Quantexa Wins £175M Public Sector AI Deal With HMRC

HM Revenue and Customs has awarded a £175 million, decade-long contract to Quantexa, the London-founded artificial intelligence company, to overhaul the tax authority’s data infrastructure and deploy AI across fraud detection, compliance and taxpayer services. The deal is one of the largest public sector AI contracts ever signed in the UK and marks a deliberate turn toward domestic technology for work that has historically gone to American providers.

The contract tasks the company with building what it calls a sovereign data foundation for HMRC. The tax authority’s systems are fragmented across legacy platforms that make it difficult to get a unified picture of suspected fraud networks, connected entities or even basic payment errors. Investigators trying to trace organised VAT or payroll fraud through that patchwork have, until now, been doing much of it by hand. The platform will stitch together billions of data points from internal and external sources, giving caseworkers a single connected view of the relationships, shell companies and layered transactions that sit behind large-scale tax fraud. It will also automate the less visible but equally costly work of resolving unmatched payments, catching return errors and improving a customer service operation that fielded more than 93,000 complaints last fiscal year. For HMRC, which has publicly stated its ambition to become the most AI-enabled tax authority in the world, the contract represents the most significant step yet toward that goal.

The political pressure behind the contract is significant. The UK tax gap, the difference between what the Exchequer is owed and what it actually collects, stands at £46.8 billion, equivalent to 5.3% of total tax liabilities. Tax evasion alone accounts for an estimated £5.5 billion a year. The government has pledged 5,500 new compliance caseworkers over the next five years to help close that gap, but headcount without better technology only goes so far. Quantexa is the bet that connecting the data will achieve what adding bodies alone cannot.

Vishal Marria founded the company in 2016 after watching the problem from the inside. As the youngest executive director EY had ever appointed, he spent years on secondments to HSBC and Deutsche Bank, where he sat with risk managers who were trying to figure out whether clients were clean or dirty by manually pulling data from half a dozen disconnected systems and cross-referencing it by hand. He has described it as trying to buy a house after only peeking through the letterbox. He left to build something better. The idea, as Marria has told it, crystallised at a conference in Las Vegas when he called his wife and told her he was starting a company. The company’s earliest work involved investigating complex money-laundering networks, and it led to real arrests, including the dismantling of a child trafficking ring identified through a major UK bank. That track record in financial crime and anti-money laundering is a significant part of what brought HMRC to the table. Tax fraud at the scale the government is trying to address relies on many of the same structures, layered ownership, cross-border transactions and obscured beneficial interests, that Quantexa was originally built to uncover.

A decade on, the company serves HSBC, Standard Chartered and BNY across financial crime, risk and compliance. Revenue reached £126 million in the most recent fiscal year, up 49%, and a $175 million Series F round led by Ontario Teachers’ Pension Plan valued the business at $2.6 billion. The HMRC contract now gives Quantexa one of the most prominent public sector reference projects in Europe at a time when governments across the continent are weighing how to deploy AI in sensitive national functions.

The sovereignty dimension of the deal is hard to miss. The UK government has spent upwards of £900 million on contracts with Palantir Technologies, the Peter Thiel-backed American data firm, across at least ten departments including the NHS. Those deals have attracted sustained parliamentary scrutiny over who controls the data, where it sits and what a foreign company with close ties to US intelligence agencies can do with it. Quantexa’s model is designed to neutralise all of that. The platform operates entirely within HMRC’s own secure environment. Taxpayer records never leave government infrastructure. Every automated decision is explainable and fully auditable, which matters when investigations end up in court.

“I don’t have access to their data,” Marria said. “We provide the transparency and the explainability. Everything the platform does is fully auditable.”

AI Minister Kanishka Narayan framed the partnership in explicitly national terms, calling Quantexa a company at the forefront of British innovation and saying it would help HMRC deploy AI safely at scale. The broader direction of travel supports that language. The government has committed £500 million through its Sovereign AI Unit to back domestic AI companies. HMRC has appointed a chief AI officer, James Mitton, and is rolling out tens of thousands of Microsoft Copilot licences across its workforce. The Making Tax Digital programme has forced quarterly digital filing on hundreds of thousands of sole traders and landlords, dramatically increasing the volume of structured data flowing into the system. The platform is designed to process exactly that kind of scale, identifying patterns and anomalies that human auditors working through legacy systems would miss.

Now it has to deliver. Marria has said he wants to move HMRC toward richer investigations that go after professional criminal networks rather than tripping up honest taxpayers caught in paperwork errors. Fewer false positives. Faster case resolution. A system that learns and gets smarter as the data grows. The contract is ten years long, which means both sides are locked into making it work. For Quantexa, the stakes extend beyond a single client. A successful HMRC deployment would give the company a blueprint it can take to tax authorities and public sector agencies around the world. A failed one would set back the case for domestic AI procurement by years. Every government department currently running Palantir will be watching to see which way it goes.