Save content
Have you found this content useful? Use the button above to save it to your profile.
A robot hand with the letter AI and a lady justice statue on the wooden table with law books. 3d illustration.

AI: Tax advice friend or foe?


By combining the technology underpinning generative AI with carefully curated tax databases, one VAT expert believes that artificial intelligence is already close to breaking into the tax advice world, opening up the prospect of whole-scale disruption in the market.

5th Jul 2023
Save content
Have you found this content useful? Use the button above to save it to your profile.

“Of all the taxes, AI is made for VAT. It’s a single transaction governed by specific articles, legislation or tax cases. It’s very logic-driven and the rules can be easily codified. AI works on data - the bigger the dataset the more accurate the results - and VAT has a colossal source of data to drive accuracy,” said VATCalc CEO Richard Asquith.

Former Avalara VP and Big Four VAT expert Asquith was speaking to AccountingWEB following the launch of VATCalc’s new VAT Advisor product. The tool uses OpenAI’s generative artificial intelligence systems, integrated into VATCalc’s legislative global tax calculation engine, to produce draft VAT advice with legal citations and tax cases in seconds.

Recent months have seen software companies tripping over themselves to make bold claims about adding new ‘generative AI’ tools to their products. For those with long enough memories, an easy comparison to draw is that of the first internet boom 25 years ago where companies were adding .com to their brand names simply to get their stock price up. 

Much of the latest AI boom coverage has either attempted to stoke fears that accountants’ time in the workforce is about to come to an end, or focuses on tasks that bear little relevance to how the tax and accounting world actually operates.

But VATCalc’s Asquith believes that behind the buzzwords is a technology that can add real value and efficiency to the tax advisory world – but also fundamentally change the way it has operated since it began.

Combining tax with technology

By combining the technology underpinning generative AI (large language models (LLMs) pioneered by the likes of ChatGPT parent company OpenAI) with carefully curated databases of information such as tax rules and data, Asquith believes that generative AI is already close to breaking into the tax advice world.

“This is where we are with our VAT Advisor product,” said Asquith. “We’ve created a global database of tax legislation, and we’ve codified it, so by combining the database with the LLM you can put any taxable transaction through and it will tell you which country is it taxable in, how to calculate the VAT (including the rates), how to produce a VAT invoice, tell you how to produce a VAT return to report transactions, and also spot if there’s a tax case relating to it.”. 

Using OpenAI’s chatbot, the software then puts a wrapper around this raw answer and turns it into tax advice. Examples of this could include confirmation of an EU VAT transaction, whether the transaction benefits from triangulation, a summary of who the parties involved in the transaction are and when the transaction takes place, a round-up of relevant case law, a summary of the rationale to communicate to tax authorities, and important information on what should be disclosed on the VAT invoice.

According to Asquith, the hardest part of the development process for its VAT Advisor tool was learning how to ‘prompt engineer’ correctly - in layman’s terms, how to ask the right questions.

Disruption in the tax advice market

Developments such as the VAT Advisor Tool open up the prospect of whole-scale disruption in the tax advice market, with the way advisory firms charge, train and operate all up for grabs. 

“If tax advisers are brave and open to new ideas they’ll adopt this quickly,” said Asquith. “While they won’t put outputs from models like this directly onto their letterheads and put it out to clients, firms should be adopting this internally both as an education and an efficiency tool.”

For junior staff more open and accustomed to using technology, Asquith believes that generative AI advisory tools could be a great tuition tool. 

“They’re all afraid to talk to partners and could spend hours interrogating the figures,” he said. “If they can get to a first draft quicker, then take it to a more senior person to review, it’s just more effective staff development.”

Pricing and operational conundrums

But just as AI tools dangle the efficiency carrot, they also raise the prospect of wholesale pricing model disruption for the market. 

Rather than manually pouring over transactions or case law and then charging for the time spent, the fact that advisers can simply use a piece of software to generate tax advice won’t be lost on clients looking for fast, inexpensive solutions.

“How would you recommend I charge my client for tax advice now?” said Asquith. “In the past, putting something like cross-border VAT advice together would take me three hours at a charge-out rate of £200, so that’s £600. Now I’ve just paid somebody a licence fee and clicked a button – how do you work it out? 

“Initially, firms will say: ‘I would have charged £600, so that’s the price and I take a nice margin’,” continued Asquith. “However, as similar tools are adopted elsewhere that advantage goes, pricing has to change significantly and profits could shrink dramatically.”

This change to the traditional ‘brain-time rental’ model for accountancy services could affect all parts of the profession, from the Big Four down to the smallest micro practice. Asquith posed a series of questions the profession may well find itself asking itself over the coming months and years.

“The work is still there but does this mean firms will have to charge less?” he said. “As services like tax advice commoditise to some extent, does this threaten my revenue? Do I need to cut my prices and reduce my profit? Do I need as many staff? Do I need to belong to an expensive global network if I already have the answers at my fingertips?”

For publishers of tax information, Asquith believes the release of generative AI tools could offer the potential to challenge accountancy firms in the provision of advisory services. 

“We could see a quiet seeping of spend from adviser to publisher,” he said. “They have the tax databases, information and case law. They have the expertise. All they need is the power of a generative AI plugin to be able to cut out the middle-men and offer tax advisory straight to the market.”

AI won’t put Excel out of business

While generative AI may be well-suited to taxes such as VAT, due in part to its data volumes and hard-coded rules, there are areas where Asquith believes it is unlikely to have much of an impact in the short to medium term.

“Corporation tax is much more complex, a lot of income tax work is already largely automated, but then it also has a selection of niche fringe cases where AI could struggle. 

“Excel won’t go out of business because of AI,” he concluded.

Replies (4)

Please login or register to join the discussion.

By JustAnotherUser
05th Jul 2023 15:19

Agreed, a few hiccups along the way but definitely see sci-fi becoming reality in the AI and Tax space right now and in coming years.

The two biggest issues that come to mind
-carefully curated tax databases
not only carefully curated but the cost to curate, peer review and continually ensure it stays relevant
can be solved with disclaimers when the AI gets advanced enough to identify if something can be interpreted differently.

Thanks (0)
By Hugo Fair
05th Jul 2023 17:48

Hmmm ... "Examples of this could include confirmation of an EU VAT transaction, whether the transaction benefits from triangulation .."

I may not be a VAT expert (to put it mildly) but that statement neatly encapsulates one of the as yet unresolved problems of LLM/database driven AI ... how do you keep it up-to-date (not just with new facts but taking on-board external factors that a human understands but software doesn't 'connect').

You may have read of the recent case where the bot was asked for the known views of QEII and her successor ... and nearly blew a fuse as it repeatedly asserted that a Queen has no successor whilst she is still alive (and refused to accept the possibility of it being ill-informed).

I'm reminded of the same kind of group-think espoused by HMRC when they claim that, say, 95% of RTI transactions are processed correctly ... and proceed to 'not worry' about the 1million+ 'bad' processing transactions every month.

Given the current unreliability of AI output (including its tendency to 'create' answers when it can't find anything suitable), anyone trialling this approach should expect to spend more (not less) time of their expensive experts in checking every output ... whilst juniors learn less and less as they rely on instant answers without rigorous understanding (as happened in the general population when log-tables and slide-rules were replaced by programmable calculators).

Things will, as always, change ... but the massive step change being promoted here is, for now anyway, mostly just hot air.

Thanks (1)
By ourpetsheadsarefallingoff
06th Jul 2023 10:21

As this technology is adopted by our industry, I struggle to see a future where junior and senior roles are maintained in anywhere near their current numbers. There's nothing in those roles that can't be replaced by bookkeeping software using a language model AI. I expect firms to be slashing staff numbers in the next few years if this is left unregulated.

Thanks (1)
By johnjenkins
06th Jul 2023 11:11

So what's the difference between AI and Google? Both will come up with the same answer as the whole thing is based on Law, fact and case history. I keep saying this, AI is not an all singing, all dancing, problem solver, it is simply a glorified calculating method, which is brilliant as it can do endless, time consuming chores. So let's not runaway with ourselves and create something that isn't there. Remember how the software people sold MTD to the Government?

Thanks (2)