Yesterday it was announced HMRC has awarded a £175 million, 10-year contract to a British AI firm to do one thing above all others: give the UK’s tax authority a clearer, connected view of its data to improve performance, help identify tax at risk, and strengthen control.
The firm in question is Quantexa — a company EmpowerRD happened to be nominated alongside at last night’s National Technology Awards, in the Tech Growth Business of the Year category. We’re proud to share a shortlist with them. But we’d be doing our clients a disservice if we let the award excitement distract from what this contract actually signals for every UK company claiming R&D tax credits.
This is not a routine IT upgrade. Britain’s tax collection agency has handed Quantexa a £175 million contract to deploy AI systems that will hunt down fraud and catch tax return errors across millions of filings — a deal that marks one of the largest government AI procurement contracts in UK history and signals a major shift toward automated compliance enforcement.
If your R&D claim isn’t built to withstand that level of scrutiny, today is the day to fix it.
What This AI Is Actually Designed to Do
Let’s be direct about technology. Quantexa’s platform doesn’t work like typical keyword-matching software. The company specialises in contextual decision intelligence — teaching machines to connect dots across massive datasets. The system can link shell companies, trace hidden ownership structures, and spot anomalies in spending patterns that might indicate tax evasion or honest mistakes that need correcting.
Applied to R&D tax credits, that capability matters enormously. HMRC already knew roughly how much it was losing. The UK’s tax gap — the difference between taxes owed and collected — sits at an estimated £36 billion annually. R&D tax credits have been a focal point of that concern for years. HMRC created a dedicated R&D Anti-Abuse Unit in July 2022, after its own published analysis revealed that almost 25% of claims in the (now reformed) SME scheme were erroneous or fraudulent.
What’s new is the scale and sophistication of HMRC’s detection capability. Traditional audit methods couldn’t scale to catch sophisticated fraud schemes or even simple errors across the millions of returns filed each year. Now they can. For HMRC, this means potentially catching fraud before refunds go out the door rather than chasing money years later through costly enforcement.
The direction of travel is clear: HMRC is building an intelligence-led compliance function. The question for every R&D claimant is whether their claim can stand up to it.
The Risk Is Already Real — This Makes It More So
Some will read this news and think: “I’m a legitimate claimant. This doesn’t apply to me.”
That is the wrong conclusion.
Even well-intentioned claims are at risk when they’re poorly documented. If your claim’s technical narrative resembles templated reports, or costs sit outside sector norms without explanation, HMRC’s risk tools may rank you at higher risk. The AI doesn’t distinguish between a poorly evidenced legitimate claim and a fraudulent one — both look the same to a pattern-matching system.
HMRC compliance checks were already rising sharply before yesterdays announcement. The number of compliance checks for ‘high-risk’ claims increased from 10% in 2023 to 17% in 2024. And the AI system HMRC is now deploying at national scale is explicitly designed to identify tax at risk and strengthen control — including across R&D filings.
There’s a further dimension here that has been building for some time. HMRC’s failure to either confirm or deny its use of AI has reinforced the belief that AI is being used by its case officers. This undermines taxpayers’ trust and discourages legitimate claimants — hindering the policy objective of the R&D tax relief scheme. The Quantexa contract makes it explicit: AI is now formally, publicly, and massively embedded into HMRC’s compliance infrastructure. There is no more ambiguity.
“This is not a reason to avoid claiming. It is a reason to claim properly — with evidence that survives scrutiny, not just software.”
The Compliant Majority Is Paying for the Non-Compliant Minority
This is not a new problem, but it’s about to get sharper. The compliant majority are paying for the minority’s misdeeds through increased processing periods and enhanced HMRC compliance checks and enquiries.
When HMRC’s new AI flags a claim as anomalous, the burden of proof falls on the claimant. If your technical narrative is generic. If your cost methodology hasn’t been stress-tested. If your supporting documentation was assembled quickly at year-end. You will be the one answering the enquiry letter.
For HMRC, the Quantexa agreement forms part of a broader strategy to become “an agile department supported by a modern IT infrastructure.” The department is pursuing this through investments in emerging technologies, including generative AI and cloud platforms, legacy system modernisation, and strategic partnerships with technology providers. This is a long-term structural change, not a campaign. The compliance environment for R&D claimants will only become more technically capable over time.
What This Means for Your Claim
Here is what EmpowerRD clients — and any company currently claiming R&D tax credits — should be doing right now:
- Your technical narrative must be specific. Generic descriptions of “innovative software development” or “overcoming technical challenges” are not defensible. HMRC’s AI will compare your language against sector norms and prior claims. Your narrative needs to articulate specific technological uncertainties, specific approaches taken, and specific reasons why the work qualifies. If you can’t explain it precisely, neither can your claim.
- Your cost methodology must be traceable. Staff time apportionment needs to be grounded in actual project records, not estimates assembled after the fact. Subcontractor costs need contracts and evidence of qualifying activities. Externally provided workers need correct categorisation under the RDEC or SME rules. An AI system built to spot anomalies in spending patterns will find the gaps that a human reviewer might overlook.
- Your claim cannot rely on volume — it must rely on substance. The era of high-volume, low-scrutiny R&D claims is over. Every line of your claim needs to be supportable in an enquiry. That means primary-source documentation, project timelines, technical evidence
