How would generative ai be used in tax

Generative AI (GenAI) is likely provoking some big questions for you and the rest of your company’s leadership. What can it do for me? How can I trust what it does? How will it impact my workforce? How do I get started? And, critically, how can GenAI safeguard my data, intellectual property and proprietary processes?

Our business — like yours — depends on our clients’ and employees’ trust, so we wouldn’t use GenAI if all it did was boost productivity. We’re using GenAI because it can support this trust not just in the quality of our work, but also in our ability to deliver data protection, regulatory compliance and a successful future for our employees.

Based on our experience, here are four distinctive insights about GenAI that tax leaders should know about this powerful technology.

GenAI comes already trained on large amounts of documents and data, including financial statements, tax regulations and regulatory decisions. This pre-training can help you deploy and scale GenAI far more quickly than conventional AI. But it’s not enough for the accuracy and security that tax requires. Instead, you’ll need to licence a private version of one of these pre-trained GenAI models. With rigorous data governance, with safeguards, you can embed your own data into this private model. Enhanced with your financial data, GenAI can then provide highly accurate answers to prompts or requests that your tax professionals give it.

Your tax professionals can, for example, use a GenAI chat-based interface — powered by this mix of commonly available and secure, proprietary data — to access knowledge (such as the latest regulatory decisions), classify and standardise documents, analyse documents or data (such as historical audit outcomes inside your company) or generate reports. Let’s say you need to generate Pillar II disclosures for the first time. GenAI can study examples of reports, then input your data, follow instructions customised to your specific needs and produce a high-quality first draft.

As is true for all technology, it’s important to understand GenAI’s strengths and weaknesses. These AI models can be very good — accurate, fast and relevant — at tasks related to understanding, summarising and generating text and data. Even so, GenAI is best seen not as a pure automation tool, but as a way to help your team grow capacity and work better. For all GenAI, it’s important to build in checks and balances: oversight that will include your tax professionals reviewing and improving at least some of GenAI’s outputs. For sensitive or complex issues, GenAI will mostly provide first drafts or inputs (such as data analysis) that your professionals can then use to make their own work more detailed and accurate.

It may also be wise in some cases to combine GenAI with conventional AI or “old-fashioned” automation. Consider indirect tax transactional sales mapping. Your company may have many products and types of spend, with inconsistent descriptions. It could take your people a lot of time to classify and categorise them all. Machine learning can potentially do some of this work — but it often requires such large training sets, its use isn’t practical. GenAI can solve that problem by reducing this need for historical training data. Machine learning can then do the job, sparing your people this tedious work. A combined GenAI-machine learning solution is often also more dynamic. It can better adapt to new business scenarios and needs.

As GenAI proliferates, your tax professionals will still be in demand. The way GenAI works best in the tax function is when the approach is “human-led, tech-powered.” People make the key decisions and perform the important work, but GenAI is helping them like the world’s best assistant: finding and crunching financial data, reading and summarising documents (such as regulatory decisions), even writing some simple software when needed. But to work well with GenAI, most tax professionals will need additional skills. For example, they’ll need to learn “prompt engineering” — carefully crafting input questions to get accurate, relevant answers. Your team should also learn when they should and shouldn’t use GenAI, how to validate outputs and how to embed your data into your private GenAI model — securely and responsibly.

GenAI may lead to deeper changes too. Since it can help you do more work more quickly, that leaves more time to focus on strategic issues with the business — requiring more business knowledge by your staff. Tax technology teams may find they’re doing less programming (GenAI can write code) but more data science. Still, if you provide a good plan for GenAI, it may excite your team. 

At Edwards Fosse, we work in a tightly regulated industry. Our clients entrust us with sensitive data and our recruitment and retention depends on our ethics — including our ethical use of technology. That’s why we have developed a responsible AI framework that can govern GenAI and manage the risks from strategy through execution.

One potential (and avoidable) risk comes from using public GenAI models. These models may not always give you full control over how your company’s data and intellectual property is used and protected. But you can often negotiate a licence for access to a private version, then run it within a secure environment, which your data never leaves. This level of data security (a must for us at Edwards Fosse) requires both rigorous governance and careful negotiations with GenAI vendors.