R&D in Software & Technology

Key facts

Software R&D must seek an advance in computer science or information technology, not just build a new product.
Routine development using established frameworks and well-known patterns does not qualify.
The advance can be in algorithms, data processing, architecture, AI/ML, security, or integration.
Both product companies (SaaS, apps) and internal IT teams (bespoke systems) can qualify.
Staff costs for developers, data scientists, and DevOps engineers are typically the largest eligible expense.

What Qualifies as Software R&D?

For software and technology work to qualify as R&D for tax purposes, it must seek an advance in computer science or information technology by resolving genuine scientific or technological uncertainty. The advance must go beyond what a competent professional in the field could readily achieve.^[1]

Common areas where software R&D qualifies include:

Novel algorithms: Developing new computational methods to solve problems that existing algorithms cannot handle efficiently
AI and machine learning: Creating new model architectures, training approaches, or data processing pipelines where the outcome is uncertain
Performance at scale: Achieving processing speeds, throughput, or latency targets that exceed known capabilities
Complex integration: Connecting systems in ways that have not been achieved before, where the feasibility is uncertain
Security and cryptography: Developing new approaches to cyber security or encryption that advance the state of the art
Data engineering: Processing, transforming, or analysing data in novel ways that push beyond known techniques

What Does Not Qualify

The following activities are common in software companies but do not typically qualify as R&D:^[3]

Standard development: Building features using established frameworks (React, Django, Rails, .NET) without technological uncertainty
Configuration and customisation: Setting up off-the-shelf software or CMS platforms
UI/UX design: Design work is not science or technology (though the technical implementation of a novel interface might be)
System administration: Routine deployment, monitoring, and maintenance
Data entry and migration: Moving data between systems using standard ETL tools
Testing and QA: Routine quality assurance (though testing as part of resolving R&D uncertainty can qualify)

The key test: Ask “Did a competent developer, with access to publicly available knowledge, know how to do this before we started?” If the answer is yes, it is not R&D. If the answer is no, and you used a systematic approach to find the solution, it likely qualifies.

Examples of Qualifying Software R&D

Project	Why It Qualifies
Developing a real-time fraud detection engine that must process 1 million transactions per second with sub-10ms latency	Exceeds known performance capabilities; uncertainty about whether the required throughput and accuracy can be achieved simultaneously
Creating a natural language processing (NLP) system that accurately extracts structured data from unstructured legal documents across multiple jurisdictions	Novel AI challenge; existing NLP models cannot handle the domain-specific language and cross-jurisdictional variations with sufficient accuracy
Building a distributed data pipeline that maintains ACID compliance across 15 microservices with eventual consistency under 200ms	Architectural uncertainty; known patterns (saga, two-phase commit) do not meet the latency and consistency requirements simultaneously
Developing an image recognition system to detect sub-millimetre defects in manufactured components at production-line speed	Computer vision challenge; uncertainty about whether sufficient accuracy can be achieved at the required speed with the available image resolution

Typical Eligible Costs for Software Companies

For software and technology companies, the largest cost category is almost always staff costs. Eligible roles typically include:^[2]

Software developers and engineers (apportioned by time on R&D)
Data scientists and machine learning engineers
DevOps and infrastructure engineers (when solving R&D-related challenges)
Technical architects (time spent on R&D design and planning)
CTOs and technical leads (time spent on R&D activities)

Other common costs include cloud computing (for R&D workloads), specialist software licences, and externally provided workers (contract developers on R&D projects).

Tip: Many software companies underestimate their R&D because they think of it as “just development.” Review your sprint backlogs, Jira tickets, and technical debt items — any work that involved genuine uncertainty about how to achieve a technical goal may qualify.

Documenting Software R&D

Software companies are often well-placed to document their R&D because they already use project management tools and version control. Useful evidence includes:

Git history: Commit messages and branches showing iterative attempts to solve a problem
Jira/Trello/Asana: Tickets describing technical challenges and investigations
Confluence/Notion: Technical design documents and architecture decision records
Pull request reviews: Code reviews discussing technical trade-offs and uncertainty
Slack/Teams: Conversations where the team discusses technical obstacles

Frequently Asked Questions

Does building a new app or website qualify as R&D?

Not automatically. Building an app or website using established technologies, frameworks, and design patterns is standard development, not R&D. However, if the project encounters genuine technological uncertainty — for example, achieving real-time performance at unprecedented scale or developing novel algorithms — the work to resolve that uncertainty may qualify.

Can agile development be R&D?

Yes, agile methodology can involve R&D. The BIS Guidelines require a “systematic approach,” and agile sprints with clear hypotheses, iterative testing, and analysis can satisfy this. The key question is whether the sprint work addresses scientific or technological uncertainty, not whether the methodology is agile or waterfall.

Does AI and machine learning qualify?

AI and ML projects frequently qualify because they often involve genuine uncertainty about whether a model can achieve the required accuracy, whether a novel architecture will work, or how to handle edge cases in training data. However, simply using pre-built ML libraries without modification does not typically qualify.

Can I claim for fixing bugs?

Generally no. Routine bug fixing is maintenance, not R&D. However, if a bug reveals a fundamental technological uncertainty — such as a race condition that cannot be solved using known techniques — the work to resolve that underlying uncertainty could qualify.

Sources

Guidelines on the meaning of research and development for tax purposes — GOV.UK / DSIT
Corporation Tax: Research and Development (R&D) relief — GOV.UK
CIRD81900 – R&D tax relief: conditions to be met — HMRC