HMRC’s Guidelines for Compliance on R&D expect a claim built in a particular order: a competent professional identifies the advance and the uncertainties first, the costs are found afterwards, and both are recorded well enough to be evidenced later. GfC3 — “Help to see if your work qualifies as research and development for tax purposes” — is HMRC’s own account of how it reads the definition of R&D. It changes nothing in law, and says so: the guidelines “do not change our view of the law”. Read GfC3 as HMRC’s marking scheme. The DSIT Guidelines remain the syllabus.
GfC3 runs to six parts. Part 2 says HMRC expects you to take fourteen steps when you claim; part 5, which carries most of the record-keeping content, is a recommended approach rather than a required one. Neither part is law. They are HMRC’s own statement of what it looks for. The duty underneath them is not optional: a company must keep the records needed to deliver a correct and complete return, whatever approach it takes to them.
What do the guidelines expect you to do before you claim?
Fourteen steps, and their order carries the meaning.
The first ten belong to the competent professional. They identify the projects that sought an advance, confirm the field each sits in, name the exact uncertainties, and explain why the answer to each was not readily deducible by a competent professional in that field. They set out what the field already knew when the project started, and why resolving the uncertainties would advance it. Two of the ten are dates: when each uncertainty was identified, and when it was resolved, abandoned, or is expected to be resolved. The rest is the shape of the work — why the approach to each uncertainty amounted to a project or sub-project, the plan for resolving it, and the steps actually taken.
The last four fall to the company and its adviser, still supported by the competent professional where the judgement is technical: identify the activities that directly contributed to resolving each uncertainty, identify any qualifying indirect activities from the paragraph 31 list, find those activities’ costs in the records, and check whether accounting practice, the DSIT Guidelines or tax law restricts them.
Costs come thirteenth; the fourteenth step only asks whether any of them are restricted. A claim assembled the other way round — the ledger first, a narrative written to fit the total — has no answer to the question the sequence forces: what did you not know, and when did you stop not knowing it? HMRC pairs the steps with responsibility: the facts of a claim remain the company’s even where an adviser prepared it, and a careless overclaim carries a penalty. The same fourteen steps produce what the Additional Information Form asks for.
Where do the guidelines go beyond the DSIT Guidelines?
Who the competent professional is. The DSIT Guidelines measure uncertainty, overall knowledge and appreciable improvement through this person’s eyes, and never say who qualifies. GfC3 does. It expects three attributes together: knowledge of the relevant principles, awareness of the state of knowledge in the field as a whole, and accumulated experience with a successful track record. It warns that having worked in a field, or an intelligent interest in it, is not enough. Who counts as a competent professional works through the test and the evidence that supports it.
What that person’s opinion has to say. GfC3 asks it to cover the depth of their knowledge and experience, the state of knowledge in the field, what the advance is, why it is an advance, and whether it is one of knowledge, capability or both. A bare assertion that the project qualifies will not do. It also asks you to keep the opinion in writing, with the person’s qualifications attached — evidencing the claim gets harder if they are unavailable.
Where a project starts, and what “readily deducible” means. There can be no qualifying project before a plan or method to resolve identified uncertainties existed. A discovery made outside a project is not claimable, though the work to develop it afterwards can be. GfC3 also tightens the phrase that claims lean on hardest: readily deducible does not mean straightforward or effortless, but able to be worked out from existing knowledge without significant effort.
Where the edges fall. GfC3 draws lines that the DSIT Guidelines do not. Staff hired to maintain equipment used on qualifying work are doing a listed indirect activity; the HR team’s work in hiring them is too remote from the advance to be part of the project, even though the paragraph 31 list names taking on staff. Extra security guarding trial equipment qualifies; the general patrol covering the trial area does not. And certifying a product whose functionality is already proven is not R&D — unless the certification itself demands a further advance in science or technology to materially improve functionality. That boundary does most of its work in regulated sectors; medtech claims show where it falls in practice.
Where do they only restate the DSIT Guidelines?
Most of part 4. The advance is measured against the field rather than your own company; appreciable improvement, uncertainty and system uncertainty carry their DSIT meanings; work that independently repeats an undisclosed trade secret still counts; failure does not disqualify a project. All of it is the DSIT Guidelines in shorter sentences. Our guide to what counts as qualifying R&D applies the same tests. A claim that already gets those right is confirmed by GfC3 rather than caught by it.
The twenty-seven worked examples across parts 3 and 4 show HMRC applying its own tests to facts. HMRC says they are not a template for your own project — write from your own facts rather than the example’s.
What should you do differently because of it?
Name the competent professional before drafting starts, and record their credentials against the three attributes. Get their opinion in writing while they are still there. Date both boundaries: when the uncertainty was identified, and when it was resolved or abandoned. GfC3 returns repeatedly to the closing boundary: testing after the answer was known, fine-tuning, certification and scale-up.
Then write down the method. Where costs are apportioned or estimated, GfC3 sets the standard the estimate must meet and the workings it expects you to keep — our page on what records you need sets those out in full.
None of this stops HMRC opening a check. What it changes is how the check goes: the evidence is already assembled, and the account is the one the work actually produced. See HMRC R&D enquiries. For a straight answer on whether your evidence would survive that reading, talk it through with a chartered adviser.
Sources
- GfC3: purpose, scope and background (part 1) — the guidelines expand on HMRC’s existing guidance on the DSIT guidelines without changing HMRC’s view of the law; getting the claim right is the company’s responsibility even where a tax adviser is used, and carelessness leading to an inaccurately high claim carries a penalty. First published 31 October 2023, last updated 23 January 2025.
- GfC3: expectations of claimants (part 2) — the fourteen steps: ten for the competent professional, then four for the company or its agent, supported by a competent professional as needed, which find the qualifying costs in the records and then test them for restriction against generally accepted accounting practice, the DSIT guidelines and tax law; the facts of a claim always remain the company’s responsibility.
- GfC3: importance of a competent professional (part 3) — the three attributes HMRC expects, the examples of evidence of competence (any one of which may be good evidence), the warning that having worked in a field or having an intelligent interest does not suffice, and the five things the professional’s opinion should set out.
- GfC3: how to identify qualifying R&D activities (part 4) — there can be no qualifying project before a plan or method to resolve identified uncertainties existed; “readily deducible” means able to be worked out from existing knowledge without significant effort; the paragraph 31 applications on hiring, site security and IT updates, including that HR costs of hiring staff who undertake qualifying indirect activities are too remote; and that testing after the uncertainties are resolved, including regulatory certification of already-proven functionality, is not R&D unless certification requires a further advance.
- GfC3: recommended approach to claims and record keeping (part 5) — you do not have to follow the recommended approach; keep a written copy of the competent professional’s opinion with their qualifications and experience, since it may be hard to evidence the claim if that person is unavailable; estimates reached using evidence and reason; the recommendation to record the claim methodology, sampling and apportionment basis; large business customers agree sampling with their customer compliance manager first.
- FA 1998 Sch 18 para 21 — the statutory duty underneath the guidance: a company which may be required to deliver a company tax return must keep such records as may be needed to enable it to deliver a correct and complete return for the period, and preserve them.
- Guidelines on the meaning of R&D for tax purposes (DSIT) — the definition GfC3 expands on: the project at paragraph 19, overall knowledge or capability at 20, appreciable improvement at 23 to 25, system uncertainty at 29 to 30, the qualifying indirect activity list at 31 (including “taking on and paying staff” at 31(c)), the start and end of R&D at 33 to 34, technological versus commercial planning at 36 to 37, and abortive projects at 38. The term “competent professional” runs throughout without being defined.
This page describes the rules as they stood at the review date above, as general information rather than advice on your circumstances. For how that distinction works, see our terms; for an answer on your own facts, talk to us.