Last year was undoubtedly the year of generative AI. And as we enter 2024, the race to embrace ChatGPT-style tools continues apace.
The transformational capabilities that gen AI offers won’t be lost on tax leaders – especially as tax increasingly goes digital.
AI can automate the processing and analysis of the large, complex datasets required by international regulations like BEPS Pillar 2 – strengthening compliance and reducing risk. It can enhance forecasting, which will drive better financial decision-making. It can also eliminate manual tasks and routine processes, freeing the team up to focus more on business partnering.
What’s more, it can be a springboard for innovation within tax functions. The insights derived from the vast amounts of data AI can handle may prompt new ideas for process improvements and efficiency gains.
A note of caution, however: while AI is a powerful tool, solutions are only as good as the data they’re trained on. Without accurate, validated and compliant data, you simply won’t be able to reap their benefits. In addition, ensuring that AI is used responsibly is an increasing focus for authorities, as the EU’s proposed AI Act demonstrates.
Explainability concerns
Of course, robust and granular data have always been essential in tax, whatever the technology that underpins them. But without it, the risks are higher than ever when using AI.
That’s because very few tax professionals can be expected to understand how AI tools work. So it becomes difficult – if not impossible – to explain the thinking behind your tax calculations, decisions and treatments. That leaves you without a transparent audit trail to show to regulators and other stakeholders.
Human oversight will play a key role in this regard. An element of human judgement in your AI-powered tax processes will have several advantages.
Crucially, it will mean you can explain your decisions, and address potential biases and ‘hallucinations’ (information fabricated by AI solutions). However, the foundation for explainability lies in high-quality, well-labelled data. Clean data acts as a transparent lens, helping humans understand how AI arrives at its conclusions, and mitigating the risk of misinformation.
Success factors
Given the risks, how can tax leaders harness gen-AI solutions in an effective and auditable way? And how can they bring in that crucial human input?
In our experience, there are four essential steps to successful AI implementation in the tax function:
1. Address the common data challenges
First, identify all of the data you’ll require. This will be made up of different data types (structured and unstructured), in different formats and from different sources.
Some of these sources will sit outside the tax function; you won’t own all of the necessary information.
Timely access is crucial – siloed data in legacy systems hinders AI. Make your systems can cope with increased data volume and complexity, driven by growing regulations and reporting requirements. And invest in data cleansing and quality checks. Remember: good data governance is critical for a successful AI implementation in tax.
2. Bring in people who understand data
Equip your team with the tools and expertise to ensure that the issues in step one are resolved.
You may need to work with external partners to ensure you’re receiving all of the necessary data, with the right quality. The ‘professional scepticism' these providers can bring will prove valuable as you go through your data journey.