Alastair Hall, the UK Managing Director at FI Group, shares a cautionary tale about a prospect who used ChatGPT to write an R&D tax claim report.
As the UK Managing Director at FI Group, I recently encountered a prospect who was handling his R&D tax claims internally. To save time, he decided to use ChatGPT to write his report. On the surface, the output was impressive: a tailored report based on HMRC guidance with a detailed description of his technology. However, upon closer inspection, we identified several anomalies that could jeopardise the claim.
Firstly, the field of science or technology was not linked to a known field identifiable from the Frascati Manual. Throughout the report, the field of technology changed multiple times, which is a red flag for HMRC. Secondly, the baseline section did not reference other solutions and merely stated “no viable alternative,” a known bugbear for HMRC. Thirdly, the advance section incorrectly linked to DSIT paragraph 9a instead of 9b. Finally, the uncertainties section lacked the necessary information required by HMRC.
It would be tempting to trust ChatGPT, but our review revealed several technical problems with the claim. Another issue was the length of the descriptions. Each answer for the AIF was around 1000 words, which is too lengthy for HMRC, given their limited time to review claims.
The theory behind using AI to link technology to HMRC guidance is sound, but in practice, it falls short for several reasons. From a human perspective, the intellectual property of R&D tax consultants lies in knowing what HMRC expects and what information to use. Controlling the input with AI is crucial, and without detailed input, the output will never meet the required standards. R&D tax claims go into enquiry 17% of the time, causing an average of 246 days of cash flow delay and additional time from the business. The time saved using ChatGPT is negligible compared to the consequences.
Moreover, ChatGPT does not understand the tone HMRC prefers for these reports.
The problems and risks of using AI in R&D Tax Claims
- Lack of Contextual Understanding: AI may not fully grasp the specific requirements and nuances of R&D tax claims.
- Inconsistency: AI-generated content can be inconsistent, as seen with the changing fields of technology.
- Over-Reliance on Data: AI relies heavily on the data it was trained on, which may not always be up-to-date or relevant.
- Human Oversight Required: AI cannot replace the expertise and judgment of experienced R&D tax consultants.
Pros and Cons of Using AI for R&D Tax Claims:
Pros:
- Cost Savings: Reduces consultancy fees.
- Efficiency: Speeds up the drafting process.
- Accessibility: Easy to use and implement.
Cons:
- Inaccuracies: Potential for errors and inconsistencies.
- Compliance Risks: May not meet HMRC’s strict requirements.
- Lack of Expertise: Misses the nuanced understanding of experienced consultants.
- Time Delays: Increased risk of enquiries leading to cash flow delays.
In summary, while AI tools like ChatGPT can assist in drafting reports, they cannot replace the expertise and nuanced understanding of human consultants. The risks of inaccuracies, inconsistencies, and non-compliance with HMRC guidelines are too high. Therefore, businesses should be cautious and ensure that any AI-generated content is thoroughly reviewed by professionals.