HMRC has been quietly building one of the most sophisticated tax compliance AI systems in the world. Most UK corporates are aware that HMRC uses data analytics. Fewer appreciate how far that capability has advanced — or how quickly it is accelerating.
This article explains what HMRC can actually see, how their AI systems work, and what in-house tax teams should do about it.
The Connect system: HMRC's data backbone
Connect is HMRC's primary data analytics platform, operational since 2010 but substantially enhanced every year since. The system aggregates data from over 30 different sources — Companies House filings, Land Registry, DVLA, financial institutions, online marketplaces, and international information exchange agreements — and uses machine learning to identify discrepancies between what taxpayers report and what the data suggests they should report.
The scale is significant. HMRC processes over a billion data records annually through Connect. The system's stated purpose is to identify cases for investigation that would otherwise require manual review — which HMRC simply does not have the resource to conduct at scale.
For large UK corporates, the relevant question is not whether HMRC's AI can see your tax position. It can. The question is whether what it sees is consistent with what you've disclosed.
What HMRC's AI is looking for
HMRC's public statements on their AI capabilities, combined with analysis of their operational guidance and tribunal cases, reveal several priority areas for machine learning-assisted compliance:
Transfer pricing mismatches. The Connect system cross-references intercompany transactions against industry benchmarks and public comparable data. Significant deviations — without corresponding disclosure — are flagged automatically.
Uncertain tax positions. Following the extension of the Uncertain Tax Treatment (UTT) notification regime, HMRC's systems are specifically designed to identify positions that meet the UTT threshold but have not been notified. The algorithm compares reported tax with what the system expects given the company's size, industry, and disclosed transactions.
R&D tax credit claims. This has been a high-profile focus since HMRC's crackdown on inflated R&D claims. Machine learning is used to identify claim patterns that deviate significantly from sector norms.
VAT input tax recovery. Automated analysis of VAT return patterns against company financial data can flag anomalies in partial exemption calculations and input tax recovery rates.
HMRC's AI roadmap: what's coming in 2026
HMRC's digital transformation strategy, published in their Transformation Roadmap, sets out several AI-related developments that take effect in 2026:
UTT expansion (June 2026). The Uncertain Tax Treatment notification threshold is being extended to apply to more companies. Businesses that previously fell below the threshold will be in scope. HMRC's systems will be calibrated to identify companies now subject to the regime that have not yet filed notifications.
AI governance standards (January 2026). New standards require tax functions to document their use of AI in compliance processes. This creates Senior Accounting Officer (SAO) exposure for companies that cannot demonstrate appropriate governance of AI-assisted tax decisions.
Real-time data access. HMRC's Making Tax Digital programme is expanding its scope. Combined with API access to accounting software, this significantly reduces the lag between a transaction occurring and HMRC being able to see it.
What this means for in-house tax teams
The practical implication is straightforward, if uncomfortable: HMRC's AI is now ahead of many in-house tax processes. A well-resourced three-person tax team using spreadsheets and judgment is facing off against a system that processes a billion data points annually and has been specifically calibrated to find what they might have missed.
This is not an argument for panic. It is an argument for process improvement.
Review your UTT position before June 2026. If your company will be in scope of the expanded UTT regime, identify your uncertain positions systematically — not just through judgment, but through a documented, reproducible process. HMRC will be looking for companies newly in scope that have not yet filed notifications.
Document your AI governance. If your tax team is using AI tools — even general-purpose tools like ChatGPT for research — you need to document how those tools are used, what human review is applied, and how the outputs are verified. The January 2026 AI governance standards create SAO accountability for this.
Assess your data readiness. If HMRC's system can produce a more coherent picture of your tax position than your own workpapers, that is a risk. The quality of your source data matters — both for your own decision-making and for your ability to respond to HMRC enquiries coherently.
Take transfer pricing documentation seriously. If your intercompany pricing is not supported by robust contemporaneous documentation, the Connect system is likely already flagging you. The question is whether it will result in an enquiry.
The right framing
The HMRC AI story is often framed as threatening. It is more usefully framed as an alignment problem. HMRC's AI is trying to build a picture of your tax position from external data. Your goal is to ensure that picture is accurate — and that your disclosures pre-emptively answer the questions it will raise.
Tax teams that approach HMRC's AI capabilities as an input into their own compliance processes — asking "what would HMRC's system see when it looks at our filings?" — are better placed than those that are simply hoping not to be selected for investigation.
The tools to do this are increasingly available. The regulatory pressure to do it is now unambiguous.
Tax Haus publishes practical analysis on AI and UK corporate tax. Subscribe to the briefing or explore our services.
Topics