I am now starting to be happy with my Optical Character Recognition system. I received a batch of rather heavily patterned bank statements, and the client had written on a few of the transactions. After scanning, most of the numbers were OK, but a few were missing, and a few were misread because of the handwriting.
The computer-assisted blink character shows each statement with just dates and numbers, and as little else as possible, in an “in your face” style to make comparison easy. The computer marks all the running totals which look wrong regardless of transaction order. I then type in the numbers that need fixing from the bank statements, and rerun the computer assistance routine, which cleared the errors in every case. This certainly beats having to type the whole lot in!
At the next stage, I process the first page, check that the computer has interpreted it correctly (right way up, the usual order of Paid out and Paid in), and then switch to automatic running. It all worked correctly. This is my first semi-automatic job and marks another stage in development.
I then run Narrative Prediction (this is a VAT job with previous narratives already there). Only about a third of narrative is right first time on this job. Of the narratives that are left, about half can be generated from the function keys F1 … F10 because they have been programmed from previous VAT quarters. What’s left is usually filled in by Excel’s autocomplete system, so only a few narratives actually need to be fully typed in.
The whole sequence does not take long and has dropped below a level of worrying about it. This is a fast versatile system!
On other jobs, most bank statements are approximately A4 size, but have been folded in half to send them through the post. The crease is often a troublespot for OCR scanning, but worse than that, the resulting error rate is high enough to be an annoyance, but not high enough to induce the clerk to switch to typing it in. Whatever the clerk decides to do turns out to be wrong. This is just the sort of situation where the computer-assisted blink comparator can help out. The reader could have worked out anyway that most bank statements have creases in them (saves on postage), which causes a problem 50% of the time (thermodynamic principle of equipartition), which is an issue (like chucking sand into your petrol tank). We have solved this one in the real world.