With an estimated 65,690 Research and Development (R&D) tax credit claims produced in the 2023 tax year1, each requiring long-form written reports and project narratives, it is no wonder that less scrupulous R&D tax agents have sought to utilise ChatGPT and other Large Language Models (LLMs) to assist in the often time consuming yet important report drafting process.
Some agents may be tempted by the ease at producing reports, particularly as many larger firms within this industry have a reputation for relying on quantity-driven approaches that sacrifice the proper time and attention that is required for ensuring claims are compliant. With the right prompts a report could be produced in seconds that could take hours to write manually.
However, due to the wider use and understanding of these tools, it has become more and more easy to tell when reports are AI-drafted, often due to specific phrasing and punctuation used. When you consider that HMRC’s R&D tax credit teams review thousands of reports a year, it is likely that they can easily spot when AI-assistance has been used. As there has been a significant increase in HMRC enquiries within this area over the last few years, and much criticism over how HMRC have handled these compliance checks, it is clear that any steps that could be taken to avoid an enquiry should be considered.
But should this be allowable with sufficient oversight? Many preparers would be unwilling to admit the role of such tools within their claim drafting process. The main issue is the prevalence of AI hallucinations, and the high likelihood of including information within a report that is not true or does not otherwise relate to a project. Even if an AI-drafted narrative was accurate, there’s clearly stigma around the use of these tools. Regardless of accuracy if a claimant pays a fee to an R&D tax advisor to prepare a claim, they’re unlikely to be happy if they find the report was produced using ChatGPT or with another tool that the claimant could have used themselves.
But how about use by HMRC? In recent weeks the Government released it’s “AI Opportunities Action Plan”: a national plan for AI use within government2. This plan details a range of potential opportunities and strategies for AI development and applications within the public sector. With a difficult economic situation and recent technological leaps within AI (including the release of China’s DeepSeek model), it is clear that it these tools are now being viewed as a potential solution with a range of uses within government. In particular, the plan notes:
- ‘Using AI assistants to do repetitive tasks better and faster, freeing up to 20% of an employee’s time’;
- ‘Drafting structured reports and forms with AI can cut final document production times by 20-80% in professional services’; and
- ‘Automated threat and anomaly detection’.
As the process of checking R&D tax claims consists of reviewing large amounts of qualitative data for accuracy, this forms an almost perfect use case for this kind of technology within government. As 20% of R&D claims are now enquired into, and the costs of fraud and error within R&D tax now reaching an estimated £1.3 billion3 a year, it is likely that these tools are being actively considered to ensure greater compliance more efficiently. In fact, there has already been much speculation about the use of AI tools within enquiries, specifically in HMRC’s process of researching existing scientific or technological baselines to a compare a claimant’s project to. With HMRC facing much criticism over the last few years in their approaches towards handling enquiries, it will be interesting to see whether any AI adoption improves or worsens the process for claimants.
It is clear that despite criticism these AI tools are being viewed as an opportunity by both R&D tax preparers and HMRC alike to provide R&D tax-related services more efficiently. Whilst for HMRC this greater efficiency may benefit the industry as a whole, it is ultimately the claimant companies that stand to lose out through either a lack of adequate care during the narrative drafting process, or AI-driven tools failing to understand their project’s advance in science and technology. Either way the input of competent tax advisors will become more valuable as the industry is more and more saturated with these tools in their varying forms.
If you have any queries on R&D tax relief or on the changes to compliance within this field, feel free to get in touch at rdtax@ct.me
- https://www.gov.uk/government/statistics/corporate-tax-research-and-development-tax-credit/research-and-development-tax-credits-statistics-september-2024 ↩︎
- https://www.gov.uk/government/publications/ai-opportunities-action-plan/ai-opportunities-action-plan ↩︎
- https://www.gov.uk/government/publications/hmrcs-approach-to-research-and-development-tax-reliefs-2023-to-2024/approach-to-research-and-development-tax-reliefs-2023-to-2024 ↩︎
