Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintech Is Rebuilding Credit Scoring From Scratch
Transcript
- Lucas: So the FICO score turns 37 years old this year. And for most of that time, it has been the single gatekeeper for whether you get a mortgage, a car loan, or even an apartment. But a quiet shift is happening—fintechs are rebuilding credit scoring from scratch using data that never touched a credit bureau. Luna: We're talking about things like your rent payments, your utility bills, maybe even the cash flow in your bank account? Lucas: Exactly. And the driver here is pretty straightforward: roughly 50 million Americans are 'credit invisible' or have thin files—meaning they don't have enough traditional credit history to generate a reliable FICO score. That's a huge market that traditional banks have largely ignored. Luna: But FICO has been the standard for decades. What changed to make alternative data viable now? Lucas: Two big things. First, open banking—through standards like Plaid and FDX—has made it possible for lenders to get permissioned access to a consumer's transaction data in seconds. That's cash flow visibility. Second, machine learning models have gotten good enough to find predictive signals in that chaotic data without overfitting. Luna: So instead of asking 'have you repaid a credit card before,' you're asking 'does this person have steady income and pay their rent on time?' Lucas: Right. And that turns out to be a better predictor for a lot of borrowers. Let's take Petal as an example—the credit card company. They use cash flow analysis and machine learning to approve people with no credit score at all. Their loss rates have been comparable to industry averages, which is pretty remarkable for a thin-file portfolio. Luna: But isn't there a regulatory risk here? The CFPB has been looking at algorithmic credit models pretty closely. Lucas: They have. And I think that's actually a healthy tension. The Equal Credit Opportunity Act requires that any credit model be demonstrably not discriminatory. Alternative data can be a double-edged sword—it can include the previously excluded, but it can also encode bias if the training data is skewed. Luna: So what's the evidence so far? Are these models actually fairer? Lucas: There's some promising data. Upstart, which uses a model that considers education and job history alongside traditional data, has published analysis showing that their model approves 27 percent more borrowers at the same default rate compared to a traditional model. And they've found that their model doesn't disproportionately deny protected groups. Luna: That's a big claim. Does it hold up under scrutiny? Lucas: The CFPB did a no-action letter with Upstart a few years back, essentially saying 'we won't sue you while we study your model.' That's a signal that regulators are cautiously open. But there's still a lot of debate about whether some alternative data—like education or zip code—is a proxy for race or income. Luna: Let me play devil's advocate. If these models are so great, why haven't the big banks adopted them more widely? Lucas: That's the core question. Part of it is inertia—FICO is embedded in every loan origination system, every securitization, every regulatory report. Replacing that is like replacing the engine on a 747 mid-flight. But there's also genuine caution. A big bank that approves a thin-file borrower who defaults gets blamed. A big bank that denies a thin-file borrower who would have been fine? Nobody notices. Luna: So the incentive structure is skewed toward saying no. Lucas: Exactly. That's where the fintechs have an advantage. They were born with a different risk appetite and a different cost structure. They can afford to experiment with models that start with a 'yes' instead of a 'no.' Luna: I want to talk about one specific alternative data point that's been getting attention: rent payments. Most people pay rent every month, but it almost never shows up on their credit report. Lucas: Right. There are now services like Experian Boost and RentTrack that let consumers add positive rent and utility payments to their credit file. But it's voluntary—the consumer has to opt in. What's more interesting is what Tala does in emerging markets. They don't even look at a credit bureau. They look at smartphone data—app usage, call patterns, repayment of microloans within their own ecosystem. Luna: And that works? I mean, that seems like a leap of faith. Lucas: It works well enough that Tala has originated over $3 billion in loans across Kenya, India, Mexico, and the Philippines. Their default rates are under 5 percent. The key insight is that behavioral data—how consistently you charge your phone, how often you top up your airtime—turns out to be correlated with repayment reliability. Luna: That's wild. But bringing it back to the US, what's the single biggest bottleneck to widespread adoption of alternative credit scoring? Lucas: I think it's standardization and portability. Right now, if Petal builds a cash flow model, that model only works for Petal's borrowers. You can't take that creditworthiness with you to apply for a mortgage at Wells Fargo. The data is siloed. What's needed is a universal, permissioned data layer that any lender can access, and that's still years away. Luna: So we're in a transition period where fintechs are serving a niche, but the mainstream hasn't shifted. Lucas: Exactly. And that's actually a healthy place to be. It lets the models be tested and improved in a lower-risk environment before they scale. And I think the regulatory path is gradually being paved—the CFPB's Section 1033 rule on open banking is going to accelerate consumer access to their own data, which is the raw material for these models. Luna: If these conversations are useful for what you're building or running, there's a simple way to make sure they keep coming. Lucas: Yeah. A couple of dollars a month is genuinely what keeps these going—buy me a coffee dot com slash fexingo, if you've gotten something out of them. Luna: No pressure, but it makes a real difference. Appreciate everyone who chips in. Lucas: Alright, back to credit scoring. One other development I want to highlight is the role of the Fair Isaac Corporation itself—FICO has started offering its own alternative data product, FICO Score XD, which uses telecom and utility data. That's a big deal because it signals that the incumbent recognizes the market. Luna: So FICO is trying to eat its own lunch before someone else does it for them. Lucas: Exactly. But critics say FICO XD still relies on a lot of the same traditional credit data—it's more of a supplement than a revolution. Meanwhile, startups like Nova Credit are doing something different: they're enabling immigrants to bring their credit history from their home country. They partner with credit bureaus in Mexico, India, and the UK to translate scores into a us equivalent. Luna: That's a really elegant solution for a specific pain point. Instead of starting from zero, you import your track record. Lucas: And it works. American Express uses Nova Credit for their international card products. The approval rates for immigrants using that data are significantly higher than for those without it. That's a win-win—the lender gets a lower-risk borrower, and the consumer gets access to credit they deserve. Luna: So where do you see this heading in the next, say, three to five years? Lucas: I think we'll see a bifurcated system for a while. On one side, traditional lenders will slowly adopt alternative data for specific use cases—mortgages for gig workers, for example. On the other side, fintechs will keep pushing the envelope with machine learning models that use every scrap of data they legally can. Eventually, I suspect the CFPB will set a national standard for what alternative data is permissible and how models must be validated. Luna: So the wild west phase is going to give way to a more regulated environment. Lucas: That's my bet. But the direction of travel is clear: more inclusion, more data, and a move away from a single number that summarizes your entire financial life. The question is how quickly incumbents adapt—and whether the fintechs can build enough trust and scale to become the new incumbents.