Some clients tick off every transaction on the bank statements, and add a few annotations or explanations as well. This threatens to spoil our optical character recognition system.

With just a few ticks, our blink comparator will highlight any numbers that may have been scanned wrongly, and we can fix them manually. With many more ticks, we have now upgraded our software to be able to deal with them to some degree.

Typically a number 123.45 followed by a tick and handwriting will be rendered as 123.45 V*f3#. Our software can pick out the leading well-formed number so it becomes 123.45. A trailing well-formed number can also be picked out. Where a number is buried among extraneous symbols, such as */= 123.45 vH+&, then leading non-numeric characters will be ignored, the well-formed number will be identified and the rest discarded. This will fail to work if some of the symbols resemble numbers, so ZB 123.45 vH+& cannot be scanned correctly as a leading number because of the leading 28, but it will still be identified correctly as a trailing number.

This system is an improvement but not perfect. The blink comparator will still highlight anything that looks wrong, and further tests on the running total are performed down the line. If we get a pile of bank statements with lots of ticks and annotations, our policy will be to try scanning the first five statements to see what it is like. If it is not good, then we will just type in the bank statements by the column, which we now expect to be a rare requirement.

 
We now have a Staff Procedure Manual to cover the commonest things that we need to do. Our aim is going to be to combined advanced technology with being well-organised both on and off the computer. Much of the SPM consists of checklists and crib sheets, and much of it was produced on a spreadsheet, as one might expect from an accountant. We will add to or amend the SPM as we become aware of our own underperformance, the infelicities of other accountants, or Revenue and Companies House handouts.

The commonest error must surely be overlooking the need to register for VAT when turnover has exceeded the threshold. We have a standard checklist for this to be completed on every job which means that we have never done this ourselves. In addition, we will be reading the VAT chapters in the latest tax textbooks every September, and in September we will put a warning notice on file where turnover exceeds 75% of the threshold, which is based on a real case (Mark Mills-Henning).

There’s a procedure for getting Professional Clearance when we take over a client from another accountant. First we send them a letter. If there is not a timely reply, then we look them up on the Internet to see if there might be a reason. We can then e-mail them with a copy of our letter attached as a PDF. This is followed up with a telephone call. After this, we will produce a fax message and post it to them, hoping that it ends up opened on the right desk. After this we either complain to the relevant professional body or give up. Generally we don’t get past the e-mail stage.

If a procedure is not yet in the SPM, then there is a crib sheet which tells us to go slowly and carefully, keeping notes. Do online work outside normal working hours and print off, date and initial computer screens as we go. Back up frequently. Then produce the checklist based upon experience.

 
Now that we have scanned quite a few more bank statements, it is timely to summarise our experience with processing them by optical character recognition.

Generally bank statements are scannable. We rarely need to resort to typing them in by the column. Often there is some issue with the bank statements. They may be too faint, or they may have handwriting on them. A session on the computer-assisted blink comparator always provides a remedy, and is always much better than typing it in. We are glad we have this system.

Narratives are entered with the assistance of Narrative Prediction. This always speeds things up, but often it is less than wonderful. Maybe this is where we need to have some new ideas. One thing we do is to customise the F1 … F10 keys for each particular client, and this is a help. In February when there was not much to do, we had a session of tweaking the customisation for each client, and this should be a help as well.

Once the bank statements are captured in our software, processing is usually straightforward. Our mapping table has a direct link to the Internet so we can look up payments with strange narratives and no invoices on file.

Generally we are happy with the way things are going. We think our choice of a hybrid OCR/Narrative Prediction system is right, because most bank statements will be just not good enough for a pure OCR system. We do often need to deal with handwritten records which resemble handwritten bank statements, so we could do with Narrative Prediction and reprogramming the F1 … F10 keys anyway.