What’s News?

Artificial Intelligence (AI) is very topical, and hundreds of variations of AI are already on offer to solve a variety of problems. There are innumerable arguments for and against AI in the audit profession. Many small and medium practices (SMP), ours included, watched its emergence intently, like a tennis match, without knowing whether it was relevant; how the regulators viewed the future with AI; how it could be used by staff to manipulate audit procedures; and how it could be used to the benefit of an audit. These questions can only be addressed, if one has, at the very least, used AI, the most popular being ChatGPT, and learnt, among others, some of its abilities and limitations.

I’m not writing this article as an expert in AI, but purely to pen my experiences with it. Keep in mind that I am still at the ‘dummy’ stage of progress, so use this article as a reference at your peril. Some aspects I’ve figured out over the last six months to a year, are not necessarily facts, but ways to approach AI in one’s business.

• It is a very fast way to obtain information from the internet
  • It saves time on research which would have been done on the internet anyway.
  • The information it obtains is mostly not complete.
  • It gets the ball rolling on new projects to point you in the right direction.
  • ‘Creative’ information not directly from the web is not accurate, especially in the audit field.
• It mostly writes better sentences than you do
  • It saves time when writing long texts.
  • It saves time when shortening long texts.
  • One can adapt the style or tone.
  • It requires editing, as its facts are not always correct, or may be communicated incorrectly in the sentence structure.
• It is not always complete
  • It’s better to do the basic outline of a request yourself first, before using AI, so that its completeness can be checked.
  • Where the results are technical and important, proper research or knowledge is required first, before using AI for possible new ideas that one did not think of.
• Its audit and accounting knowledge is limited
  • It sometimes uses wrong references. It ALWAYS needs to be double checked.
  • It sometimes quotes the wrong extracts from sections of an Act or standards.
  • It provides very limited accounting entries in a situation, when asked. The first accounting entries it gave me were wrong.
  • Audit procedures may steer you in a direction that you haven’t thought of, but they need to be determined by yourself first, with reference to the standards, before asking AI anything, in my experience. If it steers you in a direction such as, for example, ‘Estimates’, do the steps yourself from the standards, and only refer to AI for broad pointers you didn’t think of before.
• It does not advance your firm technologically
  • I would keep it to myself, and not advertise the fact that I use AI, even as mentioned above, as it gives the impression that your engagement teams cannot think for themselves. I would prohibit the use of AI by personnel, unless authorised by management or the engagement partner. Text from AI should be indicated as such.
  • The nuances and circumstances of an audit have a 100% chance of being unique. Therefore, the chances that instructions to AI will be 100% correct to obtain reasonable results are reduced. They are further reduced by the inability of AI to give proper and complete examples. The result is that your work in not at its required level of quality, nor have your engagement team learnt anything.
  • The profession’s mantra is, ‘If it is not documented, it did not happen’, and it is applied as such by the regulator. What nobody can determine or prove, is, ’If it is documented, did it really happen?’ This puts engagement partners and management in a difficult position when reviewing audit engagement files, and puts the firm at extreme risk.

In conclusion, I would say that one should focus the efforts of the engagement teams and firm personnel on developing their technical skills, and using AI as a double check. AI can save a great deal of time on administrative tasks or repetitive tasks, but I would not use it for anything that I cannot check, just by reading it. AI certainly has its limitations, and this fact makes me hesitant to use it without safeguards.