R&D tax relief for AI and data-driven businesses: key considerations under the new regime
Written by Emma Hussey
Innovation and R&D Tax Reliefs | Director (London Lead) at Azets UK
Emma Hussey leads the R&D Tax team in London, bringing with her a wealth of experience from her time at a Top 20 accountancy firm and a global innovation consultancy.
Artificial intelligence (AI) and data-led innovation are now firmly embedded in the commercial strategies of fast growth businesses. What was once confined to specialist software teams has become broad product and operational improvement.
Research and development in the sector is fast-paced. In some businesses, the innovation sits in the development of models or algorithms; in others, it lies in the systems architecture, data engineering, or the interaction and integration between AI capability and a wider product or process.
The UK’s R&D tax relief regime remains a significant source of support for innovative businesses. For accounting periods beginning from 1 April 2024, the UK introduced the merged research and development expenditure credit (“RDEC”) scheme, which provides a gross 20% credit, directly increasing EBITDA and valuation multiples. Loss making SMEs who spend 30% of their total expenditure on R&D are eligible for the enhanced R&D intensive support (“ERIS”) scheme, which offers an effective 27% payable credit as a vital cash injection.
For R&D tax relief, the central question remains: does the project seek an advance in science or technology (such as computer science or software engineering) and work to resolve scientific or technological uncertainty. It is now, however, accompanied by more rigorous compliance, restrictions on overseas and contracted out activity, and a sharp focus on project boundaries, technical narratives, and competent professionals.
Key points to consider:
Advancing AI vs Using AI: In many projects, AI and data are simply tools within a wider innovation programme (e.g. diagnostics, process optimisation, manufacturing). It is important to distinguish between when this work is advancing the field of computer science versus applying or adapting existing AI capabilities to a specific industry.
A key challenge is identifying where qualifying R&D ends and routine engineering or deployment begins. While significant effort may be spent tailoring models (e.g. setting parameters, training, testing), this does not necessarily indicate broader technological uncertainty. Applying AI to well-understood problems may rely on established methods, whereas use in complex, less understood areas (e.g. cancer diagnostics) is more likely to involve genuine uncertainty and potential advancement.
Claims should therefore be framed around the specific technological uncertainty, not the wider commercial programme or “AI” as a label. In practice, this means separating the uncertain technical core from routine software delivery activities through clear, contemporaneous evidence. Where the uncertainty lies in the wider system, process, or product, the claim should focus on that, with AI positioned as a tool used to resolve it, rather than the advancement itself.
Vibe coding: Although topical, the use of AI-assisted coding tools (i.e. “vibe coding”) does not, in itself, prevent work from qualifying. The use of AI-assisted tools to generate software code is generally analogous to using a more efficient development tool and would not, in itself, give rise to qualifying R&D. However, if vibe coding supports a technological advance and uncertainty, the associated costs may be eligible. The relevant consideration remains whether the project sought to resolve technological uncertainties that could not readily be deduced by a competent professional, regardless of whether generative tools were used.
Contracted-out R&D: Under new legislation,the party that intends or contemplates that the R&D should be done as part of a contract is the party able to claim. The company carrying out the work will often find that their R&D claim is not eligible.
For outsourced software development, bespoke customer solutions, platform builds and implementation projects, it is no longer sufficient to identify who performed the technical work; claimants must consider the contractual context, the allocation of decision-making and the extent to which they initiated the R&D when the contract was entered into. In practice, Statements of Work, customer agreements and scoping documents should be reviewed carefully before a claim is submitted. However, the key consideration will be how the arrangement operates in practice.
Data and cloud expenditure: Data licence and cloud computing service costs can be qualifying expenditure where they are employed in activities that directly contribute to resolving scientific or technological uncertainty. In practice, this can include expenditure on compute, processing and storage used in experimentation, model development, testing and other qualifying project work.
Routine hosting, commercial production environments, and non-R&D storage costs should be excluded, alongside costs for data which will be sold on or published. Where infrastructure is mixed-use, apportionments should be calculated and evidenced with usage logs, tagged environments, user activity or similarly robust records.
Overseas third-party activities: Expenditure on externally provided workers whose earnings are not subject wholly or partly to UK PAYE, and contractor payments for R&D undertaken overseas is excluded unless a narrow statutory exception applies.
To be eligible, R&D must be wholly unreasonable to conduct in the UK, for instance, for geographic or regulatory reasons – cost and labour availability arguments are specifically disregarded. For AI, software and data businesses that have historically relied on overseas development teams, this is now a key area of risk and one that requires early review rather than retrospective analysis at claim stage.
Intangible fixed assets: Where software and internally developed platforms may be recognised in the accounts as intangible fixed assets, a claim may still be considered for R&D tax relief. Expenditure that is capitalised in the accounts may still be revenue in nature for tax purposes and so can be eligible. It is important to examine the expenditure line-by-line rather than rejecting based on the financial statements alone.
Claim notification form: HMRC must be notified of intent to submit a claim for first-time claimants and those without a valid claim between six months after the accounting period end and two and a half years prior to it. Failure to do this will result in that accounting period’s claim being rejected.
Other R&D Incentives
Businesses with protectable IP, tangible CapEx or external investment plans should consider positioning these at the same time as the R&D claim is being compiled.
Patent Box permits a reduced 10% rate of Corporation Tax on qualifying profits attributable to patented inventions. For AI-driven businesses, patents can include core algorithms, technical applications, hardware-enabled solutions, control systems and product-specific inventions. Where patents are likely to play a role in the commercial model, it is advisable to align R&D tax and patent strategy at an early stage.
Research and Development Allowances (“RDAs”) provide 100% relief for qualifying capital expenditure on research and development. This is particularly relevant for investment in specialised computing environments, test infrastructure, robotics or other R&D assets, and should be considered as part of any wider claim review.
Enhanced limits for the Enterprise Investment Scheme (“EIS”) and Venture Capital Trust (“VCT”) scheme allow raises of up to £10 million in any 12-month period and £24 million in total, providing certain conditions are met (these limits are increased to £20 million and £40 million, respectively, for knowledge-intensive companies). A coherent explanation of the company’s innovation activity, IP trajectory and expenditure profile not only supports R&D tax claims but creates a stronger position for fundraising.
Strategic Considerations
Innovation tax now requires a more integrated approach. The strongest claims are those supported by good technical documentation, well-defined project boundaries, clear contractual analysis and a consistent treatment of expenditure. In the current environment, preparing a claim after the event with only a broad narrative is unlikely to be sufficient.
For many businesses, the most valuable step is simply to address these issues earlier in the process. Claims, incentive planning and supporting evidence are all more robust when they are built into the project lifecycle rather than reconstructed far after year end.
At Azets, we support innovative businesses with R&D tax relief, innovation incentives and practical advisory - from project scoping and cost analysis to claim governance and strategic planning. If your business is developing AI capability or data-led innovation, a proactive review can help ensure you approach is optimised for your growth trajectory.
FAQ
-
No. A project involving AI will only qualify where it seeks an advance in science or technology and involves genuine scientific or technological uncertainty. The use of AI by itself is not enough.
-
HMRC allows data licence and cloud computing service costs where they are used directly in qualifying R&D activity, but mixed-use costs must be apportioned reasonably and supported by evidence.
-
Not necessarily. AI-assisted coding does not change the underlying legal test. The relevant question remains whether the project involved the resolution of technological uncertainty.
-
Under the reformed rules, overseas expenditure on certain workers and contractors is generally excluded unless the statutory exception applies, and cost or labour availability alone does not satisfy that exception.
-
Under the current rules, it is generally the party that decided to undertake or initiate the R&D that is entitled to claim, rather than the party simply performing the work under contract. However, this is a nuanced area and specialist advice should be sought.
-
Not unless the whole programme is itself qualifying. HMRC expects businesses to isolate the activities that actually contribute to resolving the relevant uncertainty, rather than claiming the wider commercial project in full.
-
Capital expenditure is generally outside the scope of R&D tax relief, but accounting treatment is not determinative. Expenditure capitalised in the accounts may still be revenue in nature for tax purposes and eligible under s1308, while other capital R&D expenditure may instead be considered under RDAs.
