The honest starting point: some of it is random
HMRC runs a Mandatory Random Enquiry Programme for R&D, selecting claims at random to measure true non-compliance rather than to punish anyone. Its published results are why the scrutiny exists at all: an estimated 17.6% of relief, £1.3 billion, lost to error and fraud in 2021-22, concentrated in the SME scheme at 25.8% of that scheme’s expenditure. The estimate for 2022-23 fell to 9.9%, £759 million, with HMRC attributing the improvement to the reforms and its expanded compliance effort.
So a check can land on a perfect claim. The rest of this page is about not volunteering for one.
The risk factors HMRC has told us about
HMRC’s published approach names its concerns, and they map to selection patterns:
- Sector improbability. HMRC has run education campaigns aimed at sectors where R&D is unlikely to be carried out, its list: care homes, childcare providers, personal trainers, wholesalers and retailers, pubs and restaurants. A claim from an unlikely sector is not impossible, but it starts with the burden of explaining itself.
- First and step-change claims. A first claim, a claim after a gap, or a claim suddenly several times last year’s invites the obvious question: what changed?
- Payable-credit weight. Cash out of the Exchequer gets more attention than tax reduced, which is why loss-maker claims carry a higher scrutiny expectation.
- Thin or templated narratives. Project descriptions that read as boilerplate, describe products rather than uncertainties, or skip the baseline entirely are the visible surface of a weak claim.
- Incoherent numbers. Costs that do not reconcile to the accounts and payroll, whole-company apportionments, or categories that ignore the 65% and overseas rules.
- The agent effect. Agents are named on every claim, and HMRC sees each agent’s portfolio. Association with high-risk filing patterns is a selection factor you cannot see but do choose.
What selection is not
Selection is not an accusation, not evidence of dishonesty, and not a verdict. HMRC’s own compliance-check factsheets describe checks as making sure the right tax is paid and reliefs correctly claimed. Helpful engagement can shorten them, and can reduce any penalty if something is wrong. What a check actually involves, letter by letter, is on what happens in an enquiry, and the self-audit that finds problems while they are still fixable is on red flags in your own claim.
Frequently asked questions
What proportion of claims get checked?
HMRC reported compliance coverage of 17% of claims in 2023-24, around 9,700 checks, with over 500 people on R&D compliance against roughly 100 in 2020-21. Selection is a real probability now, not a tail risk.
Can a claim be picked with nothing wrong in it?
Yes. The Mandatory Random Enquiry Programme selects claims at random precisely to measure the true error rate, so a check is not itself an accusation. What you control is whether the check is short and boring.
Does claiming smaller amounts avoid attention?
Under-claiming is not a compliance strategy, and HMRC's published concern includes error in both directions across the SME scheme. The aim is a right-sized claim with evidence, not a small one with hope.
Does using an adviser trigger or prevent enquiries?
Neither, by itself. Every agent involved is named on the Additional Information Form, and HMRC sees patterns across an agent's filings. An adviser with clean patterns helps; one with poor patterns is a risk factor you inherit.