A digital PDF carries its text, so the numbers in it are exact. A scan is a picture of the page, and every character has to be inferred from pixels. On a bank statement the damage lands almost entirely on numbers, and one wrong digit is as costly as a missing row. The good news is that scanned statements fail in patterns, and each pattern shows up in the opening-to-closing check.
The errors, and what the check shows
Read the right-hand column as the symptom you will actually see. The middle column is the reason to fix the scan rather than the cell.
| What the scan gets wrong | Why it happens | What the check shows |
|---|---|---|
| A look-alike digit: 5 and 6, 6 and 8, 3 and 8, 1 and 7, 0 and 8 | Low resolution, faint ink, smudges, fax noise | A difference with one non-zero digit; one flagged row |
| A lost minus sign, bracket or CR/DR marker | Thin marks vanish at low resolution | A difference of exactly twice an amount |
| A lost or shifted decimal point or comma | Specks merge with punctuation | A difference of 99 times an amount, or an unreadable amount |
| A letter for a digit: O for 0, l for 1, S for 5, B for 8 | Similar glyph shapes | An amount that is not a number, or a one-digit difference |
| A dropped row | Skew, a crease, or a page break | A difference equal to the row’s amount; the next row flagged |
| A misread printed balance | The same causes as any digit | Two flagged rows with equal differences while the sum ties |
| A wrong date or payee | Ambiguous or smudged text | Nothing: totals depend on neither |
A worked example
A statement opens at 4,210.50 and closes at 3,715.20. The scan yields deposits of 2,100.00 and withdrawals of 2,395.30, so the calculated closing balance is 3,915.20, which is 200.00 too high. A single non-zero digit points to one misread, and the running balance flags the rent line: printed 1,850.00, read as 1,650.00. Withdrawals were really 2,595.30, and the sum ties.
A misread balance looks different. Suppose row 7 prints a running balance of 6,480.20 and the scan reads 6,430.20. Row 7 is flagged by 50.00, and row 8 is flagged by 50.00 the other way, because it starts from the wrong figure. Every amount is right and the sum still ties, so the fix is the printed balance, not an amount.
What the check cannot see
Dates are the blind spot. A day and month swapped, as in 03/04 against 04/03, change nothing in the totals, and neither do payee names, check numbers or categories. Review them by eye. Two errors that cancel each other can also pass, and so can a misread that happens to preserve the arithmetic.
Scan settings that reduce errors
- Resolution. Aim for 300 dpi. Lower resolutions lose thin marks first.
- Mode. Grayscale usually keeps faint print that a pure black-and-white setting can drop.
- Geometry. Flat, square, all four edges visible, one statement page per image.
- Filters. Switch off auto-enhance and heavy compression; they can thicken or erase small characters.
- Order. Pages in sequence, none duplicated, none rotated.
Step by step: review a scanned statement
- Count the pages. Compare the pages you hold with the statement’s own “page 1 of N”. A page that never reached the scanner is invisible to any reader.
- Confirm the summary figures. Check the opening balance, closing balance and any printed totals against the page image first. Everything else is measured against them.
- Run the balance check. The target is a difference of 0.00 with no flagged rows.
- Chase each flagged row. Put the page image beside the row, correct the cell from the page, and recheck.
- Spot-check what the check cannot see. Look at five dates, including any with a day above 12, every check number, and the three largest payees.
- Re-scan instead of repairing. If more than a handful of rows are wrong, the source is the problem. Scan again.
What can go wrong
- Trusting a tidy table. A neatly aligned result can still hold a misread digit. Only the balance check shows it.
- Correcting to force a tie. Never change an amount just to make the difference vanish. Fix it from the page, or the error is hidden rather than removed.
- A page scanned twice. Its rows appear twice and the difference equals that page’s net activity. Check page counts before you convert.
- Sideways or upside-down pages. They may be read poorly or not at all, which looks like a dropped page. Straighten them before converting.
- Photocopies of photocopies. Each generation loses faint marks such as minus signs and decimal points. Ask for the original or a first-generation copy.
- Writing on the page. Handwritten notes, stamps and highlighter over amounts cut accuracy. Use a clean copy.
- Curved pages. A bent spine in a phone photo distorts rows near the fold. Press the page flat before shooting.
- Faxed statements. Fax resolution drops digits. Ask for the original or a download from online banking.
Where SheetParser fits
SheetParser converts scanned PDFs and phone photos to one row per transaction and runs the same opening-to-closing check as for a digital PDF. For a PDF upload the original page sits beside the rows, so a flagged cell can be checked against the image; a photo has no side-by-side view, so keep it open in another window. Expect more flags on a scan than on a native file. Start with the scanned statement tool, free for 1 file a day up to 10 pages, with paid plans opening files up to 150 pages. The balance-check guide explains how to read each difference.