What the data shows and what it does not

1Summary

Discussion of research and development tax incentives often turns on how many claims are wrong and why. This paper sets out what the tax authorities of the five countries covered by this site actually publish on that question: the United States, the United Kingdom, Canada, Ireland and Singapore.

The answer differs sharply by country. HMRC in the United Kingdom publishes estimates of error and fraud based on random enquiries, together with compliance check volumes, adjustment rates and a breakdown of claims made with and without specialist agents.12 The Canada Revenue Agency publishes yearly figures on claims accepted as filed, accepted after modifications and denied.3 Ireland publishes the cost and number of claims but, in the sources reviewed, no figures on review outcomes.45 Singapore has published one set of outcome figures, for 2018.6 The IRS publishes no current statistics on research credit error, examination or adjustment in the sources reviewed.7

The figures that do exist measure different things and cannot be compared directly. One gap is common to every country: no authority publishes data linking claim outcomes to how the claim preparer was paid. This paper reports the published record only. It does not estimate any figure that has not been published, and it makes no recommendation for any country.

2Method

This paper uses official publications only: statistics, annual reports and guidance from the tax authorities, and reports of their auditors and statutory evaluators. It looks for four kinds of information:

Error or non-compliance rates: an estimate of how much of the relief claimed is incorrect.
Review coverage: how many claims are examined.
Review outcomes: how many examined claims are adjusted or denied, and by how much.
Adviser and fee data: whether the authority records who prepared a claim and how they were paid, and whether it publishes anything from that record.

Where a source was not found, the paper says so. A statement that data was not found means that it was not found in the official sources reviewed for this paper, not that it cannot exist.

3United States

3.1What is published

The IRS Statistics of Income program publishes tables on corporations claiming the research credit, by industrial sector and size. The tables listed on its website run only to tax year 2014, with a few series ending in 2013.7 They report claimants and amounts, not examinations.

In October 2021 the IRS said that it receives thousands of research credit claims each year for credits in the hundreds of millions of dollars, and that these claims "are currently examined in a substantial number of cases and consume significant resources for both the IRS and taxpayers".8 It did not give the number examined or the results.

3.2Older and narrower figures

The most detailed published analysis of examination outcomes is from the Government Accountability Office in 2009. GAO studied a panel of large corporations that included most of the largest credit claimants in 2003 and 2004 and accounted for about half of the credits claimed in those years.9 Using data from IRS examiners, it found that amended claims had increased the credits claimed by between 1.5 and 5.4 percent a year for 2000 to 2003, and that examiners had recommended changes that would reduce the credits claimed by between 16.5 and 27.1 percent a year.9 GAO noted that some of those recommendations were still open and could be appealed, and that the panel was not representative of all claimants.9

In April 2026 the Treasury Inspector General for Tax Administration reported on the payroll tax version of the credit for small businesses. It found that about 11,400 employment tax returns claimed $382 million of the payroll credit for tax year 2023, and about 12,600 claimed $381 million for 2024.10 Analyzing returns filed electronically as of December 2024, it found 40 businesses that elected the payroll credit for 2023 without meeting the gross receipts requirement, and that the IRS had allowed them to claim $3.1 million.10 That review concerned eligibility for one election, not the accuracy of research claims generally.

3.3What is not published

No IRS publication found for this paper reports an error rate for research credit claims, the share of claims examined, the share adjusted, or the amounts by which they are adjusted.

The IRS does review claims in a way that could produce such figures. It states that it reviews each research credit refund claim to determine whether it is valid, that it writes to taxpayers whose claims lack required information and, during a transition period, gives them 45 days to perfect them, and that it tries to decide such claims within six months.11 No published count of claims reviewed, perfected or rejected under that process was found.

The IRS also collects other information that could inform such figures. Its 2008 examiner guide, which states that it is not an official pronouncement of IRS position, told examiners to request the engagement letter for a research credit refund claim prepared by an outside consultant.12 From tax years beginning after 2025, Section G of Form 6765 will generally require business component information with the claim.13 Neither source produces published statistics. Form 6765 does not ask who prepared the claim or how the preparer was paid.13

4United Kingdom

4.1Error and fraud

HMRC introduced a Mandatory Random Enquiry Programme because, before it, HMRC's understanding of non-compliance in research and development tax credits "was incomplete".1 The programme examines a sample of claims from small and medium-sized enterprises (SMEs) in detail. Its second round, covering 2021 to 2022, found that around half of SME claims were non-compliant in part or in full, and that 30 percent were wholly non-compliant.1 HMRC estimated error and fraud across the reliefs for that year at £1.3 billion, or 17.6 percent of £7.6 billion claimed.1

HMRC's 2025 to 2026 annual report gives its final estimate for 2023 to 2024: 6.4 percent of the estimated cost of the reliefs, or £493 million, overall, with 11.1 percent for the SME scheme and 3.2 percent for the R&D expenditure credit.2 HMRC's original, illustrative estimate for that year was 7.8 percent. The 17.6 percent figure above is its estimate for 2021 to 2022.2 HMRC said the rate was lower than its estimates since 2020 to 2021, reflecting changes including the mandation of digital claims and additional information requirements introduced in 2023.2 Its estimates for 2024 to 2025 and 2025 to 2026, of 5.3 percent overall, are illustrative.2 The Comptroller and Auditor General qualified his audit opinion on HMRC's 2025 to 2026 accounts because of the level of error and fraud, said that the SME rate of 11.1 percent for 2023 to 2024 "remains amongst the highest reported across all government spending programmes", and said HMRC "will not know the full impact of these changes until it has completed a sample of enquiries on 2025-26 returns".2

4.2Compliance checks

HMRC has also published figures on its compliance checks, excluding cases handled by its large business teams.1

Measure	2022 to 2023	2023 to 2024
Claims received	90,000	61,000
Compliance checks	8,300	9,700
Coverage	10%	17%
Tax recovered from checks	£288 million	£441 million
Checks requiring an adjustment	71%	77%

Figure 1. HMRC compliance checks on R&D claims, rounded by HMRC.1

The 77 percent figure describes checks on claims HMRC selected, mostly because they appeared risky. It is not a rate of error across all claims. The random enquiry estimates serve that purpose.

4.3Disputes

HMRC reported that 89 percent of the compliance checks it settled in 2023 to 2024 were resolved by agreement with the claimant.1 In the same year it held 822 requests for a statutory review of an R&D decision; of the 251 concluded, HMRC's decision was upheld in 214 cases and varied in 33, and it was cancelled in fewer than five.1 These figures describe disputes after a check, not the accuracy of claims generally.

4.4Agents

HMRC states that more than 90 percent of R&D claimants have an authorized agent.1 In its second random enquiry programme, around half of companies used a specialist R&D agent, and HMRC said that non-compliance by value "was slightly higher in those claims that were submitted with the support of a specialist R&D agent": 28 percent of the value of claims with such an agent, against 21 percent without.1 HMRC also reported earlier analysis showing that claims made with agents affiliated to a professional body had a similar percentage of non-compliant cases to other groups but lower non-compliance by value.1

HMRC's tables distinguish claims made with and without a specialist agent. They do not distinguish how the agent was paid.1

4.5Volumes

HMRC's official statistics estimate 40,325 claims for 2024 to 2025, 17 percent fewer than the year before, and total relief of £8.2 billion, 5 percent more.14 HMRC states that "A key driver of the decrease is believed to be the continuing impact of the additional information form".14

5Canada

The Canada Revenue Agency publishes yearly statistics on the Scientific Research and Experimental Development (SR&ED) program.3

Fiscal year	Claims processed	Accepted as filed	Accepted after modifications	Denied
2021 to 2022	20,317	91%	6%	3%
2022 to 2023	20,079	91%	6%	4%
2023 to 2024	21,537	91%	6%	3%
2024 to 2025	22,738	90%	6%	4%
2025 to 2026	23,677	90%	6%	4%

Figure 2. SR&ED claims processed and their outcomes, as published by the CRA.3 The CRA notes that the 2022 to 2023 rates do not add to 100 percent because of rounding.

The CRA also reports investment tax credits claimed and allowed: for 2025 to 2026, $4.9 billion claimed and $4.6 billion allowed.3 It notes that claims processed in a year may have been filed in an earlier year, and that the figures come from its case management system and may not match its assessing systems.3

These figures are outcomes of the CRA's processing and review of claims. They are not an estimate of error across all claims, and the page reviewed does not define "accepted after modifications" or say how many claims were reviewed in depth.3

Canada collects more information about claim preparers than any other country reviewed. Part 9 of Form T661 asks whether a claim preparer was engaged, and for each preparer its name, business number, billing arrangement, billing rate and the total fee paid, payable or expected to be paid. Billing arrangement code 1 is a contingency fee arrangement, described as one "where the fee is based on a percentage of the investment tax credit earned".15 No CRA publication was found that reports statistics drawn from these answers.

6Ireland

Revenue publishes statistics on the cost of the research and development corporation tax credit and the number of companies claiming it. Its analysis of 2024 corporation tax returns reports that 2,165 companies claimed the credit, at a cost of €1,263 million, up from 1,804 companies and €976 million in 2023.16 Its separate statistics publication, dated March 2026, reports allowable expenditure, credit claimed and cost by year.4

Revenue's 2025 annual report reports 237,550 audit and compliance interventions across all taxes, yielding €733.5 million.5 It does not break those figures down for the research and development credit. No Revenue publication found for this paper reports how many credit claims were reviewed, how many were adjusted, or a rate of error.

7Singapore

IRAS states that in 2018 about 520 companies claimed enhanced R&D benefits, that SMEs made up 85 percent of R&D claims, and that, based on the claims processed, 83 percent of SME claims were granted full R&D benefits, with another 6 percent adjusted by IRAS on the approved expenditure.6 No later outcome figures were found in the IRAS sources reviewed.

8Comparing the record

Country	Error or non-compliance estimate	Review coverage and outcomes	Adviser involvement recorded	Fee basis recorded
United States	None found	None current; GAO panel data from 2009	Not on Form 6765	No
United Kingdom	Yes, from random enquiries	Yes, compliance checks and adjustments	Yes, agents named on the claim	No
Canada	None found	Yes, accepted, modified and denied	Yes, Form T661 Part 9	Yes, collected but not published
Ireland	None found	None found	Not found in sources reviewed	Not found
Singapore	None found	2018 figures only	Not found in sources reviewed	Not found

Figure 3. What each authority publishes or records, as found in the sources reviewed for this paper.713131546 "None found" means that no figure was found in those sources.

8.1Why the figures cannot be compared

The published figures answer different questions.

Random and selected samples. HMRC's error estimates come from random enquiries, designed to represent all SME claims.1 Its compliance check results come from claims selected for review, often because they appeared risky.1 A rate drawn from selected claims will usually differ from a rate across all claims.
Processing outcomes. The CRA's acceptance rates describe what happened when claims were processed, including claims accepted without detailed review.3 They are not an estimate of error.
Different schemes. Each country defines eligible activity and expenditure differently, and HMRC notes that its figures are not directly comparable across its own scheme changes.2
Different years. The figures range from 2018 in Singapore to 2025 to 2026 in Canada.36

A high adjustment rate in one country and a low one in another may reflect different selection, different rules or different administration as much as different claim quality.

9The gap common to every country

No authority reviewed publishes data linking claim outcomes to how the claim preparer was paid.

HMRC's tables come closest. They compare claims with and without a specialist agent, and HMRC has described the risk it sees from agents who offer to submit speculative claims in return for high commission.1 They do not distinguish agents by fee arrangement.1 Canada collects the billing arrangement and fee for every preparer on every claim, but no statistics drawn from those answers were found.15 Form 6765 in the United States does not ask, and no fee field was found in the Irish or Singapore sources reviewed.13

The question of whether claims prepared under contingent fees fare differently on review therefore cannot be answered from published official data in any of the five countries.

10What the facts show

The facts show an uneven record. One authority, HMRC, measures error across all claims and publishes the result. One, the CRA, publishes regular processing outcomes. Ireland and Singapore publish little or nothing on outcomes, and the IRS publishes no current figures on research credit examination, adjustment or error.

The record also shows where measurement has changed what is known. HMRC described its understanding of non-compliance as incomplete before it began random enquiries, and its published estimates since then have informed later changes to its claim process.12 The record does not show the effect of any rule on how advisers are paid, because no authority reviewed publishes data that would show it.

References

HM Revenue and Customs, Approach to Research and Development tax reliefs 2023 to 2024 (published 30 October 2024), sections 1.1, 1.2 (Table 1), 2.2, 2.4 (Table 4), 2.5 and 3, and Annex B, Themes 1 and 5 and Tables B.4 and B.5. https://www.gov.uk/government/publications/hmrcs-approach-to-research-and-development-tax-reliefs-2023-to-2024/approach-to-research-and-development-tax-reliefs-2023-to-2024 (accessed 11 October 2026). Back
HM Revenue and Customs, Annual Report and Accounts 2025 to 2026 (July 2026), "Corporation Tax research and development tax relief error and fraud" (page 107), and the Report of the Comptroller and Auditor General published within it, Summary paragraph 11 and the section on error and fraud in R&D reliefs. https://assets.publishing.service.gov.uk/media/6a4f66c406e7256c331df8a2/HMRC_annual_report_and_accounts_2025_to_2026.pdf (accessed 11 October 2026). Back
Canada Revenue Agency, Annual program statistics, Scientific Research and Experimental Development (SR&ED) tax incentive program (date modified 30 April 2026). https://www.canada.ca/en/revenue-agency/services/scientific-research-experimental-development-tax-incentive-program/annual-program-statistics.html (accessed 11 October 2026). Back
Revenue Commissioners, Research & Development ("R&D") Tax Credit Statistics (Statistics, Economics and Reporting Branch, March 2026), Tables 1 to 3. https://www.revenue.ie/en/corporate/documents/statistics/tax-expenditures/r-and-d-tax-credit-statistics.pdf (accessed 11 October 2026). Back
Revenue Commissioners, Annual Report 2025 (published 7 May 2026), "Audit and Compliance Interventions". https://www.revenue.ie/en/corporate/press-office/annual-report/2025/ar-2025.pdf (accessed 11 October 2026). Back
Inland Revenue Authority of Singapore, Research & Development (R&D) Tax Measures (last updated 13 August 2026), section on the benefits of the R&D scheme. https://www.iras.gov.sg/taxes/corporate-income-tax/income-deductions-for-companies/business-expenses/research-development-(r-d)-tax-measures (accessed 11 October 2026). Back
Internal Revenue Service, SOI Tax Stats: Corporation research credit, table list (page last reviewed or updated 9 November 2025). https://www.irs.gov/statistics/soi-tax-stats-corporation-research-credit (accessed 11 October 2026). Back
Internal Revenue Service, IRS sets forth required information for a valid research credit claim for refund, News release IR-2021-203, 15 October 2021. https://www.irs.gov/newsroom/irs-sets-forth-required-information-for-a-valid-research-credit-claim-for-refund (accessed 11 October 2026). Back
US Government Accountability Office, Tax Policy: The Research Tax Credit's Design and Administration Can Be Improved, GAO-10-136 (November 2009), pages 1 to 2 and 14 to 15 (with footnotes 18 and 19). https://www.gao.gov/assets/gao-10-136.pdf (accessed 11 October 2026). Back
Treasury Inspector General for Tax Administration, Controls for the Qualified Small Business Payroll Tax Credit Were Updated, but Ineligible Taxpayers Continue to Receive the Credit, Report No. 2026-408-012 (15 April 2026), Highlights. https://www.tigta.gov/sites/default/files/reports/2026-04/2026408012fr.pdf (accessed 11 October 2026). Back
Internal Revenue Service, Research credit claims (Section 41) on amended returns frequently asked questions, introduction and questions 3, 5 and 7 (page last reviewed or updated 1 October 2026). https://www.irs.gov/businesses/corporations/research-credit-claims-section-41-on-amended-returns-frequently-asked-questions (accessed 11 October 2026). Back
Internal Revenue Service, Large and Mid Size Business Division, Research Credit Claims Audit Techniques Guide (RCCATG): Credit for Increasing Research Activities Section 41, LMSB-04-0508-030 (May 2008), introductory note and Chapter II, "The RC Claim Engagement Letter". https://www.irs.gov/businesses/research-credit-claims-audit-techniques-guide-rccatg-credit-for-increasing-research-activities-section-41 (accessed 11 October 2026). Back
Internal Revenue Service, Instructions for Form 6765 (Rev. December 2025), "What's New" and Sections A to G (page last reviewed or updated 30 April 2026). https://www.irs.gov/instructions/i6765 (accessed 11 October 2026). Back
HM Revenue and Customs, Research and Development Tax Credits Statistics: September 2026 (updated 29 September 2026), main points and sections on number of claims and amount of support claimed. https://www.gov.uk/government/statistics/corporate-tax-research-and-development-tax-credit/research-and-development-tax-credits-statistics-september-2026 (accessed 11 October 2026). Back
Canada Revenue Agency, Form T661, Scientific Research and Experimental Development (SR&ED) Expenditures Claim, T661 E (26), Part 9, Claim preparer information, lines 935 to 975 and billing arrangement codes. https://www.canada.ca/content/dam/cra-arc/formspubs/pbg/t661/t661-26e.pdf (accessed 11 October 2026). Back
Revenue Commissioners, Corporation Tax: 2025 Payments and 2024 Returns, section 3.5 and Table 27. https://www.revenue.ie/en/corporate/documents/research/ct-analysis-2026.pdf (accessed 11 October 2026). Back

Cite this paper

Summary

1Summary

2Method

3United States

3.1What is published

3.2Older and narrower figures

3.3What is not published

4United Kingdom

4.1Error and fraud

4.2Compliance checks

4.3Disputes

4.4Agents

4.5Volumes

5Canada

6Ireland

7Singapore

8Comparing the record

8.1Why the figures cannot be compared

9The gap common to every country

10What the facts show

References

About this paper

Cite this paper

Summary

1Summary

2Method

3United States

3.1What is published

3.2Older and narrower figures

3.3What is not published

4United Kingdom

4.1Error and fraud

4.2Compliance checks

4.3Disputes

4.4Agents

4.5Volumes

5Canada

6Ireland

7Singapore

8Comparing the record

8.1Why the figures cannot be compared

9The gap common to every country

10What the facts show

References

About this paper

Related papers