Is Your AI Innovation Falling Through the R&D Funding Cracks?

The UK is investing heavily in artificial intelligence (AI). It was recently announced that UKRI has committed a record £1.6 billion targeted at the AI sector between now and 2030, and the University of Cambridge’s DAWN supercomputer has received a recent £36 million upgrade as part of that continued push.

For companies in the Cambridge technology ecosystem, working across groundbreaking technologies ranging from 6G research and machine learning models for wireless systems to Internet of Things (IoT) intelligence, AI is now central to innovation. Much of this work involves overcoming genuine technical uncertainty, placing it firmly within the scope of R&D tax relief.

For innovative businesses, particularly in software, wireless, and AI, the UK’s R&D tax relief scheme remains an important mechanism for supporting growth. However, recent changes to the regime, combined with increased scrutiny from HMRC, are making it more complex to both identify qualifying activity and submit successful claims.

The Challenge of Defining R&D in AI and Software

Unlike traditional forms of R&D, innovation in AI is rarely linear or clearly bounded. Development often involves iterative experimentation: training models, refining algorithms, testing performance, and adapting systems to real-world conditions.

This makes it challenging to distinguish between routine development and genuine qualifying R&D.

Identifying Scientific and Technological Uncertainty

HMRC defines qualifying R&D as work that seeks to achieve a scientific or technological advance through the resolution of scientific or technological uncertainty. In AI, for example, that uncertainty may relate to:

• Improving model accuracy beyond known benchmarks
• Mitigating bias or limitations in datasets
• Scaling performance in production environments
• Integrating models into complex or constrained systems

What Does Not Qualify?

Not all such work automatically qualifies, particularly where established techniques are applied without advancing underlying capability. Examples of activities that would not typically qualify on their own are:

• Deploying a pre-trained large language model (LLM) within a new product
• Integrating a commercial API into an incumbent platform
• Applying an off-the-shelf machine learning (ML) framework to a known problem

Applying Specialist Insight

Determining where qualifying R&D begins and ends requires specialist interpretation of both the challenge and the baseline level of knowledge in the field.

Advisory firms specifically focused on R&D tax relief (and qualifying criteria) are better positioned to assess technical work against HMRC criteria and identify qualifying elements within complex projects. It is surprisingly easy to misinterpret qualifying activity in technical fields, and it can ultimately trigger pushback from HMRC.

Specialist consultancy Cooden Tax Consulting, a firm that has helped claim approximately £20 million for companies of all sizes, highlights how businesses often receive post-claim enquiries from HMRC even if they are adamant the claims are substantiated. As they point out, explanations of R&D may be dismissed if HMRC suspects the work to be “just improving your own knowledge and capability”, rather than seeking a wider technological advance.

There is a need for distinction between pivotal R&D and simply making your own organisation more efficient. If this commercial imperative supersedes everything else, any historical funds granted to you under false pretences may suddenly be claimed back.

Navigating the AI Compliance Hurdle

In an AI context, enquiries from HMRC could be the result of difficulties articulating technical uncertainty, misinterpretations of eligibility, insufficient technical details, or perhaps limited documentation or evidence. The crucial point is clearly understanding what constitutes qualifying R&D work in technology.

It must:

• Have a defined start and end
• Involve a clear attempt to create a technological advance through the resolution of scientific or technological uncertainty
• Surpass the level of knowledge that a competent professional in the field could simply deduce

Increasing importance has been placed on correctly identifying and evidencing R&D activity, since a series of reforms introduced by HMRC in 2023. A key development is the introduction of the additional information form (AIF), which requires companies to submit detailed technical descriptions of their R&D activities before a claim can be processed. This has shifted the process from a predominantly financial exercise to one that warrants structured technical justification.

In parallel, the government has introduced a merged R&D scheme, which brings elements of the RDEC and SME schemes closer together, alongside the Enhanced R&D Intensive Support (ERIS) scheme for loss-making, R&D-intensive companies.

These changes are intended to reduce error and fraud, but also increase the evidential accountability on legitimate claimants, particularly in fast-moving fields like AI where experimentation is often continuous and not always formally documented.

For example, companies developing:

• Neural network architectures for edge or wireless environments
• AI models operating under constrained or real-time conditions
• Intelligent AI systems that handle interference of dense IoT networks

These may all be undertaking qualifying R&D work, but the challenge lies in articulating the technical uncertainty involved, and evidencing how it was addressed through a systematic process.

Where R&D Claims Fall Short

A common failure point for many qualifying businesses is that they describe their work as a commercial innovation rather than a technical one. HMRC doesn’t assess the commercial success of a product, but whether the work involved a genuine technological advance achieved through resolving uncertainty.

Claims can fall short where:

• The technical challenge isn’t clearly articulated
• The narrative focuses on outcomes rather than processes
• Supporting documentation is inconsistent or limited

Another point of failure lies in incorrect cost capture data. For example, staff time is typically a significant qualifying cost, but it must be intricately and specifically linked to R&D activities, rather than broadly estimated. Similarly, costs associated with cloud computing, data processing, or software licences must be justified.

According to the latest UK Government statistics, the number of R&D tax relief claims has fallen by around 26%, while the total value of SME relief has decreased from approximately £4.8 billion to £3.15 billion. While this decline partly reflects efforts to address and clamp down on non-compliant claims, it also raises important questions of whether genuinely innovative companies are choosing not to claim due to the increased complexity and perceived risk.

For AI-focused companies, particularly startups and scale-ups, the combination of stricter compliance and technical ambiguity can make the process difficult to navigate alongside rapid development cycles.

Future Proofing Your AI Claim

As the R&D tax regime becomes more compliance-driven, the margin for error is narrowing. Successfully claiming relief now requires both financial accuracy and a clear, structured explanation of any underlying technical work.

For companies working at the forefront of AI and software innovation, this often means taking a more deliberate approach, ensuring that technical ambiguity or uncertainty is properly identified, documented, and aligned with HMRC.

R&D tax relief continues to play a critical role in supporting innovation across the UK’s AI and technology sectors. But as the regulatory environment evolves, so too must the strategy that companies deploy to claim it.

Rather than treating R&D tax relief as a retrospective compliance exercise, businesses may benefit by embedding it firmly within their processes and ensuring that qualifying activity is recognised as it happens, not afterwards. Those that adapt their approaches are more likely to continue benefiting from the support the scheme is designed to provide.