You might also be interested in
Replies (6)
Please login or register to join the discussion.
"If someone asks me why we have audited a particular sample, I can explain the computer-based technique which is a lot more robust than saying ‘one of my trainees picked ten transactions’"
It is still a random sample. Yes, it is generated by a computer programme, not human.
I hope this is the first in a series as the article seems focused on the AI applications related to audits. The vast majority of uk accountants don't do audits any more.
The great advantage of random sampling is surely that it is random.
I can see the point of targeted sampling re some substantive testing, monetary unit sampling of debtors was a targeted sample, with larger debtor balances having greater probability of selection , but for compliance testing I would get worried that I was being "gamed", especially if my use of AI for such a purpose was being discussed with clients.
Having said the above I have not worked in audit since 1999, and even then they were getting scarce re the firm's clients and tended to be more industry specific, so I likely am not the best versed in suggesting concerns, nor am I versed in modern audit approaches.
In similar vein it is even longer since I studied statistics, but I presume if a sample is not random then normal calculations re sample size and confidence go out the window and the AI amends the sample size to compensate for the fact that the sample is not random?
An interesting article.
AI and machine learning in the accountancy industry runs the risk of remaining nebulous. For accounting software to truly move into the realms of AI there has to be rich and well defined syntax and semantics to the data input into and generated by the system.
As a software engineer who cut their teeth in the specification driven communications/networking industry and who is now working in the accountancy profession, I find very little evidence of that being the case.
In general software needs strong contacts/interfaces and clearly defined data sets/data vocabulary upon which which meaningful decisions can be made by computing algorithms. Then the mass adoption of those software and data contracts allows large populations of users to benefit from the collective advancement of best practice.
The concept of double entry bookkeeping which is the fundamental principle which all current market leading accountancy products revolve around, is probably not enough to facilitate the introduction of AI technology to any meaningful degree.
A syntactic and semantic layers needs to be developed on top of double entry book keeping and this needs to be defined and published as an industry wide standard. Then software companies can build tools that might possibly give us what we think AI means.
"The concept of double entry bookkeeping which is the fundamental principle which all current market leading accountancy products revolve around, is probably not enough to facilitate the introduction of AI technology to any meaningful degree."
So for most small businesses with very simple transaction processes the development of AI in an accounting context is meaningless?