Claiming R&D tax relief for AI projects: A complete guide to what qualifies

R&D tax relief can be vital for AI (artificial intelligence) projects, offering huge benefits in R&D tax credits for AI. If your business is working with AI innovation, you could be entitled to a tax cut or cash payment to offset your expenditure and fund future work.

In order to benefit from qualifying AI development costs, it’s important to know which work and expenditure qualifies for the scheme. Many businesses miss out on claiming because they confuse routine activities with R&D.

HMRC criteria can be complex, but this guide will help you understand in simple terms what qualifies for R&D tax relief (AI). Find out which of your costs, staff and expenses meet R&D definitions under HMRC criteria, and how you can shape your claim for the maximum tax relief payout while remaining compliant.

HMRC AI R&D guidelines – is your AI project truly R&D? The core criteria

To demonstrate to HMRC (His Majesty’s Revenue and Customs) that you meet AI R&D tax credit eligibility, you’ll need to show that you seek to solve scientific and technological uncertainty in AI, and overcome uncertainty in your field. As the industry grows, there are always new opportunities to qualify for the relief.

When it comes to claiming R&D tax relief, HMRC applies three strict tests. Your AI project must pass all of them to qualify.

What is the ‘advance in science or technology’ test?

HMRC defines R&D as ‘an advance in science or technology’. This means your AI work needs to either develop new knowledge within the industry, or build upon the existing knowledge of the field in order to qualify for the scheme.

To claim software development R&D, your project must deliver an advance in the overall field of science or technology, not just benefit your business. In the context of AI, R&D tax relief for data science means creating something new that pushes the boundaries of what’s currently possible.

Examples of qualifying activities for AI tax credits:

AI projects that include hardware and software development, testing of materials, and data analytics commonly qualify for software development tax credits for AI, because they meet key criteria of AI innovation and advancement.

Examples of projects that qualify for R&D tax credits for AI because they create technological advancement:

• Developing a new machine learning (ML) algorithm – AI machine learning R&D tax relief is available when the project tackles a problem or creates an advancement
• Designing an advanced method of data processing to improve model efficiency – This may qualify when the work involves, and creating a solution that goes beyond established techniques
• Inventing a unique approach to AI-driven automation that goes beyond existing tools – Eligible for R&D tax credits for AI where the project overcomes scientific and technological uncertainty in AI, or develops new capabilities and advances in current technology

Read our AI machine learning R&D case study to find out how we helped our client secure tax credits.

Non-qualifying AI examples:

Not every use of AI will qualify for R&D tax relief. Simply applying existing tools or well-known methods, without seeking to push the boundaries of capability, will not count as AI R&D tax credit eligibility.

Examples of non-qualifying AI work include:

• Using an off-the-shelf AI tool that already exists – for example, deploying a pre-built NLP (Natural Language Processing) chatbot without making any technical modifications
• Routine application of established AI technologies – such as applying machine learning to scan documents or process data in ways that are already well understood
• Basic system integration – for instance, connecting an existing AI API or software into your current systems, without creating anything new

How can you identify ‘scientific or technological uncertainty’?

HMRC looks for evidence that your project tackled genuine technical uncertainty – a problem that a competent professional could not easily solve with current knowledge.

Qualifying AI examples:

• Integrating multiple AI systems in a way that has never been achieved before
• Resolving issues of data bias where no standard method exists
• Addressing performance limitations in training a new AI model that required novel solutions

What is the role of the ‘competent professional’ in R&D tax relief for AI projects?

A core part of HMRC’s test to see if you qualify for R&D tax credits for AI is whether the project required the expertise of a competent professional – someone with recognised skills in the field. If the problem could be solved with widely available knowledge, it won’t qualify.

A ‘competent professional’ in your industry will be able to:

• Determine what is or not an advance on the current knowledge of the field
• Identify which R&D activities genuinely seek to advance the industry

This reinforces the uncertainty test. To meet HMRC AI R&D guidelines, you need to show that your project required genuine specialist input and could not have been resolved by simply applying standard methods.

Qualifying for R&D tax credits for AI – what are the complexities?

AI projects often involve multiple moving parts, and not every element will qualify for R&D tax relief. However, HMRC AI R&D guidelines recognise that genuine AI innovation can arise at different stages of AI development, provided there is a clear attempt to overcome scientific or technological uncertainty in AI. 

Below are some examples of projects which qualify for R&D tax credits for AI.

Creating new AI algorithms and models

Creating new AI algorithms or pushing beyond standard models is one of the clearest qualifying areas for R&D tax credits for AI. If your team is solving challenges that cannot be resolved with off-the-shelf tools, this work could form the basis of an R&D claim.

This includes developing predictive models that go beyond established techniques, experimenting with generative AI systems, or designing complex neural networks that achieve outcomes not possible with existing approaches. 

Integrating AI systems to solve technological problems

While basic integration of an AI tool does not qualify for R&D tax relief for data science, custom integration that tackles unique technical barriers may. The key to qualifying for R&D tax credits for AI is that the integration must create a genuine technical advance, rather than simply connecting pre-built systems.

For example, adapting an AI system to work with highly specialised datasets, or combining multiple AI tools in a way that requires novel engineering, can involve significant uncertainty. 

Building data processing methods and architecture for AI

AI systems are only as good as the data that powers them. The focus is on work that goes beyond standard practice and is necessary to make a new AI application feasible. Where your project requires the design of new methods of data, or development of AI, you may be eligible for R&D tax credits for AI. 

This includes creating new data pipelines, innovative methods of cleaning and labelling datasets, or the development of secure architectures to handle sensitive information.

Edge AI and low-resource devices

Training AI to operate effectively in constrained environments is another area where you may qualify for R&D tax credits for AI. Overcoming these limitations requires fresh approaches and experimentation, meaning it ticks HMRC’s boxes.

This includes adapting AI models to run on limited hardware, optimising them for low-energy devices, or enabling performance with smaller or incomplete datasets.

Which AI activities don’t qualify for R&D?

Not all AI-related work will qualify for R&D tax credits for AI. HMRC is clear that routine tasks or use of existing technology do not overcome scientific or technological uncertainty. If your project only involves day-to-day implementation of established tools, it is unlikely to meet AI R&D tax credit eligibility.

Examples of non-qualifying AI activities include:

• Using existing pre-trained large language models (LLMs) without development – for instance, applying a model like GPT straight out of the box without making technical changes to how it works
• Data cleaning, labelling, preparing and annotating – while essential for AI, these activities are considered routine preparation rather than innovative R&D
• System deployment, scaling, or DevOps tasks – setting up servers, automating pipelines, or managing cloud environments does not qualify unless it involves solving a unique technical problem
• Adapting off-the-shelf APIs – connecting an existing AI API (such as image recognition or translation) to your systems without advancing the underlying technology
• Routine model optimisation without uncertainty – making small tweaks to existing algorithms where the outcomes are predictable and well understood
• Creating dashboards or interfaces – developing front-end applications to display AI results is classed as software development, but not R&D

FAQs: your most pressing questions about AI R&D tax relief for data science answered

 

Does the use of ChatGPT or other LLMs qualify for R&D tax relief?

Simply using ChatGPT or another large language model (LLM) in its standard form will not qualify under HMRC AI R&D guidelines, as it involves no scientific or technological uncertainty. If you adapt, extend, or train these models in ways that go beyond standard usage, then parts of that work could qualify.

Can we claim for projects where we use ready-made AI models?

Yes, but only if you are doing more than applying them in routine ways. For example, combining pre-trained models in a novel architecture, or training them to operate effectively on highly specialised datasets, may qualify. 

Which specific AI activities qualify?

Activities which qualify as R&D tax credits for AI usually involve overcoming uncertainty and making a technical advance like developing new algorithms, building novel data pipelines or architectures, and integrating AI systems in ways that solve unique technical problems.

What if my AI project fails to launch? Can I still claim under HMRC AI R&D guidelines?

Yes – your work doesn’t need to succeed in its R&D endeavours or become profitable to qualify for credits under HMRC AI R&D guidelines. As long as you intended to resolve a genuine scientific or technological uncertainty, the work can qualify.

Don’t risk your claim: why a specialist is essential

AI-related R&D claims can be complex to define as R&D. The line between routine development and genuine AI innovation is often blurred, and without the right expertise, it’s easy to miss out on valuable qualifying AI development costs.

At Alexander Clifford, we take the time to assess your AI R&D tax credit eligibility by identifying your qualifying activities, mapping your costs correctly against HMRC AI R&D guidelines and uncovering tax credit opportunities through the SME scheme, merged scheme or RDEC scheme you may have overlooked.

Whether you’re unsure what qualifies for R&D tax relief (AI), or you need help with an HMRC enquiry, get in touch with one of our R&D advisers today and let us help you turn your AI innovation into a successful, stress-free R&D tax credit claim.

Stop leaving money on the table. Book a free, no-obligation consultation with our R&D tax credit specialists to review your AI project today.