HMRC has handed Quantexa a 10-year, £175 million AI deal at a moment when tax enforcement and taxpayer trust are both under pressure.
HM Revenue and Customs is not just buying another software tool. It is betting that a British AI company can help it find fraud, correct mistakes and make sense of the messy data trails that sit behind modern tax returns.
The department announced on May 14 that London-based Quantexa will provide AI-powered data technology under a 10-year contract worth £175 million. The system is expected to combine HMRC's own data with external sources, flag suspicious patterns, identify hidden networks of companies and individuals, and help customer-service staff work through cases more quickly.
As the BBC reported, the deal comes as HMRC faces rising frustration over its service levels, including more than 93,000 complaints in 2024-25. That timing matters. When people cannot get a tax problem fixed, a promise of more automation can sound less like progress and more like another layer between taxpayers and a human being who can help.
That is why this contract is a useful test case for government AI. Fraud detection is one of the clearest use cases for machine learning because the data is rich, the patterns are hard to spot manually and the financial upside can be large. But tax is also personal. A false positive can mean a frozen payment, an investigation letter, or weeks of anxiety for someone who made an ordinary mistake.
Quantexa is not an overnight AI story. Founded in 2016, the company built its reputation in financial crime, anti-money laundering and decision intelligence, helping banks connect data points that might otherwise sit in separate systems. In March 2025, it raised $175 million in Series F funding at a $2.6 billion valuation, led by Teachers' Venture Growth, with British Patient Capital also participating.
That funding round was not only about chasing private-sector banks. Quantexa said at the time that public-sector demand was growing, and the HMRC contract now shows why. Governments have many of the same problems as banks: fragmented data, fraud risk, identity questions and legacy systems that were not designed for the speed or scale of today's digital activity.
For startups, this is the part of the AI market that can be less glamorous but more durable. Consumer chatbots get the attention. Public-sector compliance gets decade-long contracts. If Quantexa can prove that its platform reduces losses while improving service, it gives the company a stronger story than most AI vendors selling pilots and dashboards.
There is also a national angle. The UK government has been trying to present Britain as an AI power, but much of the infrastructure market is dominated by American hyperscalers and foundation-model companies. A British firm winning a major AI procurement from one of the country's most important departments gives policymakers something more concrete to point to.
Trust will decide whether the technology works
The word sovereign is doing a lot of work in this deal. Quantexa has described the programme as a sovereign, governed AI transformation, which is exactly the kind of language governments now want to hear. They need tools that can explain decisions, manage sensitive data and satisfy public scrutiny. They also need suppliers that do not leave them looking careless with taxpayer information.
HMRC has reason to be cautious. Earlier this year, the department faced criticism over an anti-fraud scheme involving Home Office travel records and child benefit claims, after thousands of households had payments stopped incorrectly. That episode is a reminder that bad data can make a confident system behave badly. AI does not remove that risk. It can make it faster.
Quantexa's chief executive Vishal Marria has said the technology is meant to support human decision-making rather than replace it. That distinction will matter in practice, not just in public statements. The question is whether HMRC staff will have enough context, authority and time to challenge system outputs when the data looks wrong or incomplete.
Tax enforcement also has to separate fraud from error. A network of shell companies hiding income is one problem. A self-employed worker who misfiles a return is another. If the system treats both with the same level of suspicion, HMRC may recover more money in some cases while damaging confidence in the process overall.
The opportunity is still real. HMRC's data estate is large, and connecting it properly could help the department identify tax at risk earlier, reduce manual casework and give staff better information when taxpayers ask for help. For small businesses, that could eventually mean fewer repeated requests, faster corrections and less time spent proving information the government already has somewhere else.
The danger is that efficiency becomes the only scoreboard. A system that flags more cases is not automatically a better system. The real measure will be whether HMRC can reduce fraud and error without creating a new class of automated disputes that taxpayers have to fight their way through.
This is where Quantexa's government ambitions will be tested. Winning a £175 million contract is a milestone. Making AI feel fair, accurate and useful inside the tax system is the harder work. If HMRC gets that balance right, public-sector compliance could become one of the strongest commercial markets for British AI. If it gets it wrong, the debate will not be about innovation. It will be about who pays when the machine points at the wrong person.
Also read: India is becoming the first real test of AI job disruption • AI is speeding up Linux flaw discovery as Fragnesia hits servers • The FDA is bringing real-time AI oversight into clinical trials
