OCR misreads on bank statements: what the size of the gap tells you

When OCR misreads a bank statement, the converted rows no longer add up to the printed closing balance, and the size of the gap usually tells you what kind of mistake it was. A gap of twice one line is a lost minus sign. A gap of 99 times a line is a lost decimal point. A gap divisible by 9 is often two swapped digits. A round gap like 6,000.00 is one digit misread in the thousands place.

That makes the balance check more than a pass or fail. It's a pointer. Here's how to read it, with made-up numbers.

Start with the one sum

Opening balance + deposits − withdrawals = closing balance. Run it on the converted rows. If it lands on the closing balance the bank printed, every line is very likely right. If it doesn't, write down the gap: rows total minus printed closing.

Fingerprint 1: a gap of twice one line

A withdrawal of 215.40 is read without its minus sign, or without the "DR" next to it. Instead of taking 215.40 off, the rows add 215.40. The total is now 430.80 too high, exactly twice the line.

So when the gap is an even amount, halve it and look for a line of that size. 430.80 ÷ 2 = 215.40. Find the 215.40 and check its sign.

Fingerprint 2: a gap of 99 times one line

A 89.50 card payment is read as 8950, because the decimal point was faint on a scan. The line is now 8,950.00 instead of 89.50, an overstatement of 8,860.50. Divide the gap by 99 and you get 89.50.

Lost decimals are rare on downloaded PDFs and common on photocopies and faxes. They're also the misread that does the most damage, because one line can swamp a month.

Fingerprint 3: a gap divisible by 9

A deposit of 1,463.00 is read as 1,436.00. Two digits swapped places. The gap is 27.00, and 27 is divisible by 9. Any transposition of two adjacent digits produces a difference divisible by 9; it's the oldest trick in bookkeeping, and it works just as well on OCR output as on hand-typed ledgers.

It's not proof, since plenty of other gaps happen to divide by 9. But it tells you to look for two digits the wrong way round.

Fingerprint 4: a round gap

A deposit of 1,240.00 is read as 7,240.00. The gap is 6,000.00. A gap that's a single digit times a power of ten (6,000, 400, 0.30) usually means one digit was misread in that place. OCR's favourite confusions are 1 and 7, 3 and 8, 5 and 6, 0 and 8.

A round gap often has more than one candidate line. If three lines have a 1 in the thousands place, any of them read as a 7 would explain 6,000.00. Narrow it down with the running balance column, if the statement prints one, or by looking at those lines on the page.

On invoices and checks

The same thinking works on other documents, with different totals.

Invoices. Each line should satisfy quantity × unit price = line amount. If 12 × 18.75 is printed as 225.00 but the quantity was read as 72, the line computes to 1,350.00 and fails before it ever reaches the invoice total.

Checks. The amount is printed twice, once in words and once in figures. If the figures read 912.00 and the words say "four hundred twelve and 00/100", one was misread, and when the two disagree on a real check, the words generally win. A handwritten 4 that looks like a 9 is the classic case.

Why a percentage can't do this

An accuracy rate tells you how often a tool is right across many documents. It can't tell you which line on this statement is wrong. The balance check can, and the fingerprints narrow it to a line or two.

The misreads that matter most, why an accuracy percentage isn't the test, and how each document type is checked are on our page about OCR accounting software.

Disclosure: I'm Damin, and I run ZeraBooks. Our Convert tool reads PDF bank and card statements, scans included, and proves each one against its printed opening and closing balances. When a statement doesn't tie, it reads it again automatically and keeps the attempt that does; if none ties, the statement is marked Needs a look with the difference shown, so you can go straight to the fingerprint. It's on the bank statement converter page.

Write the gap down before you start hunting. Half the time, it tells you where to look.