Digital vs. scanned: the text layer difference
A digital bank statement — the PDF your bank lets you download — contains a text layer. Every character exists as data; you can select it, copy it, search it. Extraction from these files is precise because nothing has to be guessed.
A scanned statement is a photograph. Whether it came from a flatbed scanner, a phone camera, or a fax machine (still common with older records), the PDF wraps an image with no text underneath. Converting it to Excel requires OCR — optical character recognition — to read the image, identify characters, and then reconstruct the layout: which numbers are dates, which are amounts, and which column each amount belongs to.
That reconstruction step is where scanned conversion earns its reputation.
What accuracy to expect
On a clean scan, modern OCR reads characters very well — but "characters read correctly" and "spreadsheet is correct" are different bars. The failure patterns specific to bank statements:
- Digit confusion on degraded prints. Faded, photocopied, or thermal-printed statements produce 0/8, 1/7, and 5/6 mix-ups. One wrong digit in an amount is a wrong amount — there's no spell-checker to flag it.
- Column drift. On skewed pages, an amount can land between the withdrawal and deposit columns and get assigned to the wrong one, flipping its sign.
- Wrapped rows. Long descriptions span two printed lines, just like digital statements, but OCR has no text-layer hints to rejoin them.
Plan on a workflow of convert, then verify — not convert, then trust.
Scan it right the first time
If you control the scanning, your choices here determine everything downstream:
- 300 DPI. Enough detail for reliable recognition without bloated files. Below 200 DPI, accuracy falls off fast.
- Grayscale or black-and-white text mode. Color adds noise and file size without helping character recognition.
- Flat pages. Press the spine flat against the glass; a curled gutter shadows characters into garbage. A sheet feeder beats a phone photo.
- No skew. Align pages against the scanner's edge guide. Even a few degrees of tilt breaks column detection.
- The whole page. Don't crop headers, totals, or the balance summary — you need those printed numbers to verify the output.
If phone photos are all you have: shoot straight down, in even light, with the page filling the frame. It works, but expect a lower ceiling than a real scan.
Before you OCR at all: try the bank
Many "scanned" statements exist only because someone photocopied a paper original years ago. Before investing in OCR, check whether the bank can re-issue the period as a digital PDF — most US banks keep statement history available for years (seven is common), either in online banking or on request. A re-issued digital statement converts cleanly with no OCR errors and no verification burden. Ten minutes on the bank's website can beat an afternoon of correcting misread digits.
If the bank truly can't provide the period — closed account, defunct institution, records past retention — then scan well and verify hard.
Verify against the printed balances
OCR on bank statements comes with a gift most document scanning doesn't get: statements are self-checking. Every one prints an opening balance, a closing balance, and usually deposit and withdrawal totals.
After conversion, recompute opening + deposits − withdrawals = closing from the extracted rows and compare against the printed numbers:
- It ties. The extraction is almost certainly right — the odds of compensating errors landing exactly on the closing balance are tiny.
- It's off by a small or round amount. Hunt for a misread digit or a sign flip. Sort the sheet by amount and check the largest transactions against the PDF first.
- The row count is wrong. Look for wrapped descriptions split into two rows, or a page that was skipped entirely.
Also confirm the extracted date range matches the statement period — a dropped page usually shows up as a two-week hole in the dates.
When a statement won't tie after a reasonable hunt, don't force it. Retype the handful of suspect rows manually against the PDF, re-run the balance check, and only then move on. Verification feels slow until you compare it with finding the same error after the data is in Excel, imported, and reconciled against the wrong numbers.
Where ReconCSV fits today
An honest note on where things stand: ReconCSV currently converts text-layer PDFs — the digital statements banks issue — entirely in your browser, with balance reconciliation built in. It does not OCR scanned statements yet; AI-based support for scanned documents is on the roadmap. If that's the capability you're waiting for, the early-access waitlist is on the pricing page.
In the meantime, the scanning and verification practices above apply no matter which OCR tool you use — and the balance check at the end is the part you should never skip.