In looking for a system which can scan bank statements using optical character recognition, what we are looking for is a decent rise in productivity. We are not looking for a magic wand. The bank statements we scan are of variable quality, and we accept that the accounts clerk will need to review the results one page at a time. Most obviously, if cheques have been written then the clerk will need to look at the cheque book stubs and may need to type them in.
After scanning, we show each page as scanned to the spreadsheet on a “blink comparator”. The clerk can visually compare the result with the original. Our software flags up obvious errors, and it now also highlights the presumed business area of the statement. If the clerk accepts this highlighting as correct, then one click of a button removes anything extraneous. This does have to be done one page at a time, but let us repeat that the bank statements we get are of variable quality. Some are photocopies of photocopies with ticks and handwriting added.
Once everything is scanned into one big bank statement on our spreadsheet system, the clerk needs to review the narrative. There are buttons to help with tidying up narrative, and if anything is actually missing, then the clerk can run a narrative prediction routine to guess what it was, which is often successful. The clerk then overtypes anything that is wrong.
A little time spent looking at the bank statements can also be regarded as the reconnaissance phase of the job, and proverbially “time spent in reconnaissance is seldom wasted”. We don’t necessarily want everything just a bit too streamlined because then things might get missed.