Solution

Document Classification & Intake Intelligence

For lending operations and intake teams: every inbound borrower document classified by content, slotted into the loan-program checklist, and flagged the day it arrives.

1040K-1Bank statementsPay stubsFinancial statements
1M+ documents processedEvery classification decision auditableLow-confidence documents human-reviewed

The problem

Why this exists

scan_final_2.pdf

Filenames tell you nothing

A 40-page scan holds three different documents; the K-1 arrives as a photographed page. Someone opens every file to find out what it actually is.

A week

Wrong-year returns found mid-review

The underwriter discovers the 2024 return is actually 2022, and the deal loses a week while the borrower is chased for the right one.

Every deal

Manual sorting sets the pace

Intake staff sort, rename, split, and slot documents by hand — slow on a good day, inconsistent on a busy one, and the error rate travels downstream.

The product, not a promise

An intake queue you can trust

Document Classification & Intake Intelligence — workspace
Combined 40-page scan split into its three documentsSplitcited
1040 identified by content, stamped with borrower and tax yearClassifiedcited
Document slotted into the loan-program checklistSlot filledcited
Duplicate upload detected and collapsedDeduplicatedcited
Low-confidence scan routed to a person instead of guessingverify
Wrong-year return flagged the day it arrivedGap visiblecited
HUMAN-APPROVED BEFORE IT POSTS

How it works

File in. Answer out.

  1. 1

    Receive

    Documents arrive from email, portal uploads, and scans — mixed, misnamed, and out of order.

  2. 2

    Classify

    Each document is identified by its content — a 1040 is a 1040 even when the file is called scan_final_2.pdf.

  3. 3

    Extract

    Borrower names, entities, tax years, and periods are pulled and attached to each document.

  4. 4

    Assemble

    Documents slot into the intake checklist for the loan program, building the package as items arrive.

  5. 5

    Flag

    Wrong-year returns, missing pages, and unreadable files are surfaced immediately, not at underwriting.

Who it's for

Built for the people who own the outcome

Intake specialist

Chase gaps instead of sorting piles.

  • Mixed scans split, identified, and slotted automatically
  • Wrong-year and stale documents flagged on arrival
  • The borrower conversation happens while it is still easy

Head of lending operations

Intake quality that holds at volume.

  • The same classification standard on every deal, busy day or not
  • Checklist gaps visible in real time across the pipeline
  • Downstream rework shrinks because errors stop at the door

Compliance / internal audit

Every classification decision on record.

  • What each document was identified as, from what, at what confidence
  • Low-confidence items route to a person instead of guessing
  • The intake package is provably what its labels claim
Mortgage lendersSBA lendersCommercial banksConsumer lendersCredit unionsFintech lenders
1M+documents understood
Checklist-awareclassification per loan program
On recordevery classification decision
Human-reviewedlow-confidence documents

Every lending workflow inherits the quality of its intake. Borrower documents arrive as they always have — a portal upload here, an email attachment there, a 40-page scan containing three different documents — and someone has to turn that pile into an organized package before any real work starts. Done manually, it is slow and inconsistent. Done wrong, the cost lands downstream: an underwriter discovers mid-review that the “2024 return” is actually 2022, and the deal loses a week.

Classified by content, slotted by checklist

The platform reads every inbound document and classifies it by content, ignoring the filename. A K-1 is recognized as a K-1 whether it arrives alone, buried in a combined PDF, or as a photographed page. Multi-document files are split, duplicates are detected, and each item is stamped with the identifiers that matter: borrower, entity, tax year, statement period.

Knowing a document is a bank statement is half the job; the other half is knowing whether it is the bank statement the deal needs. Classification runs against the intake checklist for the specific loan program, so each document lands in its checklist slot and the gaps become visible in real time. Wrong-year returns, missing schedules, and stale statements are flagged the day they arrive, while the borrower conversation is still easy.

Trust through traceability

Every classification decision is auditable: what the document was identified as, what was extracted, and with what confidence. Low-confidence items route to a person instead of guessing — the platform’s job is to be reliably right or honestly unsure. The result handed to underwriting is a structured intake package where every document is what the label says it is, and can be proven so.

Objections, answered

What teams ask us first

How accurate is classification on real-world scans?

The platform identifies each file by its content — 1M+ documents processed — including photographed pages and combined PDFs. When confidence is low it routes the document to a person rather than guessing — the design goal is reliably right or honestly unsure.

Our checklist varies by loan program — can it follow ours?

Classification runs against the intake checklist for the specific program, so each document lands in its checklist slot and the gap list reflects what this deal actually needs. A bank statement is judged as the bank statement the deal requires, or flagged as the wrong one.

What does audit see when a package is questioned?

Every classification decision is on record: what the document was identified as, which identifiers were extracted, at what confidence, and who reviewed anything low-confidence. Each label in the package can be proven, not just asserted.

How long to stand this up on our intake?

Point it at a sample of real inbound documents and your program checklists, review how it classifies and flags them, and tune from there. Teams typically run it alongside manual intake within weeks and cut over as the flags earn trust.

Bring your messiest intake folder.

Watch mixed, misnamed scans become a classified, checklist-slotted package live in the demo.

Request a demo