Artificial intelligence (AI) is changing the way businesses innovate. From intelligent automation and predictive analytics to machine learning and natural language processing, companies across the UK are investing heavily in AI-driven solutions to improve products, streamline operations, and gain a competitive edge.

As AI adoption continues to grow, many businesses are also applying for HMRC Research and Development (R&D) tax relief to recover some of the costs associated with developing innovative technologies.

However, making a successful AI R&D claim is no longer as straightforward as it once was.

In recent years, HMRC has significantly increased its scrutiny of software and AI-related R&D tax claims. Applications that previously may have passed with limited technical information are now being reviewed much more carefully. Businesses are receiving additional compliance checks, requests for technical evidence, and, in some cases, rejected claims because they cannot demonstrate genuine technological advancement.

The biggest misunderstanding is that many organisations believe simply building or using AI automatically qualifies for R&D tax relief. In reality, HMRC looks far beyond the words artificial intelligence. What matters is whether your project genuinely solved a technological uncertainty through systematic experimentation.

Understanding these requirements before submitting a claim can save significant time, reduce compliance risks, and improve the likelihood of a successful outcome.

In this guide, we’ll explain the biggest HMRC AI R&D claims challenges, why these issues occur, how HMRC evaluates AI projects, and the practical steps businesses can take to prepare stronger, well-supported claims.

Understanding AI R&D Tax Claims

The UK’s Research and Development (R&D) tax relief scheme is designed to encourage innovation by helping businesses recover some of the costs of developing new technologies or overcoming difficult technical problems.

Contrary to what many people think, the scheme isn’t limited to laboratories or scientific research. Companies working in software, engineering, manufacturing, healthcare, finance, and many other industries may qualify if they undertake genuine research and development.

For AI projects, qualifying work often involves areas such as:

• Machine learning model development
• Predictive analytics systems
• Natural language processing (NLP)
• Computer vision applications
• Intelligent automation software
• Advanced data modelling
• AI optimisation techniques

However, using AI is not the same as developing AI.

For example, integrating an existing AI service into a customer support platform may improve business operations, but that alone usually isn’t considered R&D for tax purposes.

Instead, HMRC wants to see evidence that your team attempted to solve a genuine technological problem where the solution wasn’t readily available to competent professionals in the field.

In other words, businesses must demonstrate that they were advancing technology—not simply applying technology that already existed.

What HMRC Looks for in an AI R&D Claim

One of the biggest reasons claims fail is that businesses focus on the commercial success of a project rather than its technical challenges.

HMRC generally evaluates AI R&D claims by asking questions such as:

• Did the project attempt to achieve a genuine technological advancement?
• Was there a scientific or technological uncertainty that needed to be resolved?
• Could a competent professional have easily solved the problem using existing knowledge?
• Did the development involve experimentation, testing, or iterative improvements?
• Can the company provide technical evidence supporting its work?

These questions help distinguish genuine research and development from routine software implementation.

For example, creating a custom machine learning model capable of solving a problem that existing algorithms couldn’t address may qualify.

On the other hand, building a chatbot using an existing AI API with little or no technical innovation is generally considered routine software development rather than qualifying R&D.

Understanding this distinction is one of the most important steps in preparing a successful claim.

Why HMRC Is Scrutinising AI Claims More Closely

Artificial intelligence has become one of the fastest-growing areas of software development. As investment increased, so did the number of businesses submitting AI-related R&D tax relief claims.

Unfortunately, many of those claims overstated the level of innovation involved.

HMRC identified recurring issues such as:

• Claims based on routine software development.
• Weak or incomplete technical explanations.
• Poor documentation supporting project activities.
• Overstated development costs.
• Claims prepared without sufficient technical input.

As a result, HMRC introduced stricter compliance measures and increased reviews of software and AI-related applications.

Today, businesses should expect greater scrutiny than they might have experienced several years ago.

This doesn’t mean HMRC is discouraging genuine innovation. Instead, it aims to ensure that tax relief is awarded only to projects that meet the official eligibility criteria.

For organisations carrying out legitimate AI research and development, stronger documentation and clearer technical evidence are now more important than ever.

Common HMRC AI R&D Claims Challenges

As HMRC has tightened its review process, many businesses are discovering that preparing a successful AI R&D claim is far more demanding than simply describing an innovative project.

Below are the challenges that most commonly lead to additional compliance checks or rejected claims.

1. Demonstrating Genuine Technological Advancement

One of the biggest challenges is proving that the project delivered a genuine technological advancement rather than a commercial improvement.

Many businesses develop impressive AI-powered products, but from HMRC’s perspective, commercial success alone doesn’t make a project eligible for R&D tax relief.

Instead, HMRC wants evidence that the project pushed technology beyond what was already available.

During a review, questions often include:

• Did the project overcome a genuine technological uncertainty?
• Was the solution already available through existing software or frameworks?
• Could another competent professional have solved the problem using standard methods?
• Did the development involve genuine experimentation?

If the answer to these questions is “no,” the claim may struggle to qualify.

Example

Project Activity	Likely to Qualify?	Reason
Developing a new machine learning algorithm	Yes	Demonstrates genuine technological advancement.
Improving AI prediction accuracy through extensive experimentation	Possibly	Depends on the level of technical uncertainty involved.
Integrating ChatGPT into an existing customer portal	Usually No	Uses existing technology without advancing it.
Building a basic chatbot with pre-built AI services	Usually No	Considered routine software implementation.

This distinction is one of the most misunderstood areas of AI R&D claims.

2. Weak Technical Documentation

Another major reason claims fail is poor documentation.

Many development teams focus on building the product but forget to record the technical journey behind it.

Months later, when it’s time to prepare an R&D claim, they struggle to explain exactly what happened during development.

Good documentation should include:

• Technical challenges encountered
• Development objectives
• Testing phases
• Failed experiments
• Design decisions
• Code version history
• Project timelines
• Developer meeting notes

Without this evidence, it becomes much harder to convince HMRC that genuine research and development took place.

Keeping records throughout the project is significantly more reliable than trying to recreate them after the work has finished.

3. Confusing Commercial Innovation with Technological Innovation

This is perhaps the most common misunderstanding among businesses.

A project can be highly valuable from a commercial perspective without meeting HMRC’s definition of R&D.

For example, a company might build an AI-powered customer service platform that increases efficiency and improves customer satisfaction.

From a business point of view, that’s an excellent achievement.

However, if the platform simply combines existing APIs, cloud services, and commercially available AI models without solving a technological uncertainty, HMRC is unlikely to view it as qualifying R&D.

The key question isn’t:

“Did the project improve the business?”

It’s:

“Did the project advance technology through genuine experimentation?”

Understanding this difference helps businesses prepare much stronger claims from the beginning.

4. Incorrect Cost Calculations

Financial errors are another common reason HMRC raises questions during compliance reviews.

Some businesses accidentally include costs that don’t qualify under the R&D tax relief rules.

Examples include:

• Marketing activities
• Sales expenses
• General administration
• Customer support
• Software subscriptions unrelated to R&D
• Entire employee salaries without separating qualifying work

Instead, businesses should carefully identify costs directly connected to eligible R&D activities.

These may include:

• Salaries of developers working on qualifying R&D
• Employer National Insurance and pension contributions where applicable
• Software used directly during development
• Certain subcontractor costs (subject to the applicable scheme rules)
• Cloud computing or data costs where they qualify under current HMRC guidance

Accurate cost allocation not only improves claim quality but also reduces the likelihood of HMRC requesting additional financial evidence.

5. Writing a Weak Technical Narrative

Even when a project genuinely qualifies, businesses sometimes fail to explain it clearly.

A vague statement like:

“We developed an innovative AI platform for business automation.”

doesn’t tell HMRC anything about the technical work involved.

A stronger technical narrative explains:

• The technological uncertainty faced.
• Why existing solutions couldn’t solve the problem.
• What research and experimentation were carried out.
• Which technical obstacles were overcome.
• What advancement was ultimately achieved.

The goal isn’t to market the product.

The goal is to demonstrate the technical journey that led to the final solution.

Does Every AI Project Qualify for R&D Tax Relief?

The simple answer is no.

Artificial intelligence has become a popular business tool, but not every AI project involves genuine research and development.

The table below provides a quick way to understand the difference.

Your Project May Qualify If…	Your Project Probably Won’t Qualify If…
You developed a new AI technique or algorithm.	You only installed existing AI software.
The team solved genuine technological uncertainty.	The work involved routine software configuration.
Extensive experimentation was required.	No technical challenges existed.
Existing technology couldn’t achieve the required outcome.	Pre-built AI services solved the problem immediately.
Specialist technical expertise was essential.	The work focused mainly on business processes or design improvements.

Although this checklist provides a useful starting point, every claim should be assessed against HMRC’s published eligibility requirements and supported with appropriate technical evidence.

HMRC Compliance Checks Are Becoming More Detailed

In recent years, HMRC has significantly increased the number of compliance checks carried out on software and AI-related R&D claims.

Businesses may be asked to provide:

• Technical project reports
• Development documentation
• Financial records
• Evidence of experimentation
• Supporting calculations
• Explanations from technical team members

Receiving a compliance enquiry doesn’t automatically mean a claim is incorrect. In many cases, HMRC is simply requesting additional evidence before reaching a decision.

Businesses that maintain clear documentation throughout the development process are generally in a much stronger position to respond quickly and confidently.

How to Reduce HMRC AI R&D Claims Challenges

Although HMRC has become more rigorous when reviewing AI-related R&D tax relief claims, that doesn’t mean qualifying has become impossible. Businesses that prepare thoroughly, maintain strong technical evidence, and understand the eligibility criteria are far more likely to submit successful claims.

Here are some practical steps that can significantly improve the quality of your application.

1. Keep Technical Records From Day One

One of the biggest mistakes businesses make is waiting until the end of the financial year to gather information for their claim.

By then, developers may have forgotten important technical decisions, failed experiments, or the challenges they faced during development.

Instead, record information throughout the project, including:

• Project objectives
• Technical uncertainties
• Design discussions
• Developer notes
• Testing results
• Git commit history
• Meeting summaries
• Prototype versions
• Performance reports

Real-time documentation is much more reliable than trying to recreate events months later.

2. Clearly Separate R&D Activities From Routine Development

Not every task performed during a software project qualifies for R&D tax relief.

For example, a developer might spend part of their week experimenting with a new machine learning model and the rest fixing user interface bugs or handling customer requests.

Only the work directly related to qualifying R&D activities should normally be included in the claim.

Separating qualifying and non-qualifying work helps create:

• More accurate cost calculations
• Better supporting evidence
• Stronger compliance with HMRC requirements

It also reduces the likelihood of unnecessary questions during a compliance review.

3. Involve Both Technical and Financial Teams

Preparing an R&D claim shouldn’t be left entirely to accountants or developers alone.

The strongest claims usually combine technical expertise with accurate financial reporting.

A successful claim often includes contributions from:

• Software developers
• AI engineers
• Technical project managers
• Finance teams
• Tax specialists

Working together ensures that both the technical narrative and the financial calculations accurately reflect the work carried out.

4. Avoid Overclaiming

Some businesses believe that claiming as many costs as possible increases the value of their tax relief.

In reality, overclaiming often creates additional scrutiny.

Instead, focus only on activities that genuinely meet HMRC’s R&D criteria.

Be realistic when estimating:

• Staff time spent on R&D
• Development costs
• Eligible subcontractor work
• Software expenses
• Cloud computing costs where applicable

A carefully prepared claim supported by clear evidence is generally far stronger than an exaggerated claim with weak justification.

5. Build a Strong Technical Narrative

The technical narrative is often the most important part of an AI R&D claim.

Rather than describing how successful the final product became, explain the technical journey behind it.

A good technical report should answer questions such as:

Question	Why It Matters
What technological problem existed?	Demonstrates genuine technical uncertainty.
Why couldn’t existing technology solve it?	Shows that routine methods weren’t enough.
What experiments or testing were carried out?	Provides evidence of systematic development.
What technical obstacles were encountered?	Explains why specialist knowledge was required.
What advancement was ultimately achieved?	Demonstrates genuine innovation rather than routine implementation.

Writing clearly and factually makes it much easier for HMRC to understand why the project may qualify.

A Simple Pre-Submission Checklist

Before submitting an AI R&D claim, it’s worth checking that everything is in place.

Technical uncertainties have been clearly documented.

Development records and testing evidence are available.

Failed experiments have been recorded where relevant.

Eligible project costs have been separated from routine business expenses.

The technical narrative explains both the challenges and the solutions.

Financial calculations have been reviewed for accuracy.

Supporting documents can be provided if HMRC requests additional information.

Completing this checklist won’t guarantee approval, but it can significantly strengthen the overall quality of a claim.

The Future of AI R&D Claims in the UK

Artificial intelligence is expected to remain one of the fastest-growing areas of innovation across the UK economy. As businesses continue investing in advanced software, automation, and machine learning, demand for R&D tax relief is also likely to increase.

At the same time, HMRC is expected to continue strengthening its compliance processes.

Businesses should be prepared for:

• More detailed technical reviews.
• Higher standards of supporting evidence.
• Greater scrutiny of software development claims.
• Increased focus on genuine technological advancement.
• More requests for documentation during compliance checks.

For organisations carrying out legitimate R&D, these changes shouldn’t be viewed as a barrier. Instead, they highlight the importance of maintaining accurate records and preparing well-supported claims from the start.

Companies that combine strong technical documentation with realistic financial reporting are likely to be in the best position as HMRC’s review process continues to evolve.Keeping development software updated is also important throughout the R&D process.

FAQs

1. What is the biggest HMRC AI R&D claims challenge?

The biggest challenge is demonstrating genuine technological advancement rather than simply developing commercial software or using existing AI tools.

2. Can AI software development qualify for R&D tax relief?

Yes. AI projects may qualify if they involve genuine technological uncertainty, systematic experimentation, and the creation of new knowledge or capabilities beyond standard software development.

3. Does using AI tools automatically qualify for R&D tax relief?

No. Simply integrating existing AI platforms, APIs, or machine learning services into a product does not normally qualify unless the project involves genuine technological advancement.

4. Why are some AI R&D claims rejected?

Common reasons include weak technical documentation, insufficient evidence of experimentation, inaccurate cost calculations, and failure to demonstrate technological uncertainty.

5. How can businesses improve the chances of a successful claim?

Businesses should maintain detailed technical records, involve both developers and finance teams, accurately identify qualifying costs, and prepare a clear technical narrative supported by evidence.

Conclusion

Artificial intelligence is creating exciting opportunities for innovation, but it has also changed how HMRC evaluates R&D tax relief claims. Simply developing an AI-powered product is no longer enough. Businesses must demonstrate that their projects involved genuine technological advancement, addressed real technical uncertainties, and followed a systematic process of experimentation.

Although HMRC’s review process has become more detailed, well-prepared businesses have every opportunity to submit successful claims. Maintaining accurate documentation, separating qualifying activities from routine development, and presenting a clear technical narrative can significantly improve the strength of an application.

As AI technology continues to evolve, organisations that treat documentation and compliance as part of the development process—not something added at the end—will be better positioned to maximise legitimate R&D tax relief while reducing the risk of unnecessary enquiries or delays.