At a glance
- HMRC’s Connect system recovered £4.6 billion in underpaid tax in 2024-25.
- Connect is a powerful data-mining tool, though HMRC is also investing in AI.
- It finds discrepancies by cross-referencing data from many different sources.
- Good bookkeeping and full client disclosure are key to avoiding investigations.
Even by the standards of government revenue, £4.6 billion is a decent pile of money. That’s the estimate of the tax underpayments that HMRC recovered in 2024-25 by using a custom-built computer system.
The system is called HMRC Connect. The £4.6 billion it found in 2024-25 is roughly 10% of the UK’s total “tax gap” – the gap between what was first paid to HMRC and what the organisation estimates taxpayers ought to have paid. The number came to light through a Freedom of Information request by law firm Pinsent Masons.
HMRC launched Connect in 2010. But the new £4.6 billion figure suggests that HMRC is lifting the 15-year old system’s effectiveness through some combination of greater resourcing, new data and new capabilities.
As the government seeks to shrink the tax gap, HMRC Connect recoveries are likely to keep rising.
Says tax lawyer Ian Robotham of law firm Pinsent Masons: “With thousands of HMRC staff now using Connect, taxpayers are facing a level of oversight that would have been unthinkable just a few years ago.”
How HMRC Connect works
Despite HMRC’s recent revelation on tax revenue, we know little about the huge system that underpins UK government revenue.
An HMRC representative describes it as a “powerful data networking, analytical and risking tool”. (A risking tool is a system or method used to identify, assess, and manage risks.) Wikipedia calls it a “social network analysis software data mining computer system”. It is designed to identify underpaid tax by bringing together many different data sources in a “data warehouse”, organising that data, and then analysing it to uncover patterns and relationships. For instance, it might link an individual with a family, a company, a trust, banking details, and an expensive car, and then display the information for an HMRC analyst. Consulting firm CapGemini, which worked on the system, said in a document produced around 2016 that HMRC was “one of the first organisations to use bulk data analytics and visualisation technology to combat tax evasion in this way”.
Such systems date back to the 1980s. Financial Accountant reported in 2016 that HMRC Connect was conceived around 2005.
“With thousands of HMRC staff now using Connect, taxpayers are facing a level of oversight that would have been unthinkable just a few years ago.”
Ian Robotham
HMRC does not itself specify Connect’s exact data sources, and Financial Accountant was not able to confirm them. But Pinsent Masons legal director Ian Robotham says HMRC has “spent time building up the amount of data sources that it can access and analyse.” And various outlets claim Connect has at least some access to individual and corporate tax returns, bank and credit card records, Land Registry data, flight sales, DVLA records, e-commerce records, online property rental platform data, and UK Border Agency information. Other outlets claimed in 2023 that Connect can access at least 55 billion items of data.
‘Connect isn’t an AI tool’
HMRC Connect has sometimes been described in the media as an “AI system”. But HMRC itself emphatically rejects this label; “Connect isn’t an AI tool”, an HMRC representative told us. And while HMRC has guarded about HMRC Connect and AI, nothing in the descriptions of the Connect system requires it to use AI at all. Data-mining existed long before the AI breakthroughs of the 2020s.
But it is also clear that over the years HMRC has incorporated AI functions into the organisation’s work. As early as 2020, several years before the current AI boom, a UK government document reported HMRC was using AI to support a number of activities. These included:
“… identifying risks on some large-scale transactional services, such as repayment claims for Value Added Tax (VAT) and Income Tax Self Assessment; using analytics to help identify risks that need attention, and building case packages that are passed to teams of investigators.”
That same 2020 document noted that HMRC did – and presumably still does – use AI in its compliance work:
“AI also works well to assimilate large amounts of data – this is a newer implementation, important for compliance casework where HMRC are using AI alongside other tools like geo-mapping.
MP Sir John Hayes asked the government earlier this year for what purposes HMRC had used AI in the previous 12 months. Chief Secretary to the Treasury James Murray’s tabled response said HMRC used predictive analytics tools in its compliance work, machine learning to help organise and search customer contact data, and machine learning tools to route correspondence. He avoided using the term “AI” at all.
But HMRC’s own 2025 Transformation Roadmap says the organisation is investing in AI-powered systems for compliance work, so that caseworkers “spend less time on administrative tasks and more time on their casework”.
Pinsent Masons have reported that HMRC has recently struck a deal with a US data analytics company which is “expected to see even more sophisticated AI applied to HMRC’s data.”
On a global view, HMRC’s activities seem fairly normal, if not restrained. An HMRC representative emphasises, for instance, that “we do not monitor social media in Connect” and that “we carry out social media monitoring in criminal investigations with proper legal oversight.” Contrast that with the Australian Tax Office, which as long ago as 2018 was talking proudly about its use of online signals like Facebook posts to uncover tax evasion.
What triggers investigations
HMRC tax investigations are likely to be triggered by discrepancies between different data sources, cross referenced between Connect and tax return data. For example, HMRC may notice that the level of income on a tax return is lower than the level indicated by the size of relevant bank account deposits.
Fiona Fernie, tax disclosures and investigations partner at Blick Rothenberg, says investigations are likely to be triggered by “inconsistencies between income declared on a tax return and other information Connect has access to, such as the level of bank deposits made in a year”. For example, she says, HMRC might decide on an investigation “if you say you are earning £25,000 but you have put £75,000 into your bank account”.
She stresses that this applies to more than just UK bank deposits. HMRC is also likely to be exchanging bank information with other jurisdictions. This raises the possibility that individuals may be caught out by overseas bank accounts not referenced in their tax returns.
Another potential area of mismatched data triggering investigations is the cross-referencing of property sales against Land Registry data to which Connect is likely to have access.
Yogesh Patel, director at Telic, notes the HMRC’s strong links with government agencies such as the Land Registry. Using Connect, HMRC can cross-reference the data on returns against the number of properties held by an individual and identify whether any have been sold.
This could catch out taxpayers who have disposed of or gifted a property but have not included it on their tax returns.
Once Connect identifies potential issues, HMRC staff are involved in decisions to do further work, though how they are involved is not entirely clear. An HMRC spokesperson told Financial Accountant that Connect is “not the sole deciding factor in beginning or deciding the direction of a tax investigation” and that other factors – including human insight – are also taken into account to make the final judgement.
How to remain compliant
Fernie believes that sole traders who are completing their own tax returns should consider engaging with professionals. That’s firstly because of the complexity of filings, and secondly to minimise the risk of submitting inaccurate data. Getting it wrong is likely to cost significantly more in the long run than the professional’s fee.
The incoming MTD for Income Tax regime raises the possibility of practitioners submitting quarterly returns to HMRC with estimated figures, with the final annual submission including figures to reflect the actual values of the period.
John Toon, head of technology at HLB International, recommends against this approach, because it raises the risk of false data matches in HMRC Connect. He stresses this shows good bookkeeping has never been more important in the context of HMRC’s enhanced data mining capabilities.
Ultimately, the best way to keep clients out of HMRC Connect’s grasp seems to be to ensure they stay in compliance with the law.
Client communication: a checklist
Typically, accounting firms inform clients that they are responsible for disclosing data on tax returns. This should be agreed with clients well in advance of tax returns being completed, as it isn’t uncommon for data and documents to be sent in by clients close to filing deadlines. Another approach is to send each client a disclosure checklist which covers all information gathering. This checklist should allow clients to confirm they have:
- detailed all bank accounts globally
- set out any and all cryptocurrency holdings; and
- for non-UK residents, kept a log of their visits to the UK.
On this last point, Patel notes: “We tell clients to keep an Excel spreadsheet of their movements, including days in and out of the UK.”
He also asks clients to inform his firm of any crypto investing, so as to minimise the risk of HMRC enquiries. This is becoming increasingly important: from 1 January 2026, UK-based crypto platforms will have to share information about their customers.
For its part, HMRC remains sensitive to suggestions that its increasing use of data is inappropriate. In that 2020 government report, it noted presciently that “as AI technology matures further, it will undoubtedly bring different ways of working, which will bring different cultural and educational challenges … HMRC recognise that being able to explain how AI is used is very important in terms of maintaining the trust of customers.”
Enhance your skills in emerging areas of cybersecurity and technologies with IFA’s self-paced short courses.
