Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI for Real-Time Property Valuation
Transcript
- Lucas: If these conversations are useful for what you're building or running — and you want to keep them ad-free — there's a simple way to help. We don't run ads on these shows, by design. The link to support that choice is buy me a coffee dot com slash fexingo. Luna: Yeah, it's a small thing that adds up. Keeps the content focused on what actually matters. Lucas: Exactly. So today — let's talk about property valuation. One of the last holdouts in the mortgage process that still takes weeks, and AI is finally making real inroads. Luna: I've been curious about this. Traditional appraisals — a human drives out, takes photos, compares three comps — can delay a closing by ten to fourteen days. And it's costly. Lucas: Right, anywhere from five hundred to a thousand dollars. But a new generation of fintech lenders is using machine learning to deliver a property value estimate in seconds, not days. They're called automated valuation models — AVMs. And they've been around in a basic form for years, but the accuracy has improved dramatically. Luna: What's changed? Is it just more data? Lucas: More data and better models. The older AVMs were essentially regression models on county tax assessments and a few recent sales. Today's models ingest multiple listing service data — every home listed, every price change — plus tax records, permit data, even satellite imagery. They can update valuations in near real-time as new comps come in. Luna: So the fintech angle — is a company like Better.com using this? Or is it more niche players? Lucas: Both. Bigger lenders are licensing AVMs from providers like CoreLogic or HouseCanary. But some fintechs have built their own. One I looked at — let's call them LendFast — trained a model on over ten million property records from twenty metro areas. They claim their AVM predicts sale price within five percent about ninety percent of the time. Luna: Five percent — so on a four-hundred-thousand-dollar home, that's a twenty-thousand-dollar swing. Is that good enough for underwriting? Lucas: For a conventional mortgage, no. Fannie Mae and Freddie Mac still require a full appraisal or a hybrid appraisal for most purchase loans. But for home equity lines of credit, or for refinances where the loan to value is low, some lenders are accepting AVMs as sufficient. And the GSEs are testing them too. Fannie Mae has a pilot where they accept AVMs for certain cash-out refinances. Luna: That's a big deal. If the government-sponsored enterprises start to trust these models, it could open the floodgates. But what about bias? Traditional appraisals have been criticized for undervaluing homes in minority neighborhoods. Does an AVM fix that or make it worse? Lucas: Great question. The models are only as unbiased as the training data. If historical sales data reflects systemic undervaluation — say, homes in Black neighborhoods consistently sold for less than comparable homes in white neighborhoods — the model learns that pattern. So there's a real risk of perpetuating bias. Some fintechs are addressing it by explicitly controlling for neighborhood demographics and running fairness audits. But it's an open challenge. Luna: And regulators are paying attention. The CFPB and FHFA have both issued guidance on model governance for AVMs. They want to see validation, ongoing monitoring, and fair lending testing. Lucas: Exactly. So it's not just a technical problem — it's a compliance problem. But the efficiency gains are hard to ignore. A traditional appraisal costs six hundred dollars and takes two weeks. An AVM costs maybe twenty dollars and takes two seconds. For a lender doing a thousand loans a month, that's huge. Luna: And for the homeowner — imagine wanting to tap your equity for a renovation, and instead of waiting weeks, you get an answer in minutes. That's the fintech promise. Lucas: Right. And some companies are already there. One startup, Valon, offers a home equity product that uses an AVM for the initial valuation, then does a physical appraisal only if the loan proceeds above a threshold. Their turnaround time from application to closing can be under a week. That's almost unheard of in mortgage lending. Luna: What about accuracy in volatile markets? We've seen wild swings in home prices over the past few years. How do these models handle a market that's dropping fast? Lucas: That's the stress test. Most AVMs are backward-looking — they rely on recent sales, which lag by thirty to sixty days. So in a fast-declining market, the model might overvalue a home because it's still using comps from two months ago. Some fintechs are adding real-time listing price changes and days-on-market data to try to anticipate direction. But it's not perfect. The regulators want lenders to stress-test their AVMs for scenarios like a ten percent market drop. Luna: So we're not at full replacement of human appraisers yet. But we're getting closer. For certain products, AI is already the primary valuation tool. Lucas: Absolutely. And I think the next frontier is integrating AVM outputs with other data — like employment income verified through payroll APIs, or bank transaction data — to create a fully automated underwrite. That's where fintech is headed. Luna: It's a fascinating shift. A process that has been manual and opaque for decades is becoming instant and data-driven. The key is doing it fairly. Lucas: Right. And that's the conversation we'll keep tracking. Thanks for listening.