit exactly when you need it, or a targeted cash-back offer on a hotel you were going to book anyway — that's tangible value. Luna: So where do you see this going in the next 12 months? Are we going to see every corporate card offering dynamic travel rewards? Lucas: I think so, but the winners will be the ones with the best data pipelines. The technical challenge isn't the model — it's ingesting the transaction stream fast enough and linking it to the cardholder's profile in real time. That requires a modern card processor, not the legacy systems that most big banks still use. Luna: So the incumbents like Amex and Chase have the data, but they're slowed down by their own infrastructure. Lucas: Exactly. Amex processes transactions on a mainframe that was built in the 1980s. They can do batch analytics overnight, but real-time decisioning on a per-transaction basis is hard. Fintechs built on modern cloud stacks — Brex on AWS, Ramp on Google Cloud — can run a machine learning model on every single authorization in under 50 milliseconds. Luna: That speed advantage is a real moat. Even if Amex wanted to copy the feature, they'd have to rebuild their core processing system first. Lucas: And that's a multi-year, multi billion dollar project. So the window for fintechs to own travel spend is wide open right now. Luna: One more thing — does this dynamic offer approach work for personal cards, or is it primarily a business card play? Lucas: So far it's mostly business cards, because business travel has predictable patterns and higher spend per transaction. But I've seen some consumer fintechs experimenting with it. For example, a card that sees you bought a concert ticket and then offers you a discount on a ride-share to the venue. Luna: That's essentially the same model, just applied to entertainment instead of travel. Lucas: Exactly. The underlying architecture — real-time MCC detection, predictive model, dynamic offer engine — is the same. The domain just changes. So I think we'll see a lot of cross-pollination over the next few years. Luna: It feels like we're moving toward a world where every swipe teaches the card something new, and the card reshapes itself around your next move. Lucas: That's the direction. And if done right, with transparency and user control, it could make credit cards genuinely smarter — not just a piece of plastic with a fixed APR and a rewards brochure. Luna: I'd love to see a 'data dashboard' where cardholders can see which predictions the model made about them and opt out of specific categories. Lucas: That would be a great feature. Some fintechs are already building consumer-facing ML explainability tools. It's early, but the conversation is shifting from 'can we do this?' to 'should we, and how do we do it responsibly?' Luna: And that's the kind of conversation that makes this space so interesting to follow. Lucas: It is. And we'll keep tracking it as the technology and the regulation evolve.