You may already be fed up with hearing all about ChatGPT – how many twee rhymes turned up on your socials in December?
So please forgive another post on the subject, but I do really believe that ChatGPT and similar tools have the potential to help finance teams work better.
I think this is one of the best articles about ChatGPT, because it recognises that most of us are on a journey and urges us to get past the “isn’t this funny / stupid” stage. The same author concluded elsewhere that you should try it out in areas that you are an expert in, and be extremely careful about using it in areas you are not.
After a fair bit of playing, I would agree with this conclusion. Here are a few results of my explorations.
It’s really good at Excel and I’m already using it more than Google
As an Excel trainer and consultant, I was obviously pretty keen to see how good it was in this arena.
One of my first searches was to ask it for an Excel formula to separate first names from surnames. It provided a reasonable answer, but one that wouldn’t work with double-barrelled names. So I simply asked it “Will this work for double-barrelled names?” and it provided a more complicated but accurate alternative. The fact that you can test the answer quickly and troubleshoot using natural language is a real game-changer.
An important caveat for this specific use is that the training data is a year or two out of date and therefore doesn’t include newer Excel formulas. So in this example, it gave me an old fashioned answer using LEFT, FIND, LEN etc whereas you might now approach the problem using formulas such as TEXTBEFORE, TEXTAFTER or TEXTSPLIT which were released in 2022.
It is bad at VAT
Maybe it’s because VAT is complex, but ChatGPT mixes up various aspects of VAT and ends up with sentences that are either nonsensical or just plain wrong.
Interestingly, since I first did some trial prompts a month ago, the language has become a lot more caveated, with a reminder at the start and the end that it is complex and you should seek specialist advice.
It’s a really useful tool for writing, with the right prompts
Lots has already been written about how you could use it as a ideas generator. Or to get a first draft of something down that you can then edit. Or it can help you rewrite something.
Writing the right prompts is an art in itself, although the iterative nature of the tool will usually get you there eventually. The default style is extremely bland and generic. For example, I asked it to tell me how chat GPT could help finance teams. The initial answer was not massively inspiring either in terms of content or style.
However, if you ask it to write in the style of Nigel, a cynical technophobe, or Asma, a finance manager who is really enthusiastic about technology, you’ll get a much more distinct tone of voice, although the actual content still reads like someone trying to generate key words for SEO purposes rather than an informative piece. You can drill further by asking it to give specific examples, or expand on a particular point.
I was more impressed when I asked it to write a complaint letter to HMRC, giving it a couple of bullet points to cover. It generated a professional sounding letter quicker than I could do manually, and I am someone who writes these kinds of letters reasonably often and am confident about my writing. I can see that ChatGPT will be invaluable to people who are not confident about their writing.
Will it write my financial reports?
I copied and pasted an extract from some management accounts into the prompt bar (literally just from the excel tab, no edits or reformatting) and asked it to write a report.
It was initially a little pedestrian and repetitive but with further prompts (“You don’t need to mention the current month figures or budgets, only the variances” or “Add a note explaining where the biggest variances were”) it got better.
This still may not seem that impressive – it is just regurgitating numbers into generic text, which you could do with existing Microsoft technology. However, if you combine this with where a tool like Analyze Data in Excel might go, you can see that in the not too distant future it’s likely that an add on to your accounting software could;
- review the underlying transactions to find unusual transactions or probable causes of variances
- email the relevant budget holders for explanation
- incorporate the responses into an overall report.
What does all this mean for a forward-thinking finance team?
Even if the product I’ve suggested doesn’t yet exist, all the component pieces do and the pace of change will only get faster. It’s important for us to keep a close eye on developments and be open to change.
I feel like I’ve only just scratched the surface with ChatGPT and machine learning more generally, and I’m excited and a little nervous about what comes next.