After doing a few more jobs, it is becoming obvious where we are going with new technology. We have two principal new technologies which use artificial intelligence, namely optical character recognition and narrative prediction. We will use them like this :
The main high street banks will have their statements processed using full OCR. Obviously with them being the banks we see every day, there is an incentive to work out how to use OCR all the time, even if some of the narrative on some of the bank statements is a little difficult to interpret.
Other banks and building societies acting as banks will also have their statements processed using full OCR if we can manage it, but we won’t try too hard. Instead we will often use a combination of OCR on numbers and dates only, and NP for the narratives. Some narratives are so convoluted that trying to read and interpret them using OCR will just be futile.
With bank statements in unusual formats, we will just type them in by the column using the slick system we have developed. NP can still be used effectively. Any credit card statements will also be typed in. No doubt we could work out how to use OCR, but there would be so much faffing about with weird formats that typing and a bit of NP will be quicker. It is often the case that most credit card payments are for petrol or Diesel fuel, and we can reprogram the F9 key on the keyboard to generate the appropriate narrative even before using NP.
We will keep a list of banks and their recommended processing method so the clerk can consult it and then just get on with it.
With handwritten records resembling bank statements, again we type them in by the column, not by the row, and can make good use of NP.
With piles of invoices we just type them in by the row, but we have a fast system to do this. Often all we type in is a day and an amount, and month, year and narrative are copied down from the previous entry.
We have a working OCR system which does nicely with the top three banks. Our NP system needs a bit more work done on it considering its intended ubiquitous use, but this will be worth it. We will in the future develop a hybrid OCR/NP system where the narratives are first scanned in with OCR, and then selectively overwritten or corrected with NP where there is corroboration.
In practice, we have got to where we want to be already, and further development work will be in the spirit of continuous improvement (“kaizen”) rather than a life-and-death struggle at the frontier of technology. Usage of this system is never boring for the clerk since there is some variety and plenty of flashy new technology to play with. It is then just a matter of taking a decision about processing and getting on with it. We would be perfectly competitive if we never used full OCR, just the combination of OCR on numbers and dates and NP.
What would it be like if we stopped used OCR completely? Well, it would start to be tedious to have to type in all the bank statements, although that is still better than having to analyse them in the traditional way. We would still have a lot of other technology and could probably match the performance of other accountants who use other OCR systems because our systems are more integrated. What would be the main casualty is our imagination, because OCR work provides a stimulus to develop other systems to complement OCR.