In the last issue of Building Adeft, I recounted how Victor, a self-employed borrower I interviewed, finally closed a loan at the same bank that had previously turned him down, after months of pain and multiple lenders. His story is insightful because it shows two very different outcomes with the same bank, the same software tools, and the same tax returns. What changed? It was the person reading the file. This is what first made me question whether selling software to mortgage lenders was the right path for Adeft.
When I started exploring how agentic AI could drive efficiency in the mortgage industry, the obvious move was to sell to big lenders. I know how enterprises buy. About four years ago I served as the Chief of Staff for the leader of an 8,000 person global software organization. I saw the politics of AI committees, arcane procurement processes, stakeholder labyrinths, and how pilots came and went with few meaningful production rollouts. At my last startup I navigated the enterprise sales gauntlet with a major CPG brand, spending 6 months to get through a pilot, then a year of working through a design partnership, and finally rejoicing after the first annual renewal.
The issue isn’t the difficulty or length of the sales cycle. The real problem is what happens after all of the work selling into the enterprise, onboarding, and running the requisite change management program. The lender can get a little more efficient, the savings can stay on the lender’s books, and the self-employed borrower’s experience can feel about the same as it did before. As the software vendor, we wouldn’t control how those savings get used or how much additional attention a difficult file gets.
We want to truly transform the borrower’s experience, not just make a lender more profitable. The only way to do that is to own the outcome, not just the tool.
Software has gotten much cheaper to build. Agentic AI makes it faster and less expensive to build than ever before. Vendors and internal lender teams are cranking it out at an unprecedented rate. Industry insiders I know tell me about getting calls and DMs from AI startups almost daily. As more and more people build it, the software alone becomes harder to differentiate.
Tools for lenders are everywhere. An industry analyst told me that the average lender runs about as twice as many tech vendors than it did ten years ago. The industry knows it can, and should improve. There is a vibrant community of vendors trying to push the industry forward. Last week at an MBA event in Chicago it seemed like tech vendors outnumbered the lenders.
Tools can only do so much. The aforementioned industry analyst I met with said that the best processing automation he’s seen cuts about 20% of a cost bucket that’s about 20% of the cost to originate a loan, so roughly 4% overall. For lenders with thin margins, that is significant, but it doesn’t necessarily impact the cost to the borrower or how painful the experience feels. New tools on top of the same process won’t change the borrower’s experience enough.
A lack of attention is what fails the most complex files. In my last newsletter, Victor finally triumphed because the right person with the right amount of attention made themselves accountable to get his loan done. In my issue about Robert the processor, I highlighted the fact that there aren’t enough processors like Robert to go around.
That is why we take the seat instead of selling it. We believe to usher in a truly transformative experience we need to own the outcome instead of selling the tool. AI becomes the key to our cost structure instead of the product. The operational expert should run your loan and they should be on the hook to move it forward to closing.
The best argument against building a services business is that services businesses tend to have thinner margins than software. That has been the conventional wisdom because in most cases, it has proven to be true. We’re making a bet that we can turn this conventional wisdom on its head and one early test will be can we dramatically reduce the processor-minutes per closed file. We also need to show that fewer minutes translate to lower costs and a superior borrower experience.
We think one highly skilled licensed person can carry far more complex files than what is possible today as AI handles the grunt work. My co-founder and I are deeply embedded in these workflows now and we’re learning from the most skilled operators we can find. As soon as we can measure it, it will go on the scoreboard.
Scoreboard:
Interviews: ~43 | Design Partners 1 | Processor-minutes per closed file: coming soon
My prediction: within three years, the average lender will have fewer technology vendors than today, not more. Point tools will get absorbed into platforms and into the services that own the outcomes. Let’s follow up on this in October 2029.
Disagree? Reply with the best argument for selling software to lenders instead. I’ll share the best one.
More soon,
Jeremiah