Pave

A lender told me the lifespan of a dollar in a subprime borrower's account is about the lifespan of a mosquito. Any available funds gets spent within a couple of days. They’re currently at 85% ACH success and wants to get to 92%. Most teams try to solve this with a balance check before they pull. It's expensive and they still get NSFs, because the balance tells you what's in the account this morning. It says nothing about who else is pulling on Thursday, or whether the money will still be there when the ACH settles two days later. Whoever pulls first gets paid. Everyone behind them eats the return.

Did you miss our live session on the Q2 Consumer Health Index? Gambling was the hot topic. We got a ton of questions on it, so here's more. In a month where someone gambled, 60% also tapped a credit product. In months they didn't gamble, that number was 28%. That's across more than 10 million accounts. The cash advance gap is even sharper. 43% of gambling months included a cash advance, versus 15% in non-gambling months. This isn't proof that gambling causes credit use. But the two travel together, and closely. Sports betting and prediction markets have normalized wagering as everyday entertainment. Participation in our data climbed through June, right alongside the NBA Finals and the World Cup. Here's what matters for anyone underwriting these borrowers. Gambling activity and liquidity pressure show up in the same months. If you're working off a credit score alone, you're missing a signal sitting in plain view in the transaction data.

A team built their own cashflow underwriting model. There was some lift. They fed bank transaction data into the risk model they already had. Approvals went up. They started extending credit to people they used to decline, and did it with more confidence, resulting in new volume from borrowers they'd been turning away. Then I asked the engineer who built it what happens next. His answer stuck with me. The proof of concept was done. The hard part was scaling it. And the team wasn't staffed to do that. This is the pattern I see everywhere. Building a model that works once is the easy part. One engineer, one dataset, a few weeks of focus. Making it hold up is a different job. Every borrower, every risk event, every month as spending patterns drift, the model has to keep performing. That takes a team whose whole job is watching that one model. Most lenders don't have that team. They have smart engineers who built version one between three other priorities, and no capacity to maintain it. So the model that proved its value goes stale. Not because anyone was wrong. Because there was no one left to keep it sharp. Most good teams can build a good enough V1. Far fewer can keep it performing after the demo. That gap is where the lift quietly leaks back out.

A lender once told me his own team wouldn't adopt a tool that would obviously make them better at their jobs. The reason had nothing to do with performance. They'd built the current system themselves. Bringing in something bought felt like admitting the thing they built could be replaced. So they kept the weaker version, simply because it was theirs. I think about this more than I expected to. The best risk infrastructure in the world sits unused if the people who'd benefit feel threatened by it. Adoption is a human problem long before it's a technical one. Something I've learned watching these decisions play out. When you're evaluating a model and the loudest objection in the room comes from the person who built the old one, pay attention to where that objection is really coming from.

In a few hours, we go live with the Q2 Consumer Financial Health Index. Jordan Casey and Will Xu are walking through what 10M+ households actually did with their money last quarter. On the agenda: → Why gas was the only essential taking a bigger bite of the wallet, and why the inflation it drove never spread to anything else. → How paychecks held the quarter together, even as the cushion behind them thinned to a two-year low. → The widening K-shape: high credit utilization rose 17.6 points among the lowest cash-flow households, and just 4.4 among the highest. → Why gambling now travels alongside cash advances, and who ends up carrying the cost. → And the signals we're watching most closely heading into Q3. Roughly 10 minutes of findings, then the floor is yours for questions on the data, the methodology, or how it maps to your own book. Final hours to register: 🗓️ Today, July 31 - 11am PT / 2pm ET 🕚 35 minutes 🙋 Most of it open Q&A Register now:

American paychecks are holding steady. The savings cushion behind them just hit a two-year low. That's the Q2 read from the Atlas x Pave Consumer Financial Health Index, built on real-time behavioral data across 10M+ accounts. Income held for the sixth straight quarter. The buffer behind it thinned for the sixth straight quarter. The sharpest signal is at the bottom. Among the lowest cashflow households, the share running above 90% credit utilization rose 17.6 points over the year. Among the highest, it rose 4.4. Same economy. Two different currents. Jordan Casey and William Xu are walking through the full Q2 findings live on Friday, July 31, then opening the floor for questions. Registration link in the comments. We hope you can join us!

We just completed our FIFTH consecutive SOC 2 Type 2. Almost any control can look healthy for the few months an auditor is watching. The harder thing is holding the same standard while the company underneath keeps changing. Over the past year we launched new products, onboarded enterprise customers, and access management, monitoring, and incident response passed at the same bar the whole way. For companies routing sensitive financial data through us, that's a five-year independently verified track record they can hand straight to their banks and compliance teams. If you're interested, you can read more here:

A lender asked us to help them cut 200 basis points off the APR they charge borrowers. We never touched their model. They were collecting repayments over debit. Every pull cost them 1.6% plus a flat 28 cent attempt fee, win or lose. Move that same repayment to ACH and the cost drops to single-digit cents. On a $1,000 payment, every month, for the life of the loan, that's the difference between a real chunk of the payment going to a card network and staying in the deal. Cut that cost and you're not choosing between your margin and what you charge the borrower. You can move one without touching the other. So why isn't everyone on ACH already? Returns. ACH takes a few days to settle. A borrower can have $1,400 in the account today and still bounce a $1,000 pull, because their $900 rent clears Thursday. A balance check at initiation tells you nothing about the balance at settlement. But recurring payments are predictable. Rent lands the same week every month. Paychecks hit on a cadence. If you can see the transaction history, you can project the balance forward to settlement instead of guessing off today's snapshot. The question changes from "do they have the money today" to "will it still be there Thursday." One lender we work with is moving from 10% of repayments on ACH to 90%, and dropping live balance checks entirely. That's a lower borrower APR hiding in a payment rail.

I think the next hiring boom is going to be wild. Every purge in Silicon Valley has set up a more aggressive boom than the one before. My dad spent 35 years at AMD through an acquisition of Monolithic Memories where they built the first programmable logic devices. Everyone thought that was going to die. He took unpaid Fridays, worked weekends, watched colleagues leave. Decades later, AMD is one of the most valuable chip companies in the world, riding the AI wave that few saw coming. I've watched this cycle play out my whole life. Families packing up for Austin or Phoenix when things go sideways. Then a wave of hiring comes roaring back and pulls them home, or brings a whole new generation in. Every new technology shift is very uncomfortable and upsetting for many people. But I hope that people who are affected by layoffs have faith and prepare for all of the new opportunities that are coming. This next cycle is inevitable. AI doesn't just lower the cost of building, it raises the stakes for not building. If your competitor ships 10x faster and you can't, you lose the customer. This forcing function is going to lead to a crazy hiring wave and so much creativity in launching so many new things

Cash flow underwriting has an adverse selection problem. Or so I keep hearing. A data vendor said it confidently on a call this week: "Even having the person click a link to give you access to their bank account is going to cause the good people to drop off." I've seen the data. It doesn't hold up. Who connects their bank account without hesitation? Someone with nothing to hide. Completion rates are highest among borrowers with stronger cash flow. The people who abandon aren't your best risks. They're usually your worst. Adverse selection in cash flow underwriting is a model problem, not a funnel problem. If your score can't do anything with the data once you have it, blame the model. The friction isn't filtering out good borrowers. It's filtering out bad ones.