There is a familiar pattern: a new technology or a new tool appears, promising to make all the complexity simpler, cheaper to handle, or removed altogether. Large language model (LLM, what most of us simply call AI) agents are the current example, and every department is under pressure to adopt them and deliver on the promises: faster processes, more automation, fewer errors, improved reporting and lower costs.
Now, we are not suggesting the promises are not real. They very much are… but they need to be carefully managed.
VAT is one area where the potential benefits are obvious, while the dangers may not be. Compliance carries so much administrative burden that, if AI can reduce it, the case for using it is compelling. The risk is when efficiency is confused with accuracy.
VAT rules can be detailed and full of exceptions; they change frequently and are highly dependent on interpretation. The correct treatment of a transaction depends on who the customer is, how the contract is worded, what the product is, how it is used, what local tax authorities expect in practice, and so on. These are not always questions with simple, mechanical answers.
Nonetheless, the LLM you use will often produce an answer that sounds authoritative. It will include caveats such as “subject to applicable rules” or “information should be reviewed”, but this cautious language is not evidence of cautious reasoning. What the tool doesn’t tell you is that it may have searched only a handful of sources without reviewing their reliability, that it didn’t check whether specific information is still current, and that it may overlook exceptions or contradictions just to deliver a coherent narrative.
There is no real debate about whether AI will replace VAT professionals – because it simply cannot! Not even to write this article, we promise. AI agents are exceptionally well suited to summarising information, drafting reports and helping someone navigate large amounts of data. They are useful assistants, but not a substitute for critical analysis and experience. Wrong answers can be an inconvenience in other business areas, but in VAT they mean penalties, audits and difficult conversations with tax authorities.
We are seeing the consequences of ill-advised use of AI more and more often in our work. Just recently, a multinational business approached us with what seemed a plain question: how should it invoice between two companies in its group? They had already made decisions about their business model and proposed supply chain. They presented us with lovely slides, figures and detailed “research”. There was just one problem – what looked clean operationally was a VAT nightmare. Their favourite AI model had not identified the risk. In fact, several considerations had been missed. We are now working to disentangle the consequences.
This is precisely where professional judgement matters: to examine the underlying transactions and assumptions, understand the business operations and assess the VAT implications before decisions are made. AI agents then enter the picture to support this work, because VAT professionals who use them wisely will set a much higher standard than those who do not.
The strongest model is therefore not one of competition between automation and expertise. It looks more like a division of responsibilities. Technology handles the repeatable tasks and identifies relevant data, and experienced professionals interpret the facts, challenge assumptions and manage the positions the business takes.
Yes, AI models are to be used, because VAT compliance will never be manual again, but it is not fully automated yet. The future will be enabled by technology, driven by data, but led by experts.
Before relying on AI-generated VAT advice to make a business decision, speak to Leyton. Our VAT experts combine practical, cross-border experience with technology-enabled tools to help you move faster, without overlooking the risks that matter.