HMRC's R&D tax credit statistics: the series explained

HMRC publishes annual statistics on R&D tax relief: how many companies claim, how much they claim, and where those companies are registered. This page sets out what the most recent release shows, gives the regional numbers in full including the Wales figures almost nobody publishes, and hands you the whole series as a file you can download and use yourself.

The headline numbers

The most recent release is September 2025, covering the 2023-24 financial year.

Measure, 2023-24FigureChange on 2022-23
Claims made46,950Down 26%
Total support claimed£7.56bnDown 2%
Qualifying R&D expenditure£46.1bnDown 1%
Average support per claimAbout £161,000Up by about a third
First-time claimants3,765Sharply down, see below

Two movements run through the data, and they point in opposite directions.

Participation fell sharply. Claims fell from 63,780 to 46,950, a drop of 26%. SME scheme claims fell 31% while claims under the large-company and RDEC schemes fell only 5%, so the companies that left were overwhelmingly smaller ones: the group most exposed to the new compliance load and the reduced SME rates.

Underlying R&D spend did not. Qualifying expenditure was £46.1bn, down 1%, and total relief held broadly steady at £7.56bn, down 2%. A 26% fall in claims against a 1% fall in expenditure is the story of the release. The research spending stayed; a quarter of the claimants did not. Because it was the small claims that left and the money that stayed, average support per claim rose from about £121,000 to about £161,000.

The first-time-claimant number carries a caveat that is easy to miss. HMRC uplifts its headline claim counts to allow for returns it has not yet received, but it does not uplift the first-time-applicant table. HMRC expects the 3,765 figure to increase as more returns arrive, so treat the fall in new claimants as an outer edge rather than a settled number. Setting it against the headline claim count compares an un-uplifted series with an uplifted one.

Error and fraud fell hard. HMRC’s estimate of error and fraud in the relief fell from 17.6% in 2021-22 to 6.4% in 2023-24 on random-enquiry evidence. Its July 2026 annual report puts an illustrative 5.3% on the two years since — an estimate, not a measurement. That trajectory is the clearest evidence that the compliance campaign is doing what it was designed to do, and it is the main reason the campaign will not be relaxed.

Read together: the compliance era is squeezing out weak claims while leaving genuine R&D spend intact. For a company doing real qualifying work, the numbers are reassurance, not warning. But they also describe an environment where HMRC checked around one in six claims in 2023-24, the most recent year it has published, so the cost of a poorly prepared submission has never been higher; our guide to HMRC enquiries explains what that scrutiny looks like in practice.

Take the data

We have tidied HMRC’s published tables into one machine-readable file, free to download and use:

It holds 12,661 rows, one per financial year, breakdown and measure. Claims, cost of support and qualifying expenditure, split by scheme, region, sub-region, industry sector, region-and-sector, claim size band and R&D intensity; the national series runs back to 2000-01 and the breakdowns cover the four years HMRC publishes them for. We model and recalculate nothing: every row is a published HMRC figure exactly as HMRC published it, and every row names the HMRC worksheet and the exact cell it came from, so any number here can be checked against the government file in one step.

The source data is Crown copyright, published under the Open Government Licence v3.0. This compilation goes out under the same licence. Attribution is appreciated rather than required.

What the statistics measure, and what they leave out

Each release counts claims made under the R&D reliefs for a single financial year, the total relief claimed against them, and the qualifying R&D expenditure behind them. HMRC then splits those totals by scheme, by industry sector and by the region in which the claimant company is registered.

The figures lag. Each release covers a year that closed well before publication: the September 2025 release covered 2023-24. A release therefore describes the rules as they stood two policy cycles ago, not the rules a company is claiming under today. HMRC also rounds every figure independently to the nearest 5 claims or £5 million, which is why the parts of a table do not always add up to its total.

One number people expect to find in the statistics is not in them. HMRC’s estimates of error and fraud in the R&D reliefs are published separately, in its approach-to-R&D-reliefs papers and its annual report and accounts. Both sources are linked at the foot of this page.

Where the relief goes: every UK region

HMRC allocates each claim to a region using the postcode of the company’s registered office, and warns that this “might not correspond to where the R&D activity takes place”. A group registered in London running its laboratory in Swansea counts as London here. Read the table as a map of where claimant companies are registered, not of where UK research happens.

Region, 2023-24ClaimsShare of UK claimsSupportShare of UK supportAverage per claim
London11,33524.1%£2,320m30.7%£204,700
South East6,94014.8%£1,475m19.5%£212,500
East of England4,5709.7%£1,015m13.4%£222,100
North West4,2909.1%£490m6.5%£114,200
South West3,5107.5%£350m4.6%£99,700
West Midlands3,4457.3%£565m7.5%£164,000
Yorkshire and The Humber3,1306.7%£265m3.5%£84,700
East Midlands2,8106.0%£315m4.2%£112,100
Scotland2,7855.9%£360m4.8%£129,300
Wales1,4403.1%£115m1.5%£79,900
North East1,3352.8%£115m1.5%£86,100
Northern Ireland1,3052.8%£140m1.9%£107,300
Unknown600.1%£25m0.3%
United Kingdom46,950100%£7,555m100%£160,900

Shares and averages are ours, calculated from HMRC’s published claim counts and support totals. London, the South East and the East of England between them account for just under half of all UK claims.

Wales: a steady share of the claims, a shrinking share of the money

The registered-office caveat matters here more than anywhere. A company doing its research in Wales but registered as part of a group elsewhere appears under that group’s region rather than under Wales, and the reverse holds too. With that stated, here is what the table shows.

Welsh-registered companies made 1,440 claims in 2023-24 and were awarded £115m of support. That is 3.1% of UK claims and 1.5% of UK support. Wales has held roughly 3% of UK claims in every one of the four years HMRC publishes regionally, so its share of claimants has been stable. Its share of the money has not: it stayed close to 2.0% in each of the three years to 2022-23, then fell to 1.5% in 2023-24.

Wales2020-212021-222022-232023-24
Claims2,6652,4401,9151,440
Support£140m£155m£150m£115m
Qualifying expenditure£735m£800m£800m£670m
Average support per claim£52,500£63,500£78,300£79,900
UK average support per claim£78,600£91,300£120,600£160,900

The average Welsh claim and the average UK claim have pulled apart. In 2020-21 the average claim in Wales was worth about two-thirds of the UK average. In 2023-24 it was worth about half: roughly £80,000 against roughly £161,000. Both averages rose over the period, but the UK figure rose far faster. The regions pulling it up are visible in the table above: the East of England, the South East and London are the only three where the average claim exceeds £200,000, and between them they take almost two-thirds of all UK support.

Two further cuts that HMRC publishes and almost nobody quotes. Within Wales, South East Wales accounted for 770 of the 1,440 claims and £70m of the £115m; Mid and South West Wales for 380 claims and £20m; North Wales for 290 claims and £25m. And by industry, Wales is more of a manufacturing claimant base than the UK as a whole: manufacturing accounted for 500 of the 1,440 Welsh claims, or 35%, against 26% across the UK. Information and communication, the largest claiming sector nationally at 26% of claims, accounted for 17% in Wales.

If you are a Welsh company weighing up a claim, our Cardiff and Wales R&D tax page covers the practicalities.

Scheme, sector and claim size

2023-24 is the first year in which RDEC exceeded the SME scheme by value: £4.41bn against £3.15bn, a 36% rise against a 29% fall. Rate changes explain much of it, alongside SMEs claiming under RDEC because of grant funding or subcontracting positions under the old rules. It is also the first year with a separate figure for R&D-intensive SMEs: 3,990 claims worth £830m.

Three sectors — information and communication, manufacturing, and professional, scientific and technical — accounted for 71% of claims and 71% of support. For research-heavy fields like life sciences, the concentration is unsurprising: that is where the qualifying work is easiest to identify and evidence.

Support remains heavily concentrated in a handful of very large claims. The 500 claims worth more than £2m each are 1% of all claims and 44% of all support. At the other end, claims worth under £15,000 fell from 19,675 to 12,970, a drop of 34% against the 26% fall overall. They have not disappeared: they are still 28% of all claims. But they fell faster than anything else, which is where the rise in the average claim comes from.

What to watch when the Autumn 2026 release lands

The 2025 release was the last one built around the SME-versus-large split. From April 2024 the merged scheme covers all companies, and future releases will instead compare R&D-intensive SMEs claiming ERIS against everyone else. The interesting dividing line stops being company size and becomes R&D intensity.

Three things to look for:

  • The first merged scheme and ERIS figures. How the claimant population divides on intensity rather than size, now that the old split has gone.
  • Whether smaller genuine claimants return. The sharp fall in sub-£15,000 claims mixed companies that should never have claimed with companies that could have. Only a recovery in volumes without a rise in error and fraud would tell the two apart.
  • Whether average claim value holds. The rise of about a third came from the bottom of the distribution thinning out, not from bigger claims. Whether that level persists shows if the shift is structural or a one-year effect.

This series

HMRC publishes these statistics once a year, in the autumn, and has scheduled its next release for Autumn 2026. Every release so far has covered the financial year that ended about eighteen months before publication, which points to 2024-25 — the first year of the merged scheme — though HMRC has not confirmed the coverage.

Everything on this page, and both download files, is built from the September 2025 release. When the Autumn 2026 figures land we rebuild the CSV and JSON from HMRC’s new tables and update this page on publication day. We intend to keep the download URLs stable across releases, so a bookmark or a scheduled fetch should keep working.

Last checked against the source tables on 1 September 2026. If you want to understand where your own company sits in the current system, start with which R&D scheme applies to you, check the rates for your period, or talk it through with a chartered adviser.

Sources

  • R&D tax credits statistics, September 2025 — the release, its commentary and the spreadsheets below.
  • Main tables (ODS), tables RD1 to RD8 — claims, cost of support and qualifying expenditure by year and scheme (RD1, RD2, RD4), R&D-intensive SMEs (RD3), the regional analysis (RD5), the sector analysis (RD6), the claim size bands (RD7) and first-time applicants (RD8).
  • Supplementary tables (ODS), tables RDS1 and RDS2 — the ITL2 sub-regional analysis behind the Welsh figures, and the region-by-sector analysis behind the Welsh sector mix.
  • HMRC’s approach to R&D tax reliefs 2023–24 — 17% of claims checked in 2023-24 and 500-plus compliance staff. Its 17.6% error-and-fraud estimate for 2021-22 stands; its 7.8% illustrative estimate for 2023-24 was superseded by the measured 6.4% below.
  • HMRC annual report and accounts 2025–26 — error and fraud measured at 6.4% for 2023-24, with an illustrative 5.3% for the two years since.