# UK tax authority recovers and avoids losing £10 billion with AI and data

> UK tax authority credits AI and data analytics with recovering and protecting £10 billion in a year
> By Alfredo Collosa · Artificial Intelligence · 2026-10-17

The UK tax agency, HM Revenue and Customs (HMRC), credits the use of artificial intelligence and advanced analytics with protecting and recovering £10 billion in taxes during the 2025-2026 fiscal year. These technologies help select returns that merit investigation, review documentation and detect possible fraud. The figure combines the results of both tools and covers both money recovered and revenue whose loss has been avoided.

The agency reports these results in its annual report, covering the period from 1 April 2025 to 31 March 2026. The agency, which manages the taxes of some 40 million people and 5.7 million businesses, is also incorporating AI into taxpayer service and its employees' work.

The rollout comes in a record year for tax collection: £966.4 billion, £90.4 billion more than the previous year. Meanwhile, actions to improve compliance with tax obligations yielded £50.2 billion, the highest ever recorded by HMRC, though slightly below the £50.4 billion target.

## **Selecting what to investigate**

One of the main applications of AI is identifying returns that may contain inaccuracies. The technology helps direct investigations toward cases with the strongest signs of irregularities, a crucial task for an administration that processes millions of returns.

Once an investigation is opened, language processing tools make it possible to review large volumes of documents and locate relevant information. HMRC is also testing Copilot to summarize case files and is developing systems that let its employees consult guidance, background and datasets more quickly.

**[How does the tax authority investigate you? Spain shields the AI that selects taxpayers](https://observatorioblockchain.com/en/ai/espana-blinda-la-ia-que-selecciona-a-quien-investigar-en-hacienda/)**s

The agency has also launched a competition for private companies to propose new techniques against deliberate tax evasion. Two finalists will develop their proposals.

These actions are part of the effort to reduce the tax gap, the difference between the taxes that should be paid and those ultimately collected. The provisional estimate for 2024-2025 puts it at 6.4%. The UK goal is to raise an additional £10 billion a year by 2030 through improvements in tax compliance. Technology shares that task with a larger workforce: since the autumn 2024 Budget, more than 2,100 new tax compliance professionals have joined.

## **AI in phone calls**

Phone service is another area where HMRC is introducing these tools. The agency is testing a system that summarizes calls in real time so that the adviser can focus on resolving the query while the application documents the conversation.

It is also working with the Government Digital Service on GOV.UK Chat, a generative AI assistant that answers based on official guidance. On top of that comes a new service platform, announced in May 2026, whose phased rollout is planned from the 2026-2027 fiscal year.

These projects come after an improvement in phone service, historically problematic. During 2025-2026, 85.1% of attempts to reach an adviser were answered, compared with 71.5% the previous year. The average wait fell from 18 minutes and 38 seconds to 12 minutes and 35 seconds. The report does not allow that improvement to be attributed to the use of AI.

In international trade, the agency uses these technologies to redesign its tariff lookup tool and make it easier for businesses to understand their customs obligations.

## **28,000 Copilot licenses**

The transformation also affects internal work. HMRC distributed 28,000 Copilot licenses among its employees during the fiscal year, a rollout it presents as one of the largest in the UK government.

Training accompanies the introduction of the tools, though the report offers two figures that should not be confused. In the section dedicated to the Digital Academy, it states that 38,000 employees completed AI-focused training; in the highlights scorecard it cites 47,000 workers who completed a module on this technology. As these are different references, it cannot be assumed that they measure the same activity.

80% of staff completed at least one course on digital skills, data or technology. In April 2026, after the close of the fiscal year, HMRC appointed its first AI director to oversee its responsible development.

The modernization also includes retiring a 50-year-old VAT management system and migrating services to the cloud.

## **Who makes the decisions**

The use of AI to detect possible irregularities raises a central question for taxpayers: who decides and how an action can be explained.

HMRC maintains that these tools support its professionals and that final decisions rest with trained staff. According to the agency, results must be explainable and comply with data protection, security and ethics requirements.

To oversee that use, it has an assurance and risk management framework, an AI Council and a team responsible for its implementation. It also draws on external ethics advice through its Professional Standards Committee.

In test environments it uses synthetic data, created to develop and evaluate tools without using real taxpayer information. Systems must comply with UK data protection law and government frameworks, with specific limits for applications that interact directly with citizens.

## **Detecting impersonations**

The agency also uses machine learning to locate unauthorized uses of its name and image on social media. A tool developed with this technology identified 1,279 infringements during the year.

Across its security actions, HMRC removed more than 26,000 websites and social media accounts impersonating the organization. Its cyber defenses blocked some 23.35 billion potential threats, 99.95% of them automatically. These figures correspond to its general protection activity and should not be attributed entirely to AI.

The agency has also created a Fraud Prevention Centre following fraudulent access to PAYE payroll withholding services using stolen personal data.

## Trust remains pending

The technological results coexist with service problems. Only 44% of individuals say they trust HMRC, compared with 70% of small businesses. The average response time to complaints rose from 27 to 40 days and delays persist in correspondence and refunds.

The agency itself acknowledges that some measures intended to prevent errors and fraud harmed recipients of Child Benefit and apologized for it.

HMRC already uses AI to investigate, [handle queries](https://www.gov.uk/government/publications/hmrc-annual-report-and-accounts-2025-to-2026) and protect its services. Its report credits these tools and advanced data analytics with a considerable economic result. For taxpayers, the test will also be that a query is resolved sooner, that a refund arrives on time and that a wrong decision can be explained and corrected.
