Direct answer
CSV is already a table. PDF is a layout. If the PDF has a text layer from the bank's generator, F20 FinanceOps can lift transaction lines and still check running balance. If the PDF is a scan or photo, F20 refuses it — there is no OCR. BankXLSX and Nanonets publish OCR products for that case. A tagged BAI2, MT940 or camt.053 file is usually better than either PDF flavour because it carries control totals or structured references instead of coordinates on a page.
Banks like PDFs because they look like the paper statement and they print. Spreadsheets like CSV because they sum. Those incentives do not align. Bookkeepers live in the gap, retyping, or paying per page for someone else's model to guess at glyphs.
F20's position is unfashionable and intentional: we will not guess. A refused scan is better than a invented deposit. If that blocks you, the honest product is OCR, not us.
Text-layer PDF versus scan
Text layer: the generator wrote characters. You can select text in a viewer. F20 reads those characters, not pixels.
Scan: a picture. OCR infers characters. Layouts with stamps, logos, overlapping columns, and 'Page 1 of 14' running through the amount column are where OCR invents money. We do not measure a vendor's accuracy here. We refuse to join that game on F20.
CSV is not innocent
CSV still has decimal commas, dropped leading zeros on account numbers, and Excel turning 01-03 into a date. F20's BAI converter notes tell you to import account columns as text. A CSV from a bank portal may already have lost trailing references. Prefer the tagged file when you can.
Comparison
Source of truth, not aesthetics.
| Tagged file (BAI2/MT940/camt) | Text-layer PDF | Scanned PDF | CSV | |
|---|---|---|---|---|
| Machine meaning | Designed for software | Layout plus characters | Pixels | Already rows |
| F20 | Yes | Lift + refuse if scan | Refuse | As an input to recon tools |
| OCR vendors | Optional | Optional | Their job | Usually unnecessary |
Workflow we recommend
Ask the bank for BAI2, MT940 or camt.053. If they only offer PDF, check for a text layer. If it lifts, still run the balance check. If it is a scan, use OCR software, then still do not trust the first total — tie to the PDF's printed closing balance by hand once.
Pricing
F20: free for the files it accepts. OCR: BankXLSX and Nanonets public prices on their vs pages. Paying OCR for a file the treasury portal already offers as BAI2 is wasted budget.
Privacy
Pixels of a statement are still the statement. OCR APIs see them. F20 never sees a scan because it never accepts one.
What to tell a client who only has PDF
First: is it selectable text? If yes, try F20's PDF tool and the balance check. If the lift is clean, you still ask the bank for BAI2 or MT940 next month, because a layout change will break the lift without warning.
Second: if it is a photo, say no to F20 and yes to an OCR product, or yes to the client getting a proper file. Do not quietly type the statement into Excel and call it conversion. That is the most expensive OCR: a person.
Third: keep the PDF as evidence even after you have CSV. Auditors ask what the bank showed. A spreadsheet is not that. The tagged file is better evidence still, because it is what the bank's system emitted, not what a page happened to look like. If the bank charges for the tagged file and gives PDF away, pay for the tagged file. OCR page fees over a year usually exceed the treasury-channel cost, and they buy you guesses instead of records.
Methodology
F20 catalog (no OCR). BankXLSX and Nanonets official OCR copy. Not an OCR bake-off.
Where F20 FinanceOps is a better fit
- You can get a tagged file or a text-layer PDF and want a balance check without storing the upload.
- You would rather be blocked than receive invented rows.
Where PDF OCR or a bank CSV is a better fit
- Scans/photos: BankXLSX or Nanonets as they document OCR.
- The portal already offers a decent CSV and you trust its columns: skip conversion; recon the CSV.
- You need QBO: BankXLSX Plus, not F20.
- You convert photographed statements every week for a bookkeeping firm: an OCR plan with support may cost less than fighting F20's refusal.
FAQ
No text layer. F20 has no OCR. Get a tagged file or a text-layer export.
For arithmetic, usually. For evidencing what the bank showed a human, PDF still matters. Keep both if you must explain a line to an auditor.
That is still OCR. Same accuracy problem, worse audit trail.
No. Some 'digital' statements are flattened images in a PDF wrapper.
/tools/bank-pdf-to-excel
No. A photograph is a scan. F20 refuses it. Use an OCR product such as BankXLSX or Nanonets as they document that job, or ask the bank for BAI2, MT940, camt.053, or a PDF that still has a real text layer you can select in a viewer.
Sources
- F20 FinanceOps policies — F20 FinanceOps. Verified. Uploaded files are processed in memory and dropped when the response is sent. There is no charge to run a tool. Output is not professional advice.
- Bank Statement Converter: PDF Bank Statements to Excel & CSV — BankXLSX. Verified. BankXLSX converts PDF bank statements to Excel, CSV, JSON, or QuickBooks QBO, including scanned statements.
- Automate Data Extraction from Bank statements — Nanonets. Verified. Nanonets offers a pre-trained bank-statement OCR model, including photographs and email attachments.
Lift a text-layer PDF — If this refuses the file, that is the product working. Ask the bank for BAI2, MT940 or camt.053. A photographed statement still needs OCR elsewhere, not a retry on this page.