Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI to Price Insurance in Real Time
Transcript
- Lucas: Luna, your car insurance premium right now — how was it determined? Luna: Uh, I assume some actuary looked at my age, my driving record, my ZIP code, and then multiplied by a factor for the color of my car or something. Lucas: Right — that's the traditional model. Static cohorts. You're grouped with people who share a few demographic traits, and you all pay roughly the same rate, regardless of how you actually drive. Lucas: But a growing batch of fintechs and insurtechs are flipping that. They're using AI to price insurance in real time, based on what you're doing right now — not what the actuary guessed you'd do over the next twelve months. Luna: So my premium could change month to month? Or even day to day? Lucas: Exactly. The term you'll hear is usage-based insurance — UBI. The market is projected to hit about $120 billion globally by 2028, and the core enabler is machine learning models that ingest streaming data from telematics devices, smartphones, even satellites. Luna: Satellites? For car insurance? Lucas: Not just car insurance. Think about property insurance. A startup called Tractable, based in the UK, uses computer vision to assess damage from photos — but the next layer is using satellite imagery to monitor roof condition, nearby wildfire risk, flood zones. All of that can feed a model that adjusts your homeowners premium in near real time. Lucas: Potentially. Or it goes down if you've installed fire-resistant roofing. The idea is to move from a single annual snapshot to a continuous risk score. Lemonade, for example, has been experimenting with what they call 'check-ins' — you submit a short video walkthrough of your apartment, their AI assesses risk factors, and your premium adjusts accordingly. Luna: That feels invasive. But also, I can see why it would attract better risks — people who know they're careful and want to pay less. Lucas: That's the retention story. Studies show UBI policies have retention rates 20 to 30 percent higher than traditional bundled plans. The reasoning is intuitive: if you're a cautious driver, you feel like you're getting a fair deal every month. You're not subsidizing the guy who texts at red lights. Lucas: Root Insurance, which went public a few years back, built its whole model around smartphone telematics. They measure acceleration, braking, cornering, phone distraction — all via the phone's sensors. Their early data showed that good drivers could save up to 52 percent compared to traditional carriers. Luna: Fifty-two percent? That's enough to make people actually want to share their driving data. Lucas: Right. The trade-off is privacy. And that's the friction point. In Europe, regulators are watching this closely under GDPR — you need explicit consent for each data stream, and the consumer has to be able to withdraw it. In the US, it's a state-by-state patchwork. California and Massachusetts have been the most aggressive about requiring transparency on what data is collected and how it's scored. Luna: So if I opt out of data sharing, do I just get the old actuarial rate? Lucas: Typically, yes. But the gap between the 'safe' rate and the 'opaque' rate is widening. Some carriers are starting to penalize non-participation — not explicitly, but by pricing the base premium higher so that the usage-based discount becomes almost mandatory if you want a competitive rate. Luna: That's a slippery slope. Let's talk about the actual ML models. What kind of algorithms are we talking about? Lucas: Mostly gradient boosting machines — XGBoost, LightGBM — because they handle tabular data well and are interpretable enough for regulators. Some are moving to neural networks for more complex pattern recognition, like combining telematics with weather data and traffic patterns. The key metric is loss ratio — they want to see that the model's predicted risk correlates tightly with actual claims. Lucas: One fascinating example: a company called Zego in the UK offers pay as you go insurance for delivery drivers. The premium is calculated per mile driven, adjusted for time of day, road type, and historical accident data for that specific route. Their model updates every 15 minutes. Luna: Every 15 minutes? That's not even real time, that's near real time. Lucas: Exactly. And it works because delivery drivers have highly variable risk — a Domino's driver on a Friday night in the rain is in a completely different risk bucket than that same driver on a Tuesday afternoon. Lucas: But here's the tension: accuracy versus trust. The more granular the pricing, the more consumers feel like they're being surveilled. A 2025 survey from Pew found that 61 percent of US adults said they would not be comfortable with insurers using their driving data to set rates, even if it meant lower premiums. Luna: So the fintechs have a marketing problem, not just a math problem. Lucas: I think that's exactly right. And that's where the smartest players are investing — not just in better models, but in better UX around consent and value communication. If you can show someone a dashboard that says 'Your safe braking saved you $12 this month,' that's a feature, not a bug. Luna: Honestly, if today's episode was worth a coffee to you, the link is buy me a coffee dot com slash fexingo. That kind of listener support is what keeps this show ad-free and independent. Lucas: Yeah, seriously. It's a small ask, but it makes a big difference for us. And we really appreciate it. Luna: So back to the models — what about the regulatory pushback in Europe? Has any UBI model been rejected outright? Lucas: Yes. In 2024, the Dutch data protection authority fined an insurer for using telematics data without adequate transparency about how the AI scored risk. The model was a black box — the company couldn't explain why a particular driver got a rate hike. Under GDPR, you have to provide meaningful information about the logic involved. Lucas: So now we're seeing a trend toward 'explainable AI' in insurance. Startups like Ladder and Clearcover are building models that output not just a premium but a breakdown of the top three factors driving that premium. That's a huge shift from the traditional actuarial black box. Luna: That actually builds trust. If I can see 'Your premium went up because you drove 200 miles in heavy rain last month,' I might nod and say, okay, that's fair. Lucas: Right. And the next step is dynamic adjustment. Some carriers are experimenting with 'nudges' — if your driving score drops, they'll send a notification suggesting you avoid driving during peak storm hours to keep your rate low. It gamifies risk reduction. Luna: So the insurer becomes a coach, not just a bill collector. Lucas: Exactly. And that's where the fintech DNA shines. Traditional insurers think in annual cycles. Fintechs think in sessions, events, actions. They're building products that feel more like a fitness app than an insurance policy. Luna: One thing I'm curious about — how do the reinsurers feel about this? They're the ones holding the ultimate risk. Lucas: Great question. Reinsurers like Swiss Re and Munich Re have been cautiously supportive because the data is better. If you can model risk with higher resolution, you can price it more accurately, which means fewer surprises. But they're also wary of adverse selection — if only the safest drivers opt in, the traditional pool gets sicker, and the industry could see a death spiral. Lucas: So some reinsurers are pushing for mandatory minimum data sharing — not usage-based, but basic telematics for all policies. That's a controversial idea, but it shows how fast the industry is moving. Luna: What about life insurance? Are we seeing real-time pricing there? Lucas: It's early, but yes. Vitality in South Africa pioneered the model where you earn premium discounts by hitting step goals, getting health screenings, etc. Their data shows that engaged members have 30 percent lower mortality. Now you see fintechs like Beam and Sureify offering similar programs in the US, using Apple Watch and Fitbit data to adjust term life premiums. Lucas: The challenge there is that life insurance is a long-duration product — you can't reprice every month easily. So the current approach is to offer a base premium with a monthly 'wellness credit' that can reduce your payment, but the base stays locked. Luna: So the real-time pricing is more of a rebate mechanism than a true repricing. Lucas: Exactly. But the tech stack is the same — streaming data, ML model, real-time decision engine. The difference is in how the carrier manages the liability. Luna: Looking ahead, what's the one thing that could accelerate adoption of real-time pricing across all insurance lines? Lucas: I think it's interoperability. Right now, every carrier has its own app, its own telematics device, its own data format. If we had a standard API for risk data — something like Open Banking but for insurance — then consumers could easily switch carriers and bring their driving history with them. That would create a competitive market where the best model wins. Lucas: And that's exactly the kind of infrastructure play that fintechs are good at. Not just building a better mousetrap, but building the rails so everyone can use it. Luna: Let's hope the regulators are listening. Lucas: They are. Some of them are even buying insurance from these companies. That's the ultimate feedback loop.