If we are scanning bank statements by optical character recognition, then we will find that some bank statements are better than others. For example, if we have pristine bank statements with “Balance brought forward” and “Balance carried forward” indicated on every page, then there are likely to be few processing errors, and often none at all. We are still going to insist that the accounts clerk views every statement on our blink comparator, but we will make it easy to cycle round the statements rapidly to check them.
We feel that this is as far as we should go with automated scanning. We will not program a system which just cycles round the statements automatically because that would be an abdication of responsibility by the human operator. OCR systems have some risk of going haywire, or some risk of the clerk looking like a sorcerer’s apprentice, and we will stamp on these risks by making the clerk look at every individual bank statement and take responsibility for its correct processing. This is likely to maximise the clerk’s productivity in the long run.
Our policy puts the emphasis on human control and graceful degradation when dealing with poor bank statements. It does not try to be too clever. We have to deal with OCR as an industrial process in the real world. Anything to do with OCR gives such a big improvement in productivity that we can afford to sacrifice a little time on the best jobs in order to get a decent result on the worst jobs. We want always to be able to do a standard job in a standard predictable time, and any minor underperformance on the best jobs can be rationalised by the principle that time spent in reconnaissance is seldom wasted.