How AI Is Changing HMRC’s Approach to R&D Tax Claim Compliance

Artificial intelligence is influencing how tax authorities collect information, identify risk and manage compliance. For businesses claiming Research and Development tax relief, this contributes to an environment in which accurate data, consistent documentation and transparent calculations are increasingly important. 

However, companies should avoid assuming that an AI system independently decides whether an R&D claim qualifies. HMRC’s publicly available guidance does not state that R&D tax relief claims are automatically approved or rejected by artificial intelligence. The clearer change is the wider transition towards digital submissions, structured information and data-led risk assessment. 

This transition has practical implications for every business making an R&D tax claim. 

R&D Claims Are Becoming More Data-Driven 

Historically, the information supporting an R&D claim could vary considerably between businesses and advisors. Some companies provided detailed technical reports and cost analyses, while others submitted comparatively limited supporting information. 

The introduction of the additional information form has created a more standardised process. Companies must now provide details about their qualifying expenditure and the projects included in the claim. The information must be submitted before, or on the same day as, the Company Tax Return containing the claim. If both submissions are made on the same day, the additional information form must be sent first.  

HMRC’s reforms also require claims to identify the agent involved in their preparation and to be endorsed by a named senior officer of the company. The measures were introduced to improve compliance and tackle abuse of the relief.  

Standardised digital information is easier to compare and analyse than unstructured documents. This means HMRC can potentially identify unusual figures, inconsistencies and recurring patterns more efficiently, whether through conventional analytics, automated checks or more advanced technology. 

How AI and Analytics Can Support Compliance Activity 

AI and data analytics can examine large volumes of information much faster than a manual review process. In a tax compliance setting, these technologies may help identify claims or behaviours that warrant further examination. 

Potential indicators could include: 

• R&D expenditure that appears unusually high relative to payroll or turnover
  
• Large changes in claim value between accounting periods
  
• Staff-cost allocations that are inconsistent with employment records
  
• Similar technical descriptions appearing across unrelated companies
  
• Claim values that differ significantly from sector patterns
  
• Inconsistencies between the additional information form, tax return and accounts
  
• Repeated claims involving activities that are rarely eligible
  
• Cost categories that do not appear to match the project descriptions
  

These examples illustrate the types of discrepancies data-led review could detect. They should not be interpreted as HMRC’s published criteria or evidence that any individual factor will automatically trigger an enquiry. 

A statistical anomaly does not necessarily mean a claim is incorrect. A rapidly growing technology company, for example, may have a legitimately significant increase in qualifying R&D expenditure. The important point is that the company should be able to explain the change and support it with credible records. 

Greater Consistency Across Submitted Information 

One of the most immediate consequences of digital compliance is the importance of consistency. 

A company’s R&D claim does not exist in isolation. Figures and statements may be considered alongside information in its: 

• Company Tax Return
  
• Statutory accounts
  
• Payroll records
  
• Previous R&D claims
  
• Additional information form
  
• Subcontractor documentation
  
• Technical project descriptions
  
• Other information already held by HMRC
  

The additional information rules require companies to provide specified details in support of their claims. Updated regulations introduced in 2024 also adjusted those information requirements to reflect the merged R&D expenditure credit scheme and enhanced R&D intensive support.  

If the technical narrative suggests that a project involves substantial in-house development, but the financial data shows little relevant staffing expenditure, the difference may require an explanation. Similarly, an R&D claim that grows substantially while the company’s technical team and development activity remain broadly unchanged could attract questions. 

This does not mean that every unusual result is wrong. It means the complete claim should tell a coherent and supportable story. 

Generic Narratives Are Becoming Riskier 

A weak technical narrative often relies on broad claims that a project was “innovative”, “complex” or “industry-leading”. These expressions may describe its commercial value, but they do not demonstrate that the project qualifies as R&D for tax purposes. 

HMRC’s guidance explains that qualifying work concerns projects seeking an advance in science or technology. The company’s business type and accounting period also affect which relief may apply, and HMRC warns that an ineligible claim may lead to a penalty.  

As claims become more structured and comparable, generic wording may become easier to identify. Businesses should therefore ensure their project descriptions explain: 

1. The relevant field of science or technology
   
2. The advance the project sought to achieve
   
3. The scientific or technological uncertainty involved
   
4. Why the solution was not readily deducible
   
5. The work undertaken to resolve the uncertainty
   
6. The outcome during the accounting period
   

Technical narratives should be specific to the company’s activities and prepared using input from the competent professionals who carried out or supervised the work. 

AI-Generated Claims Create Their Own Risks 

Generative AI can help organise notes, improve readability and create an initial structure for a report. It should not be used as a substitute for technical evidence, professional judgement or an accurate review of the underlying costs. 

An AI tool may produce text that sounds authoritative while including assumptions, invented details or language that does not reflect the project. It may also generate similar descriptions for completely different businesses. 

The risks of relying too heavily on AI-generated claim content include: 

• Overstating the technological advance
  
• Inventing uncertainties that the project did not encounter
  
• Confusing commercial challenges with technological uncertainty
  
• Using terminology that the company’s technical team cannot explain
  
• Including activities undertaken outside the claim period
  
• Producing narratives that do not align with the cost calculation
  
• Exposing commercially sensitive information through an unsuitable tool
  

Responsibility for the claim remains with the claimant company. The involvement of AI or an external advisor does not remove the need for company officers to understand and approve the submission. 

Increased Scrutiny Is Already Part of the Compliance Environment 

The government has acknowledged both unacceptable levels of error and fraud within the R&D relief schemes and stakeholder concerns about increased scrutiny of claims. In its November 2025 response concerning advance clearances, it reported a £578 million reduction in estimated non-compliance for 2022 to 2023 compared with the 2021 to 2022 estimate, following changes implemented by HMRC.  

HMRC’s broader strategy also continues to emphasise modernisation, closing the tax gap and creating a more digital tax system. Its 2025 to 2026 annual report describes progress towards smarter tax administration and a system that is more digital and easier to use.  

For claimants, this means compliance should be considered from the beginning of an R&D project rather than addressed only when the Corporation Tax return is due. 

How Businesses Can Prepare 

Companies can reduce compliance risk by strengthening their records and review processes. 

Maintain contemporaneous project evidence 

Keep technical records while projects are taking place. Useful evidence may include development tickets, design documents, test results, prototypes, meeting notes and records of unsuccessful approaches. 

Use reasonable cost-allocation methods 

Staff-time estimates and shared-cost apportionments should be based on a documented methodology. Unsupported percentages can be difficult to defend during a compliance check. 

Reconcile every submission 

The additional information form, Corporation Tax computation, accounts and supporting schedule should use consistent figures. Any genuine differences should be understood and documented. 

Involve technical personnel 

Engineers, developers, scientists and other competent professionals should review the project descriptions. Tax advisors should not determine technical eligibility without meaningful input from the people who understand the work. 

Check the correct scheme and rules 

The merged RDEC scheme and enhanced R&D intensive support apply to accounting periods beginning on or after 1 April 2024. Earlier periods may remain subject to the previous SME or RDEC rules.  

Review AI-generated material carefully 

Any content produced using generative AI should be checked against project records and financial evidence. Unsupported wording should be corrected or removed before submission. 

Transparency Is the Best Response to Data-Led Compliance 

AI may make it easier for HMRC to analyse information and direct compliance resources, but the fundamental requirements of a strong R&D claim remain the same. 

The company must identify qualifying projects, apply the relevant rules, calculate eligible expenditure accurately and retain evidence supporting its conclusions. It should also be able to explain how every important figure and statement was produced. 

Businesses should not attempt to make a claim look statistically ordinary or write narratives designed to evade automated checks. A legitimate R&D project may produce unusual results. The appropriate response is not to disguise those results, but to document and explain them clearly. 

As tax administration becomes more digital and data-driven, transparent R&D claims will be better placed to withstand both automated risk assessment and human scrutiny.