Scanned statements, converted and checked.
Borrowers send faxes, crooked phone photos and 90-DPI scans. OCR alone hands you a plausible table with two digits wrong. Here the running balance is recomputed after every row, so a misread breaks the math and gets flagged instead of becoming a number in your file.
First file free, up to 10 pages · paid packs run up to 600 pages per file · files shredded after conversion
Brokers & credit analysts
Borrower files arrive as photos and faxes; you still need a ledger that ties out before you send it on.
Accountants & bookkeepers
Client records are scanned printouts. You need them in Excel without retyping and without silent errors.
Auditors & reviewers
You need to see which cells could not be read cleanly, not a confident guess.
What this actually fixes
A misread digit looks like data
One wrong character produces a table that looks correct and reconciles to nothing. Finding it manually can take hours.
Skew and shadow break rule-based OCR
Folded pages, dark photos and fax artefacts are the normal case for borrower files, not the edge case.
No way to show what was unreadable
Most tools output a result and hide the uncertainty. On a scanned source, the uncertainty is the important part.
What comes out
One upload, one set of compute-ready outputs — not a screenshot of a table.
Misreads surface as breaks
Every row is re-tested against opening + deposits − withdrawals = closing. If the chain breaks, the row is flagged.
Disagreements shown, not averaged
When a second reader disagrees on a cell, that cell is highlighted and listed with a link to the source page.
Clean Excel either way
One row per transaction, no merged cells, numbers as numbers — ready for formulas and pivot tables.
Source page behind every value
Click a figure to see the exact region of the scan it came from.
Same tie-out as a native PDF
The reconciliation does not change because the source is an image. The flags do.
Verifiable if you share it
The certificate binds the hash of the file you uploaded, so a third party can confirm which document the output came from.
Documents we read
Native PDF, scan, phone photo, Word, Excel, PowerPoint, CSV, HTML, email or image.
How it works
Upload the scan or photo
Free tier: one file per day up to 10 pages. Paid packs open files up to 600 pages.
Read twice, then do the math
A layout-aware pass reconstructs the grid; a second read is compared cell by cell; then the balance chain is recomputed.
Review the flags, then export
Everything disputed or broken is listed with its source page. Export to Excel, CSV or QBO when the file is clean enough.
Every plan, every audience
- One file = one credit, up to 600 pages on a paid pack
- No merged cells; numbers arrive as numbers
- Every extracted value traces back to its source page
- Files shredded after conversion, never used for training
- A fresh file allowance every billing cycle; cancel anytime
- An optional hash-bound certificate anyone can verify for free
Questions
My borrower sent a photo of a printed statement. Will that work?+
Yes — a phone photo is a normal input. Expect more flagged cells than a native PDF, because the source is harder to read. The flags are the point.
What if OCR reads a number wrong?+
The running balance will usually stop tying out and the row is flagged. A misread that happens to preserve the arithmetic can still pass, which is why second-reader disputes are shown separately.
How much does it cost?+
The first file each day is free up to 10 pages. Packs are per file and never expire: $79 for 10, $199 for 30, $499 for 100, each file up to 600 pages.
Do you keep the scan?+
No. Files are shredded after conversion and never used for training. Paid accounts keep the report in their own history and can delete it.
Run your next file
Upload the document, get compute-ready output with the source page behind every number. The first file each day is free.