Bank statement OCR reads most figures correctly, but it makes a small set of predictable mistakes: look-alike digits such as 1 and 7, a lost minus sign or CR marker, and shifted decimal points. The fastest catch is the running balance, because a misread amount stops the arithmetic from adding up.
Which digits does bank statement OCR confuse?
OCR, short for optical character recognition, turns a picture of text into characters, and it confuses digits whose shapes are close. The usual pairs are 1 and 7, 5 and S, 8 and 3, 0 and O, 6 and 0, and 4 and 9. Blur, small print and low contrast make each pair worse.
The cause is the same every time. The software compares each shape on the page with the shapes it knows and picks the closest one. When a scan is sharp, the closest match is nearly always right. When ink bleeds, a page is photographed at an angle or a line of text is very small, two shapes can sit almost equally close and the software has to guess.
The pairs that cause most trouble
- 1 and 7. A printed 1 with a flag at the top and a 7 with a short bar look alike when the bar fades. 1,170.00 turns into 1,770.00.
- 5 and S. A 5 in the middle of an amount can be read as the letter S. The cell then holds text, not a number, and Excel will not sum it.
- 8 and 3. The open left side of a 3 can fill with ink and look like an 8, or the reverse when the ink is thin.
- 0 and O. Statement fonts draw the zero and the letter O almost the same way. A 0 read as O makes the whole cell text.
- 6 and 0, and 4 and 9. These pairs differ in a small loop or opening, which a low-resolution scan can close.
Two things make a digit error harder to spot than a spelling error. First, a wrong digit still looks like a normal number, so nothing on the sheet appears broken. Second, the error often lands in the middle of the figure, where a quick glance does not look. This is why checking by arithmetic works better than checking by eye.
Why the letters matter more than the digits
When OCR swaps a digit for another digit, you get a wrong number. When it swaps a digit for a letter, you get a cell Excel treats as text. A column that is mostly numbers but has a few text cells will quietly skip those cells in a sum. If the total on your sheet is lower than expected, look for amounts that sit on the left side of their cell instead of the right. Text is left-aligned by default and numbers are right-aligned, so alignment is a quick visual tell.
Why do OCR'd statement debits turn into credits?
A debit becomes a credit when OCR drops the sign or marker that told you it was a debit. Banks show money out with a minus sign, brackets, a "DR" label, a separate Withdrawals column, or a trailing dash. If any of those is lost, the amount survives but its direction does not.
Statements mark direction in different ways, and each way fails differently:
- A leading minus sign. The sign is a short, thin line, so a faint scan can erase it. -48.00 turns into 48.00.
- Brackets. (48.00) can lose one bracket, leaving 48.00) or 48.00. Some tools then read the figure as positive.
- CR and DR markers. These small letters sit beside the amount. A scan that clips the margin removes them, and the letters can be misread, with DR turning into OR.
- Trailing minus. A dash after the number, as in 48.00-, is easy to lose at the edge of a column.
- Two columns. When money in and money out have their own columns, OCR can assign an amount to the neighbouring column if the page is tilted or the columns sit close.
A sign error is more expensive than a digit error. When a 48.00 debit becomes a 48.00 credit, the running balance is off by twice the amount, 96.00, because the figure went the wrong way. Totals for deposits and withdrawals both change. A bank reconciliation will show a gap that looks unrelated to any one line.
To catch it, read the converted column against the type of transaction. Rent, card purchases and fees should be money out. Payroll and transfers in should be money in. A row that contradicts its own description, such as "Monthly service fee" shown as a deposit, is almost always a sign error.
Decimal points and thousands separators in OCR'd amounts
A decimal point is the smallest mark in an amount, which makes it the easiest to lose. When OCR drops it, 1,250.00 can become 125000, and when it confuses a comma with a period, the number can become 1.250.00 or 1,250,00. The result is a figure that is wrong by a factor of 100 or 1,000.
The failures follow three patterns:
- A lost decimal point. The cents merge with the dollars, so 85.42 reads as 8542. Excel then shows a whole number a hundred times too large.
- A comma and period swap. A thousands comma read as a period gives 1.250.00, which Excel cannot treat as a number, so it stays as text.
- A stray mark read as punctuation. A speck of dust or a staple shadow near a figure becomes a comma or period, splitting 1250 into 1,2.50.
Regional formats add one more trap. Some countries print 1.250,00 for one thousand two hundred fifty, using the period as the thousands separator and the comma as the decimal mark. If a statement uses that style and your Excel expects the other, a correct reading can still land as text or as the wrong size. Check which format the statement uses before you decide that a column is broken.
These errors are large, so they stand out when you sort the amount column from largest to smallest. A debit in the hundred thousands on a personal checking account is a signal in itself. Look at the top and bottom of the sorted list first, then check each oversized figure against the paper.
For how OCR groups recognized characters into columns in the first place, how recognized words become table columns explains the step where amounts get assigned to a column.
The running balance as a bank statement OCR check
The running balance check finds a misread by arithmetic. Start from the opening balance, add or subtract each amount, and compare the result with the balance printed on that line. The first row where the two numbers disagree contains the error, or the row just above it does.
This works because a bank statement checks itself. Every printed balance should equal the balance above it plus that line's amount, so one misread figure breaks the sum on the row where it sits. You do not need to understand the cause. You only need to find the first row where the sums stop matching.
In Excel the check takes one formula. Put the opening balance in the first balance cell. In the next row, add a column called "Expected" holding the previous balance plus the amount, with debits entered as negative numbers. Then add a "Gap" column that subtracts the printed balance from the expected one. Any Gap that is not zero is a row to check against the paper.
Eight common confusions and how the balance exposes them
The table shows example misreads. The amounts are made up, and the gap is the amount by which the expected balance differs from the printed one, assuming the printed balances themselves were read correctly.
| Confusion | True amount | OCR reads | What the balance check shows |
|---|---|---|---|
| 1 read as 7 | 1,170.00 | 1,770.00 | Gap of 600.00 on that row |
| 5 read as S | 25.50 | 2S.50 | Text cell; SUM skips it and the Expected formula shows #VALUE! |
| 8 read as 3 | 880.00 | 380.00 | Gap of 500.00 (the amount is read too small) |
| 0 read as O | 100.00 | 1O0.00 | Text cell; the Expected formula shows #VALUE! |
| 6 read as 5 | 60.00 | 50.00 | Gap of 10.00 |
| 9 read as 4 | 94.00 | 44.00 | Gap of 50.00 |
| Decimal point lost | 1,250.00 | 125000 | Gap of 123,750.00 |
| Debit read as credit | -48.00 | +48.00 | Gap of 96.00, twice the amount |
Two lessons come from the table. Where the misread is still a number, the gap equals the size of the misread, so it tells you how far off the number is and often which digit moved: a gap of 600.00 points to a hundreds digit, and a gap of 10.00 points to a tens digit. And a sign error leaves a gap of double the amount, which is a clue on its own.
When the balance column is the part that was misread
Sometimes the error sits in the balance column and the amount is right. The formula above starts each row from the printed balance on the row before, so a misread amount makes one row fail and the next row pass. A misread balance makes two rows fail in a row, with gaps of the same size and opposite signs, because that balance is the target of one row and the starting point of the next.
A statement with no running balance column, as credit card statements often are, cannot be checked row by row. Compare the sum of the converted amounts with the printed totals instead, and treat a mismatch as a sign to examine the largest figures. FastBankConvert checks the running balance as part of its conversion and tells you before download if the balances do not add up. It is getting ready to open, and the homepage explains how to join the early-access list.
Faded and thermal-paper bank statements under OCR
Faded and thermal-paper statements are the worst case for OCR, because the ink itself has weakened. Thermal paper darkens with heat and fades with time, light and friction, so a receipt-style statement can lose its lightest strokes first. Those strokes are the ones that separate a 3 from an 8.
Most bank statements are printed on ordinary paper with toner or ink, and they age well. Thermal paper turns up in ATM receipts, point-of-sale slips and some branch-printed account summaries. If a client keeps a mini-statement from an ATM, treat it as fragile.
Here is what helps with weak originals:
- Scan sooner rather than later. A faded page will not improve, and light and warmth make it worse.
- Scan in greyscale or colour, not black and white. A black-and-white scan throws away the grey strokes that carry the faint digits.
- Raise the contrast on the scanner a little, but check the result. Too much contrast thickens thin strokes and joins neighbouring digits.
- Keep the page away from heat and sunlight until it is scanned, and do not stack other papers on top that could smudge it.
- Plan to verify every figure. On a weak original, treat the spot-check as a full check of all amounts, not a sample.
Handwriting is a different problem from fading. If someone has written notes or figures onto a printed page, ordinary OCR is not built for them. Whether OCR can read handwriting in a PDF covers what is realistic.
Before any of this, confirm that your file is really an image. If words in the PDF can be selected, there is nothing to recognize and no OCR error to worry about, as the ten-second test for image PDFs shows. For clean digital statements, a bank statement converter reads the text directly. Reserve the careful OCR checks above for the pages that are truly pictures.