Skip to content
R&D Tax Academy
Contact us

Software development and R&D tax incentives: what counts and what does not

Software projects produce some of the strongest R&D claims and some of the weakest. The rules are not hostile to software; they are hostile to routine development described as research. Here is how to tell the difference.

Published 02 Oct 2026 · 9 min read

Most software projects contain some work that would satisfy any of the seven schemes and a great deal that would not. Teams that claim the whole project get challenged; teams that identify the technical unknowns and claim the work that addressed them generally do not. The distinction is not about how modern the technology is. It is about whether the outcome was uncertain to people who knew the field, and whether the team had to try things to find out.

Every scheme covered here has published guidance on software, and the guidance says the same thing in different words: building something new with known methods is development, not R&D, however commercially novel the result. The sections below set out where the line falls, how to find the eligible part of a project, and how to keep the evidence that software teams already generate.

What this guide covers

  • The kinds of software work that tend to qualify, and the kinds that tend not to.
  • A work stream table showing what moves a stream from one side of the line to the other.
  • The evidence software teams already generate and how to keep it.
  • How each of the seven schemes treats software, including software built for your own administration.

What tends to qualify

  • New or substantially improved algorithms where published methods did not meet the requirement.
  • Architecture work to achieve performance, scale, reliability or security targets that the available approaches could not meet, proven through systematic testing of alternatives.
  • Novel data processing, machine learning models or inference approaches where the outcome could not be predicted from existing knowledge.
  • Integration at a scale or under constraints where standard patterns failed and new techniques had to be developed and tested.
  • Extending the capability of a platform beyond what its design or documentation supported, where the extension required experimentation rather than reading the manual.

What tends not to qualify

  • Building standard applications on established frameworks, however large.
  • Configuring or customising commercial software using documented features.
  • Routine bug fixing, refactoring, upgrades and maintenance.
  • User interface design, user research and usability testing.
  • Porting to a new platform or language using known methods.
  • Data entry, content and routine data migration.
  • Software built for the business's own administration, such as payroll and accounting systems. Several schemes exclude this outright or test it more strictly; see Country differences.

The commercial importance of a project has no bearing on either list. A product can be the first of its kind in its market and still be built entirely with known techniques, and a minor internal tool can contain a genuine technical uncertainty. Reviewers in every scheme look past the product to the method.

Where the line falls, stream by stream

Most software projects break into a handful of work streams. The table shows the usual position of each and the specific thing that would move it across the line. The right-hand column is what the project note and the evidence need to show.

Work streamUsual positionWhat would move it across the line
Application build on an established frameworkOutsideA specific capability the framework could not deliver, with documented attempts to achieve it before a new approach was developed
Integration between systemsOutsideStandard patterns failed under the actual volume, latency or consistency constraints, and a new technique had to be designed and tested
Performance, scale or reliability engineeringDependsA target that published approaches could not meet, benchmarks showing the gap, and experiments with alternatives, including failed ones
Algorithm or model developmentOften insideEvidence that existing methods were tried or assessed and fell short, and a record of the hypotheses tested
Security and cryptographic workDependsA threat or constraint that standard libraries and protocols could not address, rather than correct configuration of them
User interface, user research, usability testingOutsideRarely moves; design preference is not technical uncertainty
Bug fixing, refactoring, upgradesOutsideOnly where the defect exposed a technical unknown that required experimentation to resolve, and only that part
Software for the business's own administrationOutside or restrictedScheme-specific tests apply; see the country rows below

Documenting software R&D

Software teams already generate the best evidence any claimant could want, as long as it is kept:

  • A design document or architecture decision record that states the unknown before the work starts.
  • Pull requests, tickets and commit messages that record what was tried and why it was abandoned.
  • Benchmarks, load tests and experiment logs, including the failures.
  • Sprint records mapping people to the R&D work stream, so that time can be allocated without guesswork.

Ask the team to add one sentence of "why" to tickets touching the uncertain work. That single habit produces more defensible evidence than any year-end description. It also makes the boundary visible: tickets with a "why" belong to the claim, and tickets without one usually do not.

A test you can apply today

For each candidate work stream, ask a senior engineer: "Before we started, did we know how to do this, or could we have found out from public sources?" If the honest answer is yes, it is development, not R&D. If the answer is "we had to try things to find out", you have the beginning of a claim, and the evidence to support it is probably already in your repository. Ask the same engineer when the trying stopped; that date is where the claim ends.

Illustrative example

Illustrative example. Tessellate Logistics, a hypothetical software company, spends 1,000,000 in a year building a route planning product. The senior engineer's answers sort the work into four streams. The web application, the customer portal and the billing integration (600,000) were built on established frameworks using documented features: outside. The migration of customer data from a legacy system (100,000) used known tools: outside. The routing engine (250,000) had to meet a response time under a data volume that the published algorithms the team tried could not achieve; the team tested three approaches over five months, two of which failed, before a hybrid method worked. The load testing harness built to measure those attempts (50,000) existed only to run the experiments. The claim covers the routing engine and the harness, eligible costs of 300,000, with the benefit calculated at that country's rate. The project note names the response time target and the algorithms already assessed; the pull requests record the two failures; the sprint board shows which engineers were on the engine stream and for how long. The claim is 30 percent of the spend, and it is the part that would survive a reviewer's questions.

Country differences

Each scheme applies its general definition to software and has published specific guidance. The rows below give the scheme's own wording on the two points that decide most software claims: known methods, and software for your own administration.

  • United States. Internal use software qualifies only if it satisfies the high threshold of innovation test: the software is innovative, the development involves significant economic risk, and the software is not commercially available for use by the taxpayer. Software is not developed primarily for internal use if it is developed to be commercially sold, leased, licensed or marketed, or to enable the taxpayer to interact with third parties or allow third parties to initiate functions or review data. Section G of Form 6765 requires software business components to be identified as internal use software, dual function software, or software not for internal use. See Internal use software: United States and the four-part test.
  • United Kingdom. HMRC's software guidance states that developing software using established techniques and tools, routine integration and the application of known methods do not constitute R&D even where the product is commercially new. An advance in science or technology means an advance in overall knowledge or capability in a field, not a company's own state of knowledge or capability alone. See Eligibility: United Kingdom.
  • Ireland. Revenue's guidance treats the application of established software engineering techniques, integration through documented interfaces and configuration as excluded development, and requires a scientific or technological uncertainty in the field of computer science for software work to qualify. Business application software, routine computer maintenance and software development using known methods and tools do not qualify unless the work resolves a scientific or technological uncertainty. See Science test: Ireland.
  • Canada. The CRA guidelines state that work is SR&ED only where a scientific or technological uncertainty existed that could not be removed by standard practice, and that the routine application of known software engineering techniques is not SR&ED even where the resulting product is new. The guidelines ask why the work was done and how it was done, that is, whether it was a systematic investigation or search by means of experiment or analysis, involving a hypothesis, testing and conclusions. See Eligibility: Canada.
  • Australia. The department's software guidance states that eligible core activities in software are experiments to resolve technical uncertainty that cannot be resolved from existing knowledge, and that developing software using known methods, or where the uncertainty is commercial, is not a core activity. Developing, modifying or customising computer software for the dominant purpose of use by the entity, its connected entities or affiliates for the internal administration of their business functions is excluded from core R&D activities. Software development claimed at a project level is a published risk area. See Core vs supporting activities: Australia.
  • New Zealand. Inland Revenue guidance states that for software the uncertainty must be scientific or technological rather than commercial, and that applying known techniques, frameworks and methods to build a new product does not resolve a technological uncertainty. Schedule 21 excludes developing, modifying or customising software for the purpose of internal administration of the business, such as payroll and accounting systems. See Eligibility: New Zealand.
  • Singapore. IRAS guidance states that software development qualifies as R&D only where there is novelty or technical risk in a scientific or technological sense, and that development using established methods, system integration, customisation and routine enhancements do not qualify even where the resulting product is new to the market. See Eligibility: Singapore.

Educational material, not advice. Eligibility depends on the scheme and the facts.

Educational content only, not tax advice. Rules change and eligibility depends on your circumstances, so check the official guidance or speak to a qualified advisor before you claim. See the disclaimer.