Op-Ed: AI in accountancy is a strange mix of comfort, streamlining, and big new issues

Accountancy is an odd and often very picky profession. Much of the AI-managed accounting information runs on rails from the transaction stage to formal entries in accounts journals. Account entries have protocols, as well as essential checks and balances, excuse the puns.

The current generation of accounting AI is also under continuous intense scrutiny for reliability and productivity and it’s gaining growing acceptance at various degrees of difficulty. It’s achieving this in the face of a type of obsessive professional wariness that makes other sectors look positively sedate.

Money does matter. If you’re an accountant and you sign off on an account, it’s all yours. The degree of responsibility is extraordinary, and it’s brutal. Everything on any account must stack up and be trustworthy before it goes to the C-level. This is not a sector where “AI slop” or “AI hallucinations” can be tolerated. It is a true test of operational business AI at the most fundamental level.

Yes, AI accountancy works, and it’s undeniably useful

Some routine daily AI accounting practices are very predictable and are really just doing the drudge work. This is clearly a significant plus, ironically “rehumanizing” the job for accountants and reducing the incredibly tedious ditch-digging element of accounts maintenance and upkeep. Accountants seem comfortable with this approach.

Important note: All of this level of work is done under human supervision. It’s only automated to the dashboard level, not the actual accountancy.

Interestingly, the AI is clearly and visibly productive and very efficient in this field. It’s not hard to generate efficiency metrics based on saved time compiling reports, organizing work, and even the potentially ultra-nitpicky stage of coding spreadsheets.

At that level, AI is winning the argument for efficiency, and it’s not actually replacing anyone. It can’t. Many accounting firms are in the process of adopting AI for their core businesses, including tax and other compliance-heavy work.

The problems with AI accountancy and their fixes, defined

It’s simply doing the jobs quicker, not checking, questioning, and verifying, which is the gut-level reality of accountancy. This work requires a hierarchical range of expertise which can’t be easily fitted into AI accountancy.

The argument that AI can’t replace accountants has merit, and it can prove it. The AI is not doing an intense inner critique of “Do I trust these numbers?” like a higher level accountant would be doing. It can’t ask where some of the numbers that show up on accounts come from.

For example:

Suppose a prim and pristine $600,000 appears on a government authority balance sheet out of thin air. The accounts are supposed to be strictly compliant with public accounting practices.

This adorable figure, which just happens to miraculously deliver a balance on the accounts, also comes with no explanation, no prior entries, and absolutely no other qualifying information. There’s not so much as a subatomic hint of where this number came from.

What is the AI supposed to do about a case like this? How can it respond, if at all? Can it even flag the account entry for attention?  

The above example comes from direct experience. This is what accountancy is all about. It’s also where old-style accountancy asserts itself, with any number of good and potentially expensive reasons. The level of pure fiction that can slither into accounts can only be understood by accountants.

This level of oversight also just happens to be pure best-practice AI management incarnate.

The trade-off here is that the AI has freed the accountant to focus clearly by managing the workload better.  It’s very much a matter of opinion whether AI can do forensic accountancy on any level. It can deliver numbers, but can it trace them?

AI and the future of accountancy

The big and potentially fatal mistake with the endless predictions of the future of AI in accountancy is assuming that automation somehow does the job. No, it doesn’t, and it can’t. Most accountants would agree that complacency is a recipe for fraud. Fraud doesn’t take holidays. Nor do serious mistakes on accounts suddenly become harmless due to automation, whether the mistakes are honest or otherwise. Even basic data entry can do a lot of damage.

AI is clearly about to become a major load-bearing asset in accountancy. That raises more than a few other issues:

Training: Will AI become another “ongoing education” element in accountancy? Probably, although at least the training can be structured to meet business needs. This essential training will have to be factored into future needs as a cost.

AI agents: This is tricky. AI custom agents are given a range of functions that will change constantly and may become redundant. At what point do AI agents have to be modified, upgraded, or retired? In accounts, their roles are critical at the most fundamental levels.

Compliance: There are two bandwidths here, and neither is simple. Compliance is not optional. Accounts that don’t comply are instantly and rightly suspect and lack credibility. AI regulation, in whatever form, is inevitable, whether from sovereign AI rules or from the usual tectonic changes in accountancy laws. This means configuring AI accountancy will have to comply with both.

The comfort zone in accountancy goes only so far.