We walked into lenders and said the same thing every time. “Put AI into your collections. We’ll recover more, faster, than anyone.”
Most doors stayed shut. One opened.
So we sat next to the staff and asked questions. A lot of them.
“If they keep ignoring calls, what do you do next?”
“Check the contract and the payment history.”“Can you seize right away once you have a judgment?”
“It depends on the case.”“How do you tell which claims are undisputed?”
“You just know.”
We wrote everything down and started building an agent. Then we hit the wall. The answers that mattered most were the ones nobody could put into words. When to call again. Which asset is worth the trouble. When a promise is real and when it’s just a way to end the call.
That kind of knowledge doesn’t transfer through interviews. It lives in people who have done the job for years.
So we asked a different question.
“We’d like to work here.”
They said yes. We spent our days collecting debts. Calling people who didn’t want to be called. Reading case files that had been sitting on shelves for years. Filing the paperwork that turns a broken promise into a payment order.
Every day at the desk we learned something no interview had given us. Every night we turned it into a rule the agent could follow.
That loop is how Recova got built. Tacit knowledge in, software out.
It’s slower than building from the outside. It’s also the only way we found to build something the people at that desk would actually trust.