• Stuart Tait, Partner |
5 min read

The debate rages on about the impact of artificial intelligence (AI) on individuals, organisations and society.

My last blog on AI introduced the three types of artificial intelligence that tax leaders need to be aware of: rule based automation, discriminative machine learning models and large language models (LLMs). It’s the latter of these that’s causing all the hype. So let’s look at this technology, and its applications for tax functions, in a little more detail.

LLM technology – also known generative AI – has genuinely transformative potential, for two reasons. It can create new content: text, images, video and music. And it can interact with users in a very ‘human’ way.

That makes it extremely useful, and highly accessible. This explains the phenomenal adoption rate of ChatGPT – the first ‘mass consumer’ generative AI tool. It took the internet 15 years to reach 100 million users, Facebook just four. ChatGPT got there in a few months.

Indeed, according to KPMG’s latest Global Tech survey, 57% of business leaders believe AI will help them achieve their goals over the next three years.

Cue a worldwide panic over generative AI replacing jobs in just about every sector and function – including those of tax professionals.

Expectation inflation

But along with the hype, there’s been plenty of attention on its limitations. Generative AI comes with risks that organisations aren’t yet equipped to deal with. That’s why our survey found that most organisations still have many unanswered questions, and can find it difficult to prioritise AI transformation in the right areas.

ChatGPT has been trained on pretty much the whole of the internet, which means it’s not especially good at answering specialised queries. It’s prone to errors, and to ‘hallucinations’ – making answers up. But given its natural language patterns, its wrong responses sound pretty convincing.

In our own testing, for example, we asked ChatGPT questions about certain areas of the UK tax code. Its responses were confident, and came complete with citations. The only problem being, they were nonsense: we’d asked about provisions in the tax code that don’t actually exist.

Gen AI also comes with data privacy risks. To be effective in an organisational context, generative AI needs to be trained on the organisation’s data – which means sharing that data with the third party providing the solution.

In my view, generative AI is going through the early stages of Gartner’s famous ‘hype cycle’ (see diagram).

Gartner’s Hype Cycle

Ground-breaking new technologies are often saddled with unrealistic expectations, which then rapidly sink (along with the hype) once its limitations become clear. But the experts who understand the technology continue to develop it, and to discover its practical uses. As a result, it comes into common use over time.

Right now, generative AI is at the ‘inflated expectations' point in the cycle. It will take a while to become a genuinely valuable tool for businesses. And getting the most from it will require human input. Tax professionals shouldn’t worry about their jobs just yet.

A pragmatic perspective

So what are generative AI’s practical uses, and how can tax professionals take advantage of them?

The technology’s applications stem from five broad capabilities:

  1. Content generation. Feed an LLM the right source information, and it can instantly produce the content you require: an email, presentation, precis of a meeting, etc. For tax teams, it could create research memos, summaries of technical positions and complex decision-making processes.
  2. Information extraction. Generative AI can also analyse large amounts of text, rapidly distilling and synthesising the essential information for you. For instance, it could isolate differences between late-payment clauses in hundreds of company loan agreements – in a matter of minutes.
  3. Smart chatbots: Until now, chatbots have delivered a clunky experience, where it’s obvious you’re interacting with software. But a ChatGPT-powered virtual assistant will ‘feel’ far more human. Combine that with generative AI’s content creation and information extraction capabilities, and you’ll have an effective business partnering tool. One which might, for example, automatically calculate the direct tax on the invoices raised by the finance team.
  4. Language translation: This one speaks for itself: generative AI promises to be significantly more powerful than Google Translate. The applications for tax teams in multinational businesses, working across numerous jurisdictions, are obvious.
  5. Code generation. Generative AI should transform tax teams’ ability to process and analyse data. Its algorithms can convert text inputs into most coding languages. So you could extract the data you need – say, a list of suppliers meeting certain criteria – from a massive database without having to write a piece of code.

Thinking through the fundamentals

Naturally, these are highly desirable functionalities for just about any business. But firms are taking very different approaches to harnessing them.

Some (KPMG included) are building bespoke, secure GPT environments, tailored to their particular needs and organisational context. Others are encouraging staff to experiment with opensource solutions (principally ChatGPT). In some rare instances, they’re banning generative AI together, as they feel it’s incompatible with their culture and practices.

Whatever your company’s stance, you should be thinking about how generative AI can make your tax function more effective. Start by asking yourself the following questions:

  • Does your organisation have a set position on the use of generative AI?
  • What risks and opportunities do you see in generative AI – for the business and the tax function?
  • What will implementing generative AI in the tax function mean for your resourcing requirements, capacity and costs (i.e. headcount, roles, skills and structure)?
  • What will it mean for your tax data strategy and risk policies?
  • How could generative AI help you comply with new tax regulation (e.g. BEPS Pillar 2, and country-by-country reporting)?
  • How might tax authorities and auditors start using it – and how would you respond?
  • Does your firm have a secure, generative AI capability – or are your employees using opensource applications? If opensource, what are the data risks?
  • How can generative AI help the tax function play a more strategic role in the business, and add greater value to it?

KPMG’s tax technology experts can help you think through these matters, and understand the possibilities that generative AI offers.

We’ll work with you to set your AI vision and strategy, then design and build the right operating model. We’ll help you integrate AI capabilities into your technology environment. And we can manage your access to them for you.

Please get in touch to see how we can bring the power of generative AI to your tax function.

Transform tax with KPMG.

  • Stuart Tait

    Stuart Tait

    Partner, Chief Technology Officer, Tax & Legal, KPMG in the UK

    Blog articles