Skip to content

Dromi · Processing

The processing side reads the file for you.

A normal loan costs roughly 45 minutes of data entry and balancing plus another 30 chasing order-outs. Dromi reads the documents into the loan, balances the file in a click, and a person approves the result before it drives anything.

What 75 minutes a loan is worth.

Data entry and balancing run about 45 minutes on a normal file. Order-outs take another 30. Dromi does each of those in a few minutes, so the time comes back as processing capacity rather than as an abstraction.

Move the slider to your own volume. The arithmetic is plain and the assumptions are printed underneath it.

10500
115hrs
Saved per month
$5,520
Labour per month
$66,240
Labour per year

Modelled on 75 minutes of manual data entry, balancing and order-outs per loan, cut to about 6 minutes, at a $48 per hour processor cost. That is roughly 92 loans of capacity back each month.

Modelled, not measured results

What happens to a document.

01

It is read

Mortgage statements, payoffs, homeowners declarations, drivers licences, notes, title commitments and CDs, tax certificates, FHA refi credit queries and lock confirmations are read into structured loan data and filed per document, per borrower.

02

A person approves it

Every extracted value appears as a pre-filled field, not a decided fact. One click marks it reviewed and stamps who approved it and when. Nothing unverified ever drives an action on a mortgage file.

03

The file balances

Auto balancing runs the arithmetic a processor would, with every line item sourced back to the document it came from, so you can see exactly where each number originated.

The calculations that get argued about.

Both FHA max base loan amount calculations run, and the one that governs is flagged. Aggregate escrow and waived cushions sit side by side with the cheaper option marked and the numbers behind it visible.

Cash to close is complete rather than approximate: payoff with interest padding, escrow netting, title and closing costs, prepaid interest, lender credits and the UFMIP refund.

A borrower portal and a document hub.

Borrowers land on a branded, identity-verified portal and upload against a checklist of exactly the documents you asked for. Files are named and sorted for them, and every upload appears in the pipeline in real time with a running progress count.

Behind it sits a document hub, one per borrower: team folders for conditions and QC, one-click preview, PDF download that converts photos automatically, and share links that expire in 24 hours.

Document types read
11
Order-out escalation
7 business days
Share link expiry
24 hours
Processing pipelines
LOA, Junior, Processor
Approval
Human, stamped

Questions we get asked

What reads the documents, and where does our borrower data go?

Document reading is powered by Anthropic's Claude models under a no-training data agreement, which means borrower documents are never used to train anything. The architecture is built to GLBA expectations and reviewed against the FTC Safeguards Rule. We do not claim a certification we do not hold.

What if the AI reads something wrong?

Then a person catches it, because a person has to approve every value before it does anything. Extracted numbers arrive as pre-filled fields with a reviewed stamp recording who confirmed them and when. The design assumes the model will occasionally be wrong.

Are the time savings measured or modelled?

Modelled, and the assumptions are printed under the calculator: 75 minutes of manual work per loan, cut to about 6, at a $48 per hour processor cost. Your own numbers will differ. We would rather show you the arithmetic than a testimonial.

Does this replace our processors?

It replaces the retyping. The processors we built it with carry more files, not fewer, and spend their time on the parts of a file that actually need judgement.

Send us a file you already closed.

We will run it through and show you what came out, side by side with what your team entered by hand.

Book a demo

Or call 480-848-2864