Figure is turning abandoned home-equity applications into loans. That's cheap growth, and a credit question.

Every lender knows the cheapest loan it will ever make is one that is already half-done. Figure, the home-equity fintech, is turning that intuition into a product. It is working with Sierra — the AI-agent company founded by Bret Taylor — on agents that chase homeowners who started a home-equity application and then vanished, and steer them to a funded loan.

The headline reads like a productivity footnote. It is not. Figure's stock is up more than a fifth over the past month, and this is the story investors are attaching to the move. To see whether it deserves the attention, it helps to stop calling the abandoned application a pile of dead leads and ask what it really is.

Industry estimates put online loan abandonment above 70 percent, and higher still for mortgages. In home equity specifically, average drop-off runs from 50 to 68 percent — well over half the people who begin never finish. For a lender whose revenue depends on funded-loan volume, that is the single largest source of lost lending revenue there is, bigger than rate competition and bigger than any marketing gap. Money was already spent to bring those borrowers in; the loan is sitting partway through the funnel. Recovery costs almost nothing to source.

So the logic is sound and the economics are real. The interesting question is which abandoned applications the agents recover, and what it costs to find out.

The mechanism is well documented, just not by Figure. Sierra's loan-officer agent is already live at Rocket Mortgage, where it has handled hundreds of thousands of chat conversations and makes over a million outbound calls a month. Rocket says clients who start with the assistant close at three times the rate of those who do not, and four times when an AI chat hands off to a banker. Sierra has also made the abandoned application its explicit research subject: it acquired Takeoff in part to understand why borrowers walk away, and it integrated its agents with Plaid so that instead of stalling the moment a loan needs bank data, the agent can pull it mid-conversation and keep the application moving.

That last detail is where the real business lives. Figure sells itself on speed — approval in five minutes, funding in as few as five days — but the mortgage side of the funnel is the part no one can make instant. Loans stall at the junctures where a human would normally take over: verifying income, pulling an appraisal, re-explaining a document request after three days of silence. That is the abandonment that most people mean, and it is the abandonment an agent can genuinely fix. Remove the friction and the borrower who always qualified closes; the lender books funded volume it was never going to get; both sides are better off.

But abandonment is not a single thing. Look at why the borrowers left and it splits into two piles. One pile is friction: a forgotten signature, a stalled verification, a form that took too long. The other pile is signal: a homeowner saw the offered rate and balked, could not confirm they qualified, or was comparison-shopping and let a competitor win. Those are different animals. An agent that un-stalls the first pile is recapturing borrowers who always wanted the loan. An agent that pushes the second pile across the finish line is converting a borrower who already looked at the deal and decided it did not fit into a funded loan anyway.

Figure's own chief executive has stated the risk almost exactly. He argued that lenders cannot "AI their way into Triple-A securitization ratings" and warned that AI risks making bad processes faster. That is not a hypothetical caution for this company. Figure has repositioned itself from a HELOC lender into what it calls a blockchain-native capital marketplace — it originated about $4.3 billion of marketplace volume in Q2, up 132% year over year, and its loan-trading platform grew 262 percent to $2.8 billion. The more loans it funds, the more it can sell and trade, which is what turned net revenue of $218 million into a 95 percent year-over-year increase and net income into a near-tripling. In that model, every recovered application is inventory for the marketplace, and the incentive to chase reluctant borrowers is strong.

This is why the abandoned-application story is worth watching as a credit question rather than a growth one. There is a clean falsifiable test buried in Figure's operating data. If the agents are recovering friction-based abandonment, the recovered loans should convert at high rates and perform like the rest of the book. If they are recovering signal-based abandonment, the marginal funded loan will be the lowest-visibility credit risk in the company — a borrower who walked away and got talked back.

None of that is visible in the price. Figure trades at roughly 35 times trailing earnings and eleven times sales, and it reached a 52-week high of $78 before this month's level near $37. The market is already paying for years of fast, profitable growth. Whether that growth compounds or merely accumulates depends on the one thing the announcement does not say: which abandoned applications the agents are actually funding, and how those loans perform next year. The cheapest loan in lending is also, for the borrower who left a reason, the riskiest one to want back.