Software R&D tax relief is open to UK companies whose development work tries to advance software technology itself, not just their product, and has to get past an uncertainty that a competent professional in the field couldn't readily work out. Plenty of hard, expensive engineering doesn't pass. When work does pass, R&D tax credits for software are claimed like any other R&D: through the merged scheme's 20% expenditure credit, or through ERIS if you're a loss-making, R&D-intensive SME.
It's tempting to assume nearly all your engineering qualifies. HMRC's own software guidance warns against assuming a commercial project "fully aligns with the definition of R&D", and only the parts that meet the test count. We'd rather tell you which parts those are.
The legal test: advance in science or technology plus technological uncertainty
R&D for tax purposes is defined in section 1006 of the Income Tax Act 2007, and the detail sits in the DSIT Guidelines on the meaning of R&D, which regulations give legal force. The 2023 version covers accounting periods beginning after 31 March 2023, and it treats software exactly like any other field. Three things have to be there:
- An advance in overall knowledge or capability. The work has to seek an advance in a field of science or technology, not just in your company's own knowledge (Guidelines paragraph 6). A product doesn't become an advance because it's built with software (paragraph 8). An appreciable improvement counts too, if a competent professional would see it as genuine and non-trivial (paragraphs 9 and 23).
- Scientific or technological uncertainty. Whether the thing was technologically feasible, or how to do it in practice, wasn't readily available or deducible by a competent professional working in the field (paragraph 13). If an experienced engineer would say "we know how to do this, it'll just take time", there's no uncertainty.
- Work that resolves it. Only activities that directly help resolve that uncertainty are R&D, plus some qualifying indirect activities such as the R&D's own admin and training (paragraphs 4, 5 and 31). Improvements, optimisations and fine-tuning that don't materially affect the underlying technology don't count (paragraph 14).
"Overall knowledge" means what's publicly available, or readily deducible from it (paragraph 20). So your team learning a framework that already exists isn't R&D, however new it is to you (paragraph 22). Two points help software teams. Since April 2023, mathematical advances in themselves count as science (paragraph 15B), which matters for algorithm work. And the competent professional who explains the advance can be one of your own engineers, as long as they set it against what was public at the time, not just assert it (CIRD81300). Routine software development doesn't qualify under HMRC's guidelines, however much effort it took.
What passes and what doesn't in software work
HMRC's software page gives its own examples of where the line falls, and HMRC's R&D guidance is plain that routine work is out. Here's how typical software work sorts, with the Guidelines paragraph behind each call where there is one. "Can pass" still needs the facts to back it up.
| Can pass (if the facts back it) | Usually doesn't |
|---|---|
| A novel algorithm that significantly increases overall capability (paragraph 23) | Feature development with established frameworks and patterns, however large (paragraph 22) |
| A data architecture that goes beyond what existing database engines can do | A standard CRUD application build, however complex |
| Extending a framework or SDK beyond its original design, where how to do it wasn't known | Configuring or customising existing software (React, Rails, Django, Laravel etc.) to your own requirements |
| Combining standard technologies where a competent professional couldn't readily deduce how they'd fit together (paragraph 30) | Assembling components to an established pattern, or an API integration from published documentation (paragraph 29) |
| Work on a problem that's a documented subject of computer science research | Applying an existing ML model, including fine-tuning a pre-trained one, to a new dataset with routine changes (paragraph 24) |
| Software built solely as a tool for an R&D project in another field (paragraph 27(a)) | Cosmetic and UI work, including layouts, styling and animations (paragraph 42) |
| A new problem after launch that a competent professional couldn't readily fix (paragraph 35) | Routine bug fixing and standard QA testing (paragraph 35) |
| A performance target no known method can meet, where feasibility itself is uncertain | Optimisation that applies standard techniques more carefully (paragraph 14) |
One thing to watch: what passed a few years ago may not pass now. HMRC's software page says projects that met the test before "may not qualify now", because solutions have since become available or readily deducible. Judge each year's work against what was public that year.
Where the boundary falls inside a single project
HMRC's software guidance says most commercial projects "will not qualify as R&D in their entirety". The R&D project is only the set of activities that together resolve the uncertainty, and it can sit inside a bigger commercial project without turning the rest into R&D (paragraph 19). It starts when work to resolve the uncertainty starts, and ends when it's resolved or the work stops (paragraphs 33 and 34).
In practice, business requirements gathering doesn't count, but technical analysis aimed at the uncertainty can. Planning the R&D parts counts; financial, marketing and legal planning doesn't (paragraphs 36 and 37). Design and testing count only where they help resolve the uncertainty.
So the question isn't "does this project qualify?" It's "which workstreams involved genuine uncertainty, and what did they cost?" The Additional Information Form makes you describe that work precisely, so loose boundaries show up there first.
Worked example: a scaling team's engineering roadmap
A fictional B2B SaaS company has a six-workstream roadmap for the year and ten engineers, each costing £90,000 all in (salary, employer NIC and pension). That's £900,000 of engineering staff cost.
| Workstream | Description | Verdict | Reason | Qualifying staff time |
|---|---|---|---|---|
| Query engine redesign | Building a new execution layer to achieve sub-10ms latency on a class of analytical queries that existing engines cannot handle at that specification | Potentially qualifying | Not known whether the latency target is achievable; goes beyond what existing database engines can do | 3 engineers at 70% |
| Billing and subscription module | Implementing recurring billing using Stripe, with standard subscription logic | Does not qualify | Routine development against a well-documented third-party API; no technological uncertainty | None |
| ML anomaly detection | Applying a pre-trained anomaly-detection model to a new category of operational data | Does not qualify | An existing model on a new dataset; the advance is in the data, not in science or technology | None |
| Novel compression algorithm | Investigating whether a domain-specific compression technique can reduce storage by 60% below what DEFLATE achieves, where no published algorithm achieves this on the data type | Potentially qualifying | An algorithmic advance whose feasibility a competent professional couldn't readily deduce | 1 engineer at 50% |
| Multi-tenant UI redesign | Rebuilding the dashboard and settings pages across the product | Does not qualify | Cosmetic and UI work; no uncertainty beyond standard engineering execution | None |
| Real-time protocol bridge | Bridging two systems with conflicting real-time message-ordering guarantees where the interaction effects are genuinely unknown and no published approach addresses this combination | Potentially qualifying (verify specifics) | System uncertainty: a competent professional couldn't readily deduce how to combine them; check the facts before claiming | 2 engineers at 40% |
Three of the six potentially qualify, and their staff time comes to £306,000: £189,000 on the query engine (3 × 70% × £90,000), £45,000 on compression (1 × 50% × £90,000) and £72,000 on the protocol bridge (2 × 40% × £90,000). That's 34% of the engineering payroll, not all of it. Under the merged scheme it's a £61,200 credit (20%), worth £45,900 after Corporation Tax if the company pays the 25% main rate. Our R&D relief estimator gives the same figures. Add any qualifying cloud, data licence or software costs for those workstreams, and keep the timesheets and technical notes that support the split.
Client work: who claims when R&D is contracted out
If you build software for clients, this matters as much as the test. For accounting periods beginning on or after 1 April 2024, under both the merged scheme and ERIS, the general rule is that only the party that decides to undertake or initiate the R&D can claim (CIRD161000).
So if a client contracts you to build something and, going by the contract and the circumstances, intended or contemplated R&D of that sort, the client claims and you generally can't. That takes more than knowing some R&D might happen: the client has to have had a specific appreciation of what R&D would be done. You can still claim for client work if the client's an "irrelievable client": a university, charity, scientific research organisation or health service body (CIRD163000), or someone not acting in the course of a trade within the charge to tax.
Two things work in your favour. Scoping work you do at your own risk, before any contract exists, can be yours to claim, even if the client later pays you for it. And where it's unclear who decided, HMRC treats an evidenced agreement between you and the client about who claims as persuasive. Transitional rules apply to work that straddles the old schemes. Building your own product is simpler: you decided, so you claim.
What goes in the AIF project description
For each project you describe, the AIF asks for the field of science or technology, the baseline, the advance you were after, the uncertainties and how you tried to resolve them. For software, name the sub-field precisely: HMRC's software page says putting the advance in "a certain area or field of software is not specific enough". Then cover:
- Baseline: what existing tools, frameworks or published techniques could do when you started, ideally measured against a comparable family of software.
- Advance: what you were trying to make possible, against that baseline, rather than the commercial feature.
- Uncertainty: why it was uncertain for the industry, not just your team. HMRC's AIF guidance says a problem isn't an uncertainty if it can be resolved in a discussion with peers.
- Method: what you tried, what failed, and whether it was resolved.
With 1 to 3 projects you describe them all; with 4 or more, you pick 3 or more that between them cover at least half the qualifying spend. The AIF guide takes each section in turn, with a worked software narrative.
A quick eligibility checklist
Before a piece of work goes into your claim, run it through these questions. You need a yes to all four for it to be potentially qualifying.
- Was there an advance sought in science or technology, not just in your company's own knowledge or in business capability?
- Was there genuine technological uncertainty about whether or how the objective could be achieved, where the outcome was not known in advance?
- Could a competent professional in the relevant sub-field NOT readily deduce the solution from existing scientific or technological knowledge?
- Was the work not routine, i.e. not the application of standard techniques in the normal way, however carefully or expensively?
If it's yes to all four, the next step is to put accurate figures on the costs. If you're unsure on any of them, settle it before the claim is filed, not after the Additional Information Form is submitted.
Why honesty in your claim protects your company
Weak or inflated R&D claims don't sit quietly with HMRC. The Additional Information Form has to reach HMRC before or on the same day as the CT600, and without it the claim isn't accepted. HMRC uses the form to risk-assess claims, and one that describes routine work can draw a compliance check, which costs time and management attention whatever the outcome. If a past claim was too high and it's too late to amend the return, there's an HMRC disclosure service for that.
For a funded, scaling company the downside is lopsided: repaying an overclaimed credit, with interest and possibly penalties, outweighs anything the extra work adds. A defensible claim over a smaller qualifying base is worth more than an inflated one that sits in HMRC's enquiry pipeline.
Once you know your qualifying workstreams, pick the route. The merged scheme gives a 20% expenditure credit, above the line and taxable. A loss-making SME whose relevant R&D expenditure is at least 30% of its total relevant expenditure (connected companies included) can choose ERIS instead: an extra 86% deduction and a payable credit worth up to 14.5% of the surrenderable loss. Both are set out in HMRC's merged scheme and ERIS guidance, and the R&D relief estimator gives an indicative figure for your scenario.
If this is your first claim, or your last one was made more than 3 years before your notification window closes, check the notification deadline: the window ends 6 months after your period of account does, and missing it makes the claim invalid however good the work. That step's due to go for claims made from 1 April 2027 (announced at Budget 2026, still to pass in the Finance Bill).
The merged R&D scheme guide has the full mechanics of both routes. To go through your own workstreams and costs with us, start with the R&D claims service. There's more on R&D for software development companies and SaaS companies too.
Calculator
R&D Relief Estimator 2026/27
Enter your qualifying R&D expenditure and company position. The tool shows your merged-scheme net benefit and, where you qualify, the ERIS payable credit. Speak to a specialist before filing.
Total qualifying R&D costs for the period: staff, subcontractors or externally provided workers (EPW), software, consumables.
The R&D-qualifying fraction of payroll only. Must be no more than total qualifying expenditure above.
Loss-making companies with 30%+ R&D intensity may qualify for the higher ERIS payable credit.
Used for the 30% R&D-intensity test (ERIS). Must be greater than zero.
25% main rate, 19% small profits rate, and 26.5% on each extra £1 of profit inside the marginal relief band. Only affects the net value of the merged-scheme taxable credit for a profit-making company; a loss-making company's credit is netted at 19%.
Your result
Your answer's ready. Press below to see it.
Want this checked against your specific situation?
Leave your details and a one-line summary. One of our specialists will come back to you, with no obligation.