It is now becoming obvious what sort of technology we need. If client-provided information is in the form of bank statements in the usual format, then we will scan them in using optical character recognition. For certain minor banks, the narrative will be in some funny format which is too awkward to scan, so we will use our narrative prediction system instead. It’s full OCR for all the high street banks and some minor banks, and an OCR/NP hybrid for all other minor banks.

We consider that NP based upon repetitive narrative is superior to OCR, but NP based upon statistics is inferior to OCR, and we will develop a hybrid OCR/NP system which reflects this. Weak NP will be able to overwrite OCR only when there is some degree of correlation.

Bank statements in unusual formats, credit card statements, other printed matter and all handwritten records will be typed in by the column. We’ve replaced our numeric keypads with the best models according to Amazon, and we will encourage the clerk to type in numbers from paper records without looking at the screen. We will make it very easy to go back and correct mistakes after the end of each statement when the checking is done.

The two commonest mistakes are missing the decimal point and missing the minus sign to indicate income (the clerk normally works in debit-positive convention). The clerk can move the cursor back to the error, and will have a choice of typing in codes from the keypad like *00 and *+, or of clicking on buttons on an onscreen toolpad which are also labelled *00 and *+ with reminders alongside of what these codes do.

If the clerk types in *** from the numeric keypad, then it will be possible to cycle around from numbers to dates to narratives and back to numbers. There will also be onscreen buttons with *** to click on as an alternative. This “triple star” system might look a little gimmicky, but it should give the clerk a feeling of being in control and of having decent technology when OCR cannot be used.