Mortgage File Assistant.

The problem I’d investigate

Imagine a borrower emails a bank statement after their processor opens a request. The attachment is in the inbox, but the file checklist still says “missing.” A colleague picking up the file has to reconstruct what happened. Send another request and the borrower repeats work. Mark the request complete too soon and an incomplete file moves to the next reviewer.

I’d want to know how often that mismatch happens, where it starts, and how much time the team spends resolving it. If the checklist simply updates too slowly, fixing that handoff may solve more than adding an assistant.

What a useful pilot would have to prove

These are proposed targets. I’d agree on them with the processing lead after measuring the team’s current workflow.

Processor time. Aim for a 20% reduction in median hands-on minutes per file, from opening the document review to recording the next action. Include verification and corrections so time spent checking a suggestion counts.

Borrower effort. Aim for 30% fewer duplicate requests per 100 files reviewed. Count a request as a duplicate when the lender already held a usable copy at the time of outreach.

Review quality. Keep the share of files handed off with a required document missing or unusable at or below baseline. Have a second reviewer check a sample from both workflows against the same checklist.

How I’d approach deployment

First, follow the file. I’d sit with processors and the people who receive their handoffs. We’d trace recent files, including ones that went smoothly, and distinguish time spent searching from time spent waiting for the borrower. That tells us which delays a change inside the lender can actually address.

Then, narrow the pilot. If finding documents is a recurring bottleneck, I’d start with bank statements on one team. At the point where a processor would send a request, the assistant would surface the likely document, its source, and the statement period. The processor would decide whether it satisfies the request, needs clarification, or warrants follow-up.

Compare before expanding. I’d first run the suggestions alongside completed reviews to find cases where a plausible match was the wrong month or an unreadable copy. Then I’d compare assisted and ordinary reviews on similar files in the same period. I’d expand to another document type only if both effort targets were met and handoff quality held. If reviews got faster but borrowers still received repeat requests, I’d revisit how the checklist gets updated before widening the pilot.