Here is a confession. I have spent 30 years routing lender decisions by gut. Not because I did not have data — I had more data than any analyst would know what to do with. Because the data I had was the wrong kind for the decision.
The bureau report tells you what the borrower looks like. The funding outcomes tell you what the lender looks like. And the lender changes faster than the borrower does.
The structure of the problem
Every month, each lender we submit to has a slightly different appetite. Capital One is tightening on sub-540 this quarter because their pool delinquency ticked up. Santander added a DTI cap that was not there 60 days ago. Chrysler Capital just expanded their used-vehicle window by three model years. None of this is in the bureau report. None of this is on any rate sheet a dealer has access to.
The way you learn it is by submitting deals and watching what comes back.
The way you apply it is by remembering what came back the last 40 times you submitted a deal that looked like the one in front of you.
The thing software keeps getting wrong
Every dealer-software vendor has taken a swing at lender routing. Most of them wrote rules engines. "If FICO is between X and Y, route to lender Z." That is a 2004 solution to a 2026 problem.
The correct structure is a pattern-recognition layer sitting on top of the submission history. You feed it every deal you submitted, every outcome (approved, declined, conditional, stipped, funded, rescinded), every stipulation the lender asked for, and the final gross after reserve haircuts. Then, for the deal on the desk, you ask it which lenders look most like the pattern of the ones that funded clean in the last 90 days on similar structure.
That is what Louie does. It is not magic. It is pattern recognition on an honest dataset the dealer already owns.
What an outside analyst misses
I have sat through a lot of pitches from fintech companies trying to sell me lender routing over the years. They all make the same mistake. They assume the goal is to pick the lender with the best rate, and they optimize for that.
That is not the job. The job is funding. A 2-point-better rate on a deal that stips for six days and then falls apart is worse than a 2-point-worse rate on a deal that funds in 24 hours.
This is the thing you cannot learn from outside the dealership. And it is why operator-built software is a different class of artifact than engineer-built software.
If your lender-routing product does not index on funding velocity as its primary output, it is solving the wrong problem. Rate is secondary. Funding speed is the business.
The strategic read
For an acquirer thinking about what the dealer-software intelligence layer should look like over the next five years: the AI routing layer I just described is the gate. Whoever builds it first, and builds it on the submission history of enough rooftops, owns a piece of moat that incumbents cannot ship.
They cannot ship it because their product orgs are run by people who have never watched a stip check come back at 6pm on a Friday with the bank closing at 7. And that is the moment when Louie's lender-routing matters.
If you are curious how the pattern layer actually works, the public moat endpoint has the macro anchors: GET /api/moat/public. The per-lender playbooks are in the data room under NDA.