HMRC has granted a £175 million AI contract to the British company Quantexa in an effort to address the £46.8 billion tax gap.

      TL;DR: HMRC has awarded a £175 million, ten-year AI contract to London-based company Quantexa, aimed at detecting tax fraud, correcting errors, and bridging the £46.8 billion tax gap. This marks a strategic move towards British AI independence after previously spending over £900 million on contracts with the American firm Palantir across various UK government departments.

      HM Revenue and Customs has granted a £175 million, decade-long contract to Quantexa, a London-based AI firm, to enhance the tax authority’s data systems and utilize AI for fraud detection, error correction, and closing the tax gap. This contract represents one of the largest AI agreements in the UK public sector. It also signals a move towards sovereignty, as the UK government previously invested over £900 million in Palantir, the American data analytics company, across multiple departments including the NHS. The Quantexa contract differs fundamentally as it involves a British company, with data managed within HMRC’s own system. This decision was intentional.

      The tax gap, which is the disparity between taxes owed to the government and what is collected, reached £46.8 billion for the fiscal year 2023-24. The government aims to recover an extra £10 billion annually by 2030, and Quantexa's role is to assist in this effort.

      The company Quantexa was established in 2016 by Vishal Marria, who previously served as the youngest executive director at Ernst and Young, where he directed anti-financial crime tech initiatives for global banks. Quantexa's "Decision Intelligence" platform integrates data from various sources, resolves identities within datasets, and employs graph analytics and machine learning to unveil patterns that human analysts may miss at scale. Initially, it focused on anti-money laundering in the banking sector, with HSBC being one of its first clients. The technology that enabled banks to identify suspicious transactions between shell companies is now being employed by a tax authority to spot dubious patterns in numerous tax returns.

      Quantexa reported revenues of £126 million for the year ending March 2025, reflecting a 49% year-over-year increase, with annual recurring revenue surpassing $100 million. In March 2025, the company secured $175 million in a Series F funding round, achieving a valuation of $2.6 billion, supported by investors including Warburg Pincus, HSBC, and the Ontario Teachers’ Pension Plan, with total funding amounting to $546 million. Quantexa operates in over 70 countries, serving clients in finance, telecommunications, and the public sector, including Vodafone.

      With a valuation of $2.6 billion, Quantexa has become the UK’s most valuable generative AI company, just ahead of Synthesia, which is valued at $2.1 billion. These companies exemplify the distinct strengths of British AI: Synthesia in generative media and Quantexa in enterprise decision intelligence. Neither creates foundational models, but both develop applications that leverage such models, which has proven commercially successful for European AI firms.

      The HMRC contract encompasses three key capabilities. First, Quantexa will upgrade HMRC's primary data infrastructure, establishing a connected framework that currently faces fragmentation due to separate systems. HMRC manages hundreds of millions of transactions yearly across various tax types. Although the data exists, it is scattered. Quantexa's entity resolution technology will link records referring to the same individual, company, or transaction, thereby creating a comprehensive view and revealing patterns.

      Second, the platform will utilize AI to detect tax fraud and evasion, including covert networks of individuals and companies engaged in fraudulent activities. Graph analytics will trace the connections between multiple layers of corporate ownership, a feat that is virtually impossible for human auditors using spreadsheets but manageable by software engineered to analyze intricate networks.

      Third, the system will enhance HMRC's customer service operations, enabling staff to resolve inquiries more efficiently. Public frustration with HMRC has increased significantly, with over 93,000 complaints submitted in 2024-25, up from 70,000 complaints in 2020-21. Slow response times are a key concern. AI-driven case management, including generative AI that can summarize documents and address caseworkers' inquiries, will be part of the solution.

      Marria informed the BBC that while automated taxpayer-related decisions will require human oversight, he stressed that AI should not function as a "black box" in government contexts. Decisions must be transparent, auditable, and understandable, especially in areas impacting citizens directly. He assured that HMRC data would solely reside within HMRC's domain and never be shared externally.

      Regarding sovereignty, NHS England has permitted Palantir contractors broader access to identifiable patient data via the Federated Data Platform, a move criticized by patient advocacy groups and Labour MPs. The £330 million contract, awarded by the former Conservative government, positioned an American surveillance firm at the core of NHS data infrastructure. Science minister Patrick Vallance has indicated that future government