Extracting transactions from a bank statement or line items from an invoice is the kind of task that looks solved — until the parser quietly swaps a decimal, merges two rows, or invents a balance. The output looks clean. The numbers are wrong.
I built LedgerLens around the unglamorous truth: the value isn't extraction, it's knowing which extractions to trust.
30-second takeaway
- A parser that always says "done" is lying some of the time — and you can't see which times.
- The useful output is a confidence score per field, not a confident grid.
- "Verify before use" is the feature, not a disclaimer.
The silent error is the real one
Most extraction tools hand you a table and call it a win. The error rate lives in the rows that look fine — the amount that's off by a factor of ten because of a comma, the line item that got dropped, the balance that was inferred instead of read. You find those in reconciliation, which is the worst place to find them.
Scoring every field means the tool tells you "this one I'm sure of, this one you check" — which is what a human would do if they had time.
What the tool does
Paste a statement or invoice, or upload a text-layer PDF/DOCX. LedgerLens extracts structured transactions, balances, and line items — each with a confidence score and a clear flag on the ones to verify before you use them.
LedgerLens runs in the browser, no account, no key.
FAQ
What is LedgerLens?
A tool that extracts transactions, balances, and line items from statements and invoices, with a confidence score on every field. No account, no key.
Does it reconcile the numbers for me?
It flags the low-confidence fields so you verify before use. The human still owns the final numbers — that's the point.
Will it read a scanned PDF with no text layer?
It works best with a text layer; for pure scans, the extraction quality depends on what the file actually contains.
Author's Process
This product was built with AI coding tools. The honest design choice was to surface uncertainty instead of hiding it.
Top comments (0)