Quick answer: Software R&D qualifies where a project seeks an advance in computer science or software engineering, such as a novel algorithm or architecture, not merely a commercially innovative product built on established frameworks. HMRC issued updated software-specific guidance in 2024, and the same advance-and-uncertainty test applies as for any other field of R&D.
What is software R&D?
Software R&D covers research and development work in software engineering and computer science. HMRC recognises that software development can qualify where the project seeks an advance in the underlying field, such as novel algorithms, new architectures, or substantial performance improvements not achievable by standard techniques. Software that is commercially innovative but technically built on established frameworks, libraries and design patterns typically does not qualify. HMRC issued updated software-specific guidance in 2024 to clarify the boundary.
How does HMRC define software R&D?
HMRC guidance on software R&D is at CIRD81960 and CIRD81970 of the CIRD Manual, and in the 2024 Guidelines for Compliance publication GfC3, Help to see if your work qualifies as R&D for tax purposes. The guidance illustrates typical acceptable and unacceptable software projects with worked narratives.
What does software R&D look like in practice?
A fintech SME builds a new payment routing engine whose core algorithm is a novel approach to latency-optimised transaction sequencing not found in published literature. The front-end and reporting modules use standard frameworks and are excluded. The qualifying R&D is the algorithm work and its associated testing, with the rest treated as routine development outside the claim.
How do you split a mixed software project for a claim?
A £250,000 development budget for a new analytics platform breaks down as £160,000 on a novel latency-optimised query engine (qualifying) and £90,000 on a standard front-end dashboard built with established frameworks (non-qualifying). Only the £160,000 enters the claim; at the merged scheme's 20% rate that produces a £32,000 pre-tax credit - a reminder that software claims for SaaS businesses are usually a fraction of total build cost, not the whole engineering budget.
Where do software R&D claims most often go wrong?
The most common HMRC challenge on software claims is treating ordinary commercial build work - gluing together well-documented APIs and frameworks - as R&D because the end product is commercially novel. The test is always the underlying scientific or technological advance, assessed against the same technological uncertainty standard that applies to every other field of R&D, not against how innovative the business model looks. The distinction becomes sharper as AI tooling makes ordinary integration work faster: using an off-the-shelf large language model to build a customer-facing feature is not R&D merely because the feature is new to the market, whereas fine-tuning a novel training methodology to overcome a specific, undocumented model limitation can be.
Frequently asked questions
No. Applying a well-documented model architecture to a new dataset is not automatically R&D; qualifying AI work usually involves a genuine advance in model architecture, training methodology, or a technical problem no published approach solves.
Identify which parts of the build seek a genuine technical advance, such as a novel algorithm or architecture, and which use standard frameworks and established design patterns. Only the qualifying portion, for example the core engine rather than a standard front-end dashboard, enters the claim.
No. Commercial novelty is not the test. Software that is new to the market but built on established frameworks, libraries and design patterns typically does not qualify; what matters is whether the project sought an advance in the underlying computer science or software engineering.