Your whole pipeline, balanced by Day 1

Balancing a file is not difficult work. It is careful work, done under time pressure, on numbers that arrive as PDFs from six different places. That combination is why it goes wrong.

The payoff statement, the homeowners insurance declaration, the tax certificates, the title fee sheet. Dromi reads them into structured loan data instead of leaving them as documents a processor has to open one at a time and transcribe.

What the arithmetic actually covers

Auto balancing runs the calculations a processor would run, and shows you where every number came from:

• Both FHA max base loan amount calculations, with the governing one flagged.
• Aggregate escrow against waived cushions, side by side, cheaper option marked.
• A complete cash to close: payoff with interest padding, escrow netting, title and closing costs, prepaid interest, lender credits and the UFMIP refund.

Every line item is sourced back to the document it was read from. That matters more than the speed. A number you cannot trace is a number somebody has to redo.

A person still approves it

Nothing the model extracts drives an action on its own. Values arrive as pre-filled fields, a human confirms them in a click, and the confirmation stamps who approved it and when. We designed it assuming the model would occasionally be wrong, because it occasionally is.

What changes is not the judgement. It is the forty-five minutes of typing in front of the judgement.