Software and AI continue to be among the most common areas for R&D tax relief claims, and among the most scrutinised. Innovation in this space is often fast-moving and commercially significant, but for tax purposes HMRC applies a much narrower test than “new” or “bespoke”.
Recent updates to the Guidelines for Compliance are helpful in clarifying where software development falls within scope, and just as importantly, where it does not.
Understanding Qualifying Work vs Commercial Objectives
For software activity to qualify as R&D for tax purposes, it must form part of a project that seeks an advance in science or technology through the resolution of scientific or technological uncertainty.
In practice, that usually means uncertainty around matters such as system architecture, performance limits, scalability, latency, or how complex components interact. The question HMRC asks is whether a competent professional could readily deduce how to achieve the outcome using existing knowledge at the time.
Using technology to deliver a commercial objective, even where the work is complex or commercially novel, does not of itself amount to R&D.
Routine IT Does Not Become R&D by Proximity
The guidance is explicit that routine IT activity does not qualify simply because it supports a wider R&D programme.
One example describes an in-house IT team carrying out a business-critical update to communications systems used across the organisation, including by staff working on qualifying R&D. HMRC does not accept that such work contributes to resolving technological uncertainty. The costs are therefore not allowable, even indirectly.
This point is often misunderstood. General infrastructure, platforms, or upgrades remain outside the scope of R&D tax relief unless they form part of a qualifying project and meet the specific definition of qualifying indirect activity.
Not all supporting software activity is excluded, but the boundary is tight.
Where software is developed solely to support a qualifying R&D project, and where that development falls within the defined list of qualifying indirect activities, related staffing costs may be allowable. The guidance gives an example of a bespoke internal application built purely to enable secure information sharing within an R&D project. The app itself is not R&D, but the development costs can qualify as indirect activity because they support the delivery of the R&D.
The distinction matters. Costs may be claimable, but only under the correct category and only where the link to a qualifying project is clear and specific.
Where Software and AI R&D Commonly Arises
Software R&D is most likely to arise where companies are working at, or beyond, the limits of existing capability. This may include:
- developing new approaches to machine learning model training, deployment, or optimisation
- resolving performance constraints in real-time or high-volume systems
- addressing system uncertainty created by non-standard integration of technologies
- adapting algorithms or architectures where established solutions are insufficient
By contrast, feature development, configuration, UI design, user acceptance testing, and routine integration using known methods will generally fall outside the scope of R&D.
Internal Software Is Not Excluded, but It Is Not Special
Software developed for internal use is not excluded from relief, but it is assessed on exactly the same basis as customer-facing products. There must be a clearly defined project, a genuine technological advance being sought, and uncertainty that could not readily be resolved using existing knowledge.
Internal tools built using standard frameworks, however valuable to the business, will rarely meet this threshold.
Evidence Matters More Than Labels
Because software development can be difficult to assess from the outside, HMRC places particular weight on evidence from competent professionals. These are individuals with relevant technical expertise who can explain, in practical terms:
- what was known at the outset
- why the solution was not readily achievable
- what uncertainties existed and how they were addressed
Describing work as “innovative”, “AI-driven” or “cutting-edge” carries little weight without this underlying analysis.
A More Disciplined Approach
Software and AI remain areas where genuine R&D takes place, but they are also areas where claims have historically been overstated. The updated guidance reinforces the need for careful boundary-setting, clear technical reasoning, and proportionate evidence.
R&D tax relief remains available where the statutory conditions are met. Applying those conditions with discipline, rather than enthusiasm, is now essential if claims are to remain credible and defensible.
Get in touch with one of our team as Asquith Bhondi if you want to explore this ever-changing landscape with an expert.
