Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI for Real-Time Identity Verification
Transcript
- Lucas: So we all know the feeling — you sign up for a new fintech app, and within seconds it asks you to take a picture of your driver's license and then a selfie. That moment, right there, is where AI is doing some genuinely impressive work. Luna: It feels almost magical when it works fast. But I've also had times where it rejected my passport three times for weird lighting. So what's actually happening under the hood? Lucas: Great question. Let's talk about real-time identity verification — the AI systems that fintechs use to confirm you are who you say you are, within seconds, without a human in the loop. It's become the gatekeeper for everything from opening a bank account to getting a loan. Luna: And it's a huge market. Grand View Research put the global digital identity verification market at around $13 billion in 2025, growing at over 15 percent CAGR. Lucas: Right. And the key shift is from static checks — like uploading a scan and having a person look at it hours later — to real-time, ai driven decisions. The poster child here is probably Stripe's Identity product. They claim they can verify a user in under 3 seconds with a 98 percent pass rate for legitimate users. Luna: That 98 percent is important because the fear is always false positives — rejecting real customers. How do they get there? Lucas: It's a multi-layered AI pipeline. First, document verification: the AI checks the passport or license for tampering — holograms, font consistency, even the alignment of the photo. Then liveness detection: the selfie isn't just a face match; the AI looks for subtle cues that you're a real person, not a photo or a deepfake. Luna: Liveness detection is where the arms race is hottest. Deepfakes are getting scary good. Lucas: Exactly. Companies like Jumio and Onfido are investing heavily in passive liveness — meaning you don't have to blink or turn your head. The AI analyzes micro-movements, skin texture, even the way light reflects off your eyes. It's looking for things that a video replay or a 3D mask can't replicate. Luna: And this is all happening in milliseconds, on a phone, often with low bandwidth. Lucas: That's the engineering challenge. You're running neural networks on-device or at the edge to keep latency under a second. If the verification takes too long, users drop off. Stripe found that every extra second of verification time increases abandonment by something like 2 to 3 percent. Luna: So speed is a product feature. But what about the deepfake problem specifically? I've seen demos where ai generated faces can fool older systems. Lucas: It's a constant cat and mouse. Jumio publishes transparency reports showing they block about 8 to 10 percent of attempts as suspected fraud — a mix of stolen IDs, synthetic identities, and deepfakes. The AI models are retrained weekly on new attack patterns. They also use ensemble methods — multiple models voting on the result — to reduce false accepts. Luna: And regulators are paying attention too. The FCA in the UK and FinCEN in the US have issued guidance on using AI for KYC, but they also require explainability. If you reject a customer, you have to be able to say why. Lucas: That's a tension. The most accurate models are deep neural networks — black boxes. So companies like Onfido have built what they call 'explainability layers' that highlight which parts of the document triggered a flag — say, a mismatched font on the date of birth. Luna: Let's zoom in on a specific use case: gig economy platforms. A driver for Uber or DoorDash needs to be verified quickly, often remotely, and the identity needs to be rechecked periodically. Lucas: Good one. Uber uses a mix of document verification and real-time selfie checks before drivers go online. The AI compares the selfie to the ID photo on file, and if it fails, the driver can't start working. That's a high-stakes decision — a false reject costs the driver money, but a false accept could let someone with a suspended license drive. Luna: And that's where the ROI for fintechs becomes clear. Better AI means fewer manual reviews. Jumio claims their AI reduces manual review by 60 percent, which directly cuts operational costs. Lucas: Let's talk about the biometric piece more. Many fintechs now use facial biometrics as an ongoing authentication method, not just at onboarding. So after you verify your identity once, every time you log in or make a high-value transaction, the app asks for a selfie. Luna: But isn't that more friction? Users hate extra steps. Lucas: They do, which is why the trend is toward passive authentication. Some apps now use behavioral biometrics — how you hold your phone, your typing cadence, the way you swipe — to continuously verify you without asking for anything. That's still early, but companies like BioCatch are already doing it for banks. Luna: So where does this go next? I'm hearing about decentralized identity — where your verified identity lives on your phone as a digital credential, and you share only what's needed. Lucas: That's the holy grail. The W3C's Verifiable Credentials standard is gaining traction. Imagine a fintech app asks 'Are you over 18 and a US resident?' and your phone proves it without revealing your name or address. That would reduce data breach risk and improve privacy. Luna: Apple is pushing that with its identity verification framework. And some EU banks are piloting it under PSD2. Lucas: If this episode was useful to you — especially if you're building or running a fintech — listener support is what keeps these conversations ad-free and deep. You can find us at buy me a coffee dot com slash fexingo. Luna: Yeah, it's a simple way to keep the content independent. We appreciate everyone who chips in. Lucas: Back to the tech — the interesting frontier is liveness detection that doesn't even need a selfie. Some researchers are working on verifying identity from a short voice recording, analyzing vocal tract resonances as a biometric. That could be huge for accessibility. Luna: Fascinating. And it would work for people who can't use a camera easily. So identity verification is becoming more inclusive, not just faster. Lucas: Exactly. The end goal is trust that feels invisible. You get verified without thinking about it, and the fraudsters get blocked without the good users ever noticing. Luna: That's the vision. And for now, the AI is doing a pretty good job of making it real.