The 30-to-45-day mortgage close timeline is a familiar industry number. Borrowers know it going in, and most accept it as the natural rhythm of the process. What the number obscures is how that time is actually distributed. Most of those days are waiting: waiting for the appraisal, waiting for title work, waiting for borrower responses. The active staff hours per file, when you account only for the time a human being is actually doing work on it, are concentrated in a much smaller window.
We spent several months working closely with a regional lender in the Midwest, mapping their pre-underwriting workflow at a task level. Not hours per loan in the aggregate, but what specifically a loan officer, processor, and underwriter were doing during the time they spent actively touching a file. The results reshaped how we think about where document extraction produces value.
The 8-to-12 hour pre-underwriting window
For a conventional purchase application with an employed borrower, a typical file moves through roughly 8 to 12 hours of active staff time before it reaches the underwriter's queue. That includes the initial intake review, document collection follow-up, data entry into the loan origination system, completeness verification, income calculation, and file organization for submission.
That range is not uniform across file types. A W-2 borrower with two years of employment history, straightforward assets, and a complete submission can move through in closer to 6 hours of active work. A self-employed borrower with 1099 income, Schedule C filings, multiple bank accounts, and a gap in employment documentation can require twice that, sometimes more.
The hours also are not evenly distributed across the team. Loan officers tend to spend concentrated time at intake and then again when conditions come back from underwriting. Processors carry the bulk of the middle hours: document chasing, reorganization, data entry, and internal handoffs. Understanding where each role's time goes changes which part of the problem is worth solving first.
Task breakdown: where the hours concentrate
Looking at active work tasks rather than total elapsed time, a few categories dominated:
Document organization and completeness review (1.5 to 2.5 hours per file). This is the time spent opening submitted documents, identifying what was received, noting what is missing, and organizing the file for review. For a borrower who submitted 12 PDFs of varying types through an upload portal, this is not trivial. Documents arrive out of order, some are partial submissions, and the processor has to build a mental and physical map of what is present before they can begin substantive review.
Data entry and field reconciliation (1 to 2 hours per file). Even when a loan origination system is in place, significant data entry happens manually. Income figures from W-2s, employment dates from paystubs, account balances from bank statements: these are re-keyed from paper into the LOS rather than extracted automatically. Each field is a potential transcription error. Each field also takes time.
Borrower follow-up for missing or expired documents (variable, often 0.5 to 1.5 hours of direct time but multiple days of elapsed time). The active time to draft and send a document request is short. The elapsed time waiting for the borrower to respond, then reviewing what they sent, then identifying what is still missing, is where the calendar impact accumulates.
Income calculation and initial pre-qualification check (1 to 2 hours per file). Gross monthly income from multiple sources, DTI calculation, reserve verification: these calculations are the core of the pre-underwriting pass. For complex income files, this is genuinely analytical work. For straightforward files, it is still time-consuming when the inputs need to be located and transcribed from documents rather than extracted.
Where extraction produces the most value
The highest-leverage intervention in this workflow is not at the income calculation step, even though that is where most people assume AI adds value. It is at the document organization and completeness step.
When a processor spends an hour and a half organizing a file before they can begin substantive review, they are doing work that is highly checkable but not analytical. The documents have defined structures. The date fields on a bank statement are in a predictable location. The period covered is stated on the first page. Whether a required field on a 1003 is blank is a binary question. None of this requires human judgment. It requires reading.
Extraction that produces a structured inventory of what was received, with dates verified, gaps identified, and fields read into structured output, collapses that organization phase from an hour to a few seconds of review. The processor's first active engagement with the file becomes the actual review, not the preparation for it.
The data entry step is the second area where extraction time returns directly replace manual work. A processor who re-keys 40 data points per file across 20 active files per month is spending roughly two to three full workdays per month on transcription. Extraction does not eliminate the need to verify values, but it shifts the task from generating numbers to checking them, which is both faster and less error-prone.
The bottleneck that does not move when you speed up underwriting
There is a common framing in mortgage tech that positions underwriting capacity as the central bottleneck. Faster underwriting turnaround is the goal, and everything upstream should feed underwriting more efficiently. That framing is not wrong, but it misidentifies where the constraint actually lives for the firms we have worked with.
When we asked underwriters where their time went, the answer was not "reviewing complex credit questions." It was "chasing information that should have been in the file when it arrived." Conditions issued back to processors were frequently for items that the processor, given more time or better tooling, could have caught before submission. The underwriting bottleneck was partly a symptom of an incomplete handoff from the pre-underwriting stage.
Speeding up underwriting review without improving the quality of the file at handoff reduces the underwriter's time on a given file by maybe 20 minutes. Improving document completeness and data accuracy before the file arrives at underwriting can eliminate entire condition cycles, each of which might add two to four days to the timeline.
The practical implication for how to deploy extraction tooling
The takeaway from this time-mapping exercise shaped how we built Maestro's priority order. The extraction focus is front-loaded: what was submitted, is it complete, are the dates within program windows, and what do the key fields say. That output lands in the processor's queue before they open the first PDF. The substantive review they do is oriented around the structured report rather than the raw document stack.
The 30-day close timeline is not going to compress to 10 days. Most of those calendar days are third-party dependencies: appraisals, title searches, final payoff statements, rate lock windows. What the pre-underwriting tooling changes is the distribution of staff hours within that window, and how often a file has to cycle back for additional information. Those cycles are where the days that can actually be saved are sitting.