Can advisers rely on AI for capital allowances advice?

Artificial Intelligence (AI) has exploded in popularity over the last few years. What was once largely associated with technology businesses and specialist applications has rapidly become part of everyday working life, with tools such as ChatGPT, Copilot, and Claude now being used to research, analyse, draft and answer questions across almost every profession. For financial advisers, AI presents an obvious opportunity: the ability to ask a question in plain English and receive an immediate response can be particularly valuable when dealing with technical subjects that sit outside their day-to-day area of expertise.

The rapid evolution of AI is prompting growing concern among professionals, and the tax profession is certainly not immune to this apprehension, with many of the tasks traditionally undertaken by tax consultants now falling within the expanding capabilities of AI.

However, while AI has developed at an extraordinary pace, it is important to understand its limitations. AI does not necessarily "know" whether an answer is correct simply because it can provide one confidently and coherently. It can misunderstand the facts presented to it, rely on outdated information, overlook important exceptions or produce references that do not support the conclusion reached. This is particularly relevant in tax, where relatively small differences in circumstances can materially change the treatment of an item of expenditure. An answer that sounds entirely plausible can therefore still be technically incorrect. 

Where can AI help with capital allowances?

For financial advisers, AI can nevertheless be a very useful starting point when capital allowances questions arise. Capital allowances cover a broad and often complex area of tax, encompassing Plant and Machinery Allowances, Structures and Buildings Allowances, the Pooling Requirement, elections, and a range of more specialist provisions. Advisers will inevitably encounter circumstances where they are unsure whether a particular item or project may qualify, what legislation is relevant, or what questions need to be asked of their client.

AI can assist by helping an adviser frame the issue and identify potential areas for investigation. It can also help explain technical terminology, identify potential issues that may otherwise be overlooked and provide a starting point for further research.

Used in this way, AI can potentially save considerable time. Rather than starting with a completely blank page, an adviser can use AI to develop an initial understanding of the issue before turning to the underlying legislation and HMRC guidance.

The danger of relying on the answer rather than the process

The significant risk arises when AI moves from being a research tool to becoming the source of the advice itself.

Capital allowances are full of technical nuances. The treatment of expenditure can depend on factors such as the nature and use of the asset, the taxpayer's interest in the property, when expenditure was incurred, whether an election has been made, whether a previous owner has claimed allowances and the precise circumstances surrounding the acquisition or disposal. There are also important distinctions between legislation, HMRC's published interpretation and the facts of an individual case.

An AI-generated answer may identify the correct general principle but miss one of these critical details. Equally, it may provide a citation that appears authoritative but, when checked, does not actually say what the AI has suggested. The danger is therefore not necessarily an obviously wrong answer. The greater risk may be an answer that is almost right, sufficiently convincing to go unchallenged, but sufficiently wrong to affect the tax treatment.

This is why AI should not be treated as an authority in its own right. The fact that an answer has been generated confidently, or that it contains references to legislation and HMRC manuals, does not remove the need to verify those references and consider whether they actually apply to the circumstances in question.

Author experience

The capital allowances landscape is continually evolving, with legislation being amended, new provisions introduced and case law establishing or refining important principles. Keeping abreast of these developments and understanding how they interact with the existing legislative framework, is a significant challenge in itself. This is particularly the case where newly published legislation or guidance addresses only the more straightforward or anticipated scenarios, leaving more complex or unusual circumstances unexplored. In such situations, the practitioner may be required to interpret the underlying intention of the legislation or ruling and apply it to circumstances that have not been explicitly considered, often without the benefit of absolute certainty.

Even in the absence of new legislation, the breadth and complexity of the existing capital allowances regime means that it is unrealistic to expect any practitioner to have an immediate and comprehensive recollection of every relevant provision. The relatively infrequent application of some aspects of the legislation makes this particularly challenging. As a result, even experienced professionals routinely need to consult external sources, whether that be the Capital Allowances Act 2001, the HMRC Capital Allowances Manual, published technical literature, case law, or simply a discussion with a colleague or another industry specialist.

AI has undoubtedly emerged as a useful tool for navigating this extensive body of information where it can accelerate research and analysis and efficiently progress some of the more routine administrative responsibilities, allowing the practitioner to spend more time on the areas where technical judgement and experience add value. It can help identify relevant legislation, explore potential interpretations and provide a starting point for further research.

However, my own encounters have demonstrated that its output cannot simply be accepted at face value. There have been numerous occasions where my knowledge and experience have enabled me to identify responses that were fundamentally incorrect. On some occasions, AI has cited legislative provisions or other sources that either do not exist or have no relevance to the issue being considered. More concerningly, when challenged on these references, it has readily acknowledged the error, responding with comments such as:

“Excellent catch — you’re absolutely right to flag that. That was a mis-citation. Let me correct that for you.”

“You're absolutely right — thank you for pointing that out.”

“You're absolutely right again — and I appreciate your persistence in seeking accuracy. Let's clarify this fully…”

These responses illustrate both the usefulness and the potential danger of AI. Its ability to produce a confident and convincing answer does not necessarily mean that the underlying analysis, interpretation or supporting references are correct.

Without sufficient technical knowledge and experience to critically assess the output, an untrained professional could potentially accept an AI-generated response as authoritative and, in doing so, provide a client with incorrect advice. In the context of capital allowances, this could lead to incorrect tax positions being adopted and potentially influence significant commercial decisions, with consequences for a client's cash flow, tax liability and wider financial position.

A better way to use AI

The most effective use of AI for financial advisers may therefore be as an assistant to the research process rather than a replacement for it.

An adviser could use AI to ask questions such as:

• What capital allowances issues should I consider in this particular scenario?
• What additional information should I obtain from my client?
• Which areas of the Capital Allowances Act 2001 should I investigate?
• What HMRC guidance may be relevant?
• What are the potential arguments for and against an item qualifying?

The answers can then be used to guide further research against authoritative sources, including the legislation itself and current HMRC guidance. Where the issue is particularly complex, unusual or commercially significant, the appropriate next step may be to involve a capital allowances specialist.

This approach changes the role of AI considerably. Instead of asking "What is the answer?", the better question may be "What do I need to consider?"

That distinction is important. AI can help an adviser identify the questions that need answering, but professional judgement is still required to determine which facts are relevant, assess the reliability of the available guidance and reach an appropriate conclusion. 

AI alongside expertise, not instead of it

The increasing availability of AI does not diminish the value of existing technical resources or specialist expertise. In fact, it arguably makes the ability to critically assess information even more important.

For capital allowances, AI should sit alongside rather than replace the resources advisers already use. The Capital Allowances Act 2001, HMRC's Capital Allowances Manual, relevant case law and other authoritative sources remain fundamental to understanding the technical position. AI can help advisers navigate these resources, formulate questions and identify potential issues, but the underlying sources should remain the foundation of any technical conclusion.

There is also an important distinction between using AI to educate and investigate and using it to provide definitive tax advice. The former can potentially make advisers more efficient and better informed. The latter carries considerably greater risk if the underlying answer has not been independently verified.

The opportunity

AI is unlikely to disappear, and its capabilities will continue to develop. For financial advisers, the question is therefore less about whether AI should be used and more about how it should be used responsibly.

Capital allowances provide a good example of where AI can add genuine value. It can help advisers understand unfamiliar concepts, generate lines of enquiry, identify potential qualifying expenditure and structure further research. It can also provide a useful second perspective when an adviser is testing their understanding of a particular issue.

But the final answer should not simply be the answer produced by a chatbot.

The most appropriate model is one where AI accelerates the research process, authoritative guidance establishes the technical framework, and professional expertise applies that framework to the specific facts. Used in this way, AI has the potential to make financial advisers more efficient and better equipped to recognise capital allowances opportunities while retaining the technical rigour and professional judgement that complex tax matters require.