Save content
Have you found this content useful? Use the button above to save it to your profile.
Pixel design of a sword icon
istock_merovingian_AW

Generative AI: The ultimate double-edged sword for accountants

by

A new era of artificial intelligence tools offers accountants the tantalising prospect of slashing their workloads and opening up new horizons. But it could come at a heavy cost for firms that don’t tread carefully.

27th Jan 2023
Save content
Have you found this content useful? Use the button above to save it to your profile.

The past few months have seen the unlikely rehabilitation of chatbots. From jazzed-up spreadsheets designed to respond to a limited set of queries (and occasionally trapping unsuspecting users in an inescapable death loop), they’ve now become the latest must-have accessory in the business tech world – all thanks to the buzzword du jour, generative artificial intelligence (AI).

Spearheaded by ChatGPT, this new breed of generative AI-powered bots is currently surfing a wave of popularity with users and investors alike, with Microsoft reportedly ploughing a further $10bn into the project (while quietly laying off 5% of its human workforce).

The chatbots, designed to generate coherent, original outputs from a simple prompt and provide iterative responses based on further feedback, are certainly a step up from what existed previously. 

If you apply Gartner’s famous hype cycle to the phenomenon, we’re somewhere between ‘tech trigger’ and ‘peak of inflated expectations’.

Gartner hype cycle
Gartner

The hype cycle’s own descriptions of ‘early proof-of-concept stories’ and media interest, accompanied by ‘unproven commercial viability’ and ‘a lack of usable products’, make the rise of generative AI tools look utterly textbook, according to the Gartner model.

But does it really fit the template so neatly? The fact that just days after the Microsoft investment announcement, ChatGPT started to appear on its Azure platform, available for developers to play with, suggests that the computing behemoth has plans to strap boosters to the AI bandwagon and attempt to leap, Evil Knievel-style, over the trough of disillusionment.

In the accountancy world, we’ve seen a slew of practitioners kicking the tyres on all kinds of ways to use generative AI. One firm owner had the chatbot write to clients about why their fees have increased, while another attempted to visualise what it might be like to take one on as a member of staff (spoiler alert, it didn’t end well).

ChatGPT recently fell just short of a pass mark in an ICAEW assurance paper, and in a recent podcast finance director and Excel trainer Jess Slack outlined how the tool had a range of potential uses for the profession from Excel to marketing copy (although it didn’t do particularly well interpreting UK VAT laws).

Small firms in particular have taken notice. Similar to the adoption of cloud accounting tools, smaller, more flexible operators can move quickly to stitch in the latest AI software and steal a march on their clumsy competitors.

Speaking to small firm owners or partners during the last events season, more often than not their bottlenecks seemed to come around client communication and keeping up with correspondence, rather than the volume of actual accounting work. 

If there was a way of streamlining things, adding in a kind of ‘autocomplete on steroids' function to cut down the time it took to respond to clients, it could be genuinely game-changing. Client written in asking “when’s my next payment on account due?”. The bot can have a polished, polite reply in your drafts folder waiting for you to approve in seconds.

And taking it a step further, the idea of hitching up an AI chatbot to a practice management tool and a general ledger to provide automated responses to more complex client queries about their finances must have already occurred to at least one software developer.

Or how about the same tool acting as a kind of co-pilot during conversations with a client, piping up with prompts for the accountant such as ‘you might want to check Tom’s expenses this month’? 

If you can think of the idea, it’s probably already happening somewhere…

‘Your emails may be answered by a robot’

But as double-edged swords go, the brave new world of generative AI seems as sharp as a Samurai sword after slicing a particularly juicy lemon.

The accountant-client relationship is a delicate thing – carefully cultivated but easily crushed, built up over many years but squandered at the touch of a button. 

The system I’ve described above may sound enticing to a firm owner looking for tech-powered growth, but it comes loaded with a different kind of risk. Potentially a more damaging kind.

Ultimately, machines are still programmed to answer the question you’ve specifically asked it, not the one you think you have. Stuart Cobbe, the accountant who ran ChatGPT through the ICAEW assurance exam, told me the tool “gives answers with an air of confidence even when it’s completely wrong.” And as anyone in the profession knows, a wrong answer can sometimes be the difference between success and failure for a business – or a business relationship.

“It’s not afraid to give a garbage answer and back it up with garbage. It’s like having a fresh-faced junior who’s always convinced they’re right, so users need to approach it with a degree of caution,” said Cobbe.

Then there’s the ethical dimension. The phrase ‘can of worms’ doesn’t even begin to cover it. 

Generally speaking, most people accept that technology does a fair amount of heavy lifting in the modern business world, but accounting is still a people-to-people business. How would clients feel if they found out they were dealing with an almost completely automated system, responding to them? Would you be obliged to tell them this was the setup? It feels like it should be something more than a footnote in your terms and conditions stating ‘your emails may be answered by a robot’. What would the various accounting institutes or insurers have to say on the matter? And let's not even get started on these tools in the hands of the tax authorities…

More practical questions also abound. With sensitive financial information sluicing through the AI tool data centres located across the world, what will the data protection implications be? 

Whether we like it or not, a new era of artificial intelligence tools stretches out before us – one that could offer accountants the chance to expand their reach and remit beyond anything previously possible. 

However, these opportunities also come with risks, both known and unknown, and in such a fast-moving area of technology, the ground can shift pretty quickly. It’s fantastic that so many accountants are willing to wield the AI sword in their attempts to cut through the forest of work besetting the profession, but they’d do well to remember – that double-edged sword is sharp.

Replies (5)

Please login or register to join the discussion.

avatar
By RFL H
27th Jan 2023 10:45

As evidenced by Any Answers you usually need to ask at least 5 questions in response to any query from a client before you have understood what you are being asked.
Once AI can do that, and the further questions as required, I might be interested in it.

Thanks (2)
avatar
By mgbacchus
27th Jan 2023 15:18

If it is any slight consolation I tried using ChatGPT to answer a cryptic crossword and even when told they were cryptic clues, it was hopeless.

So we can settle back and do our cryptic crosswords whilst the AI does our day job, confident that there are some things we can still do.

Thanks (0)
By Duggimon
27th Jan 2023 15:24

You can't use an AI to answer client queries without human input, and you will never be able to. There are two insurmountable hurdles, first, if a client queries something about the terms and conditions of your engagement with them, the response given could be contractually binding on your firm. Second, you will be potentially personally liable for any bad advice given, and your PII insurer won't cover advice given by an automated system.

Client communication, unless heavily caveated up front of all responses, cannot be handled by an AI, and if all your responses to clients start off with "Hello, I'm an AI and nothing I say in this email can be construed as legally binding on your accountant, nor should any advice be deemed to have been given by them" then they'll bin it and phone you.

I can see a use to it in generating responses or finding information needed on trickier situations, but even then you'll have to check everything, and you always will or risk breaching the T&Cs of your PII.

Thanks (1)
avatar
By Hugo Fair
27th Jan 2023 15:30

Way ahead of you, Tom, I've left behind the ‘Peak of Inflated Expectations’ and am fully stuck in the 'Trough of Disillusionment' (on this topic) ... you might say that after too much PIE comes ToD!

I fear you are falling for the blandishments of the increasingly desperate technology sector - and should try to leaven your nourishment with a little intake from the coalface (as reported by the scientists in the field, not the marketeers).

New Scientist is a good starting place (science can be deep, but they explain and can be informative especially when you delve into their proper articles - not just the bite-sized snippets).

A couple of recent examples (from 14th Jan issue) to blow away the cobwebs:

* "AI will advise a defendant in court"
- the defendant is charged with speeding and will say only what the AI tells them to via an earbud;
- supplier is offering $1m to anyone with an upcoming case at the US Supreme Court to do the same!
* "AI chatbots could stagnate due to shortage of training data"
- the high quality language data used to train models like ChatGPT will run out by 2026;
- the advance in collective knowledge of AI could come to a halt soon after!

The fundamental problem (as I've mentioned previously) is the manner in which these AI bots will lie (to use a human concept) simply to achieve their purpose.
Bad enough in a legal setting, but the proposed solution to the learning dead-end (to teach the bots by using bot-generated texts) is fraught with more than mere danger.
Basically AI is still 0% of the way towards embedding human concepts like honesty, humility, self-awareness let alone balancing the good of one vs many ... not surprisingly since most humans still struggle with these, and both scientists & philosophers have failed to define them unequivocally.

Thanks (1)
Mark Lee headshot 2023
By Mark Lee
27th Jan 2023 16:17

I've had a good play with ChatGPT and even had it produce some useful material to inspire my articles, blog posts and talks. I have also had it abbreviate my own drafts - something I rarely make time to do!

Ok, so it's not (yet) perfect - and may never be. But it's already much better than many cynics seem to assume. What is a tad disappointing is how many people I have seen expressing a firm view re ChatGPT by reference to just one or two of its capabilities; and those who suggest they know how limited it is and that it will remain so.

This AI is surprisingly intuitive (all things considered). Try asking it how it could help accountants do certain tasks. And then ask it to explain what more it maybe doing to help accountants within a couple of years. Replies to such questions, along with the caveats it includes, are quite remarkable.

I'm not making any predictions here. And I'm normally a huge cynic re new shiny objects. This is evident from many of my blog posts re Google+, twitter, social media generally and more recently clubhouse specifically. But ChatGPT is genuinely exciting and could well be as much of a game changer as Google search was when it launched in 1997.

Thanks (0)