LONDON — His Majesty’s Revenue and Customs announced Wednesday that it will deploy artificial intelligence tools developed by Edinburgh-based software firm Claribel Analytics to help identify fraudulent tax submissions and errors in self-assessment returns, in what officials described as the most significant upgrade to the agency’s compliance and fraud-detection systems in more than a decade. The five-year contract, valued at approximately £340 million, is expected to generate hundreds of millions of pounds in additional tax recovery annually once fully operational and represents a major step in the government’s broader ambition to modernise public-sector data infrastructure.
Under the agreement, Claribel’s platform will be integrated into HMRC’s existing data architecture beginning in September, with a phased rollout expected to extend coverage to the full self-assessment filing population of roughly 12 million individuals and 1.8 million small businesses by early 2027. The system combines large-language model document analysis with structured statistical anomaly detection to flag returns that deviate from expected patterns derived from the filing history of comparable taxpayers, segmented by sector, income bracket, geography, and business type. Cases above a defined risk threshold are then surfaced to human compliance officers for further investigation.
HMRC’s chief digital officer, Rachel Forsythe, said the platform would allow the agency to direct its compliance resources toward cases with the highest probability of genuine underpayment rather than relying on the random sampling and largely manual review processes that have formed the backbone of self-assessment compliance for the past 20 years. “We are not replacing human judgment — we are giving our people better intelligence,” Forsythe said in a statement. “Every pound we recover through more accurate and targeted compliance goes directly to funding public services.” The agency projected the system could help recover between £1.8 billion and £2.3 billion in additional tax revenue per year at full operational scale, though it cautioned the figure was based on modelling conducted during the pilot phase and was not a guaranteed outcome.
Claribel Analytics, founded in 2019 and backed by a consortium of British institutional investors including several pension funds, has previously delivered narrower fraud-detection systems for tax authorities in Ireland and the Netherlands, but the HMRC deal is by far its largest and most complex contract to date. Chief executive Dr. Amir Ghosh said the company’s system had been trained on anonymised HMRC datasets spanning 15 years of filing history, covering more than 180 million individual return records. During a six-month controlled pilot conducted across three HMRC regional compliance offices, Ghosh said the platform demonstrated a precision rate of 91 percent in surfacing cases that human investigators subsequently confirmed warranted formal inquiry — a significant improvement over the 67 percent hit rate the agency reported from its existing screening methods.
The announcement drew immediate and pointed scrutiny from civil liberties organisations and digital rights campaigners, who raised concerns about algorithmic bias and the risk that the system could disproportionately flag returns from self-employed workers, gig economy participants, and ethnic minority-owned small businesses — groups that prior advocacy research has shown already face higher rates of unwarranted HMRC inquiry relative to their filing population size. The Open Tax Rights Alliance called on the government to publish the model’s full auditing methodology and training data provenance before permitting any live deployment and to establish a permanent independent oversight board with powers to audit outcomes and compel adjustments. “Automated systems that directly affect people’s finances and livelihoods must be transparent, explainable, and accountable,” said the group’s director, Miriam Osei. “Announcing a procurement contract is not the same as demonstrating a system is fair or free from embedded bias.”
HMRC said it had commissioned an independent equalities impact assessment from third-party auditor Vertex Assurance and would publish a summary of findings before the September integration launch. Officials confirmed that any self-assessment return flagged by the AI platform would be subject to review by a qualified human compliance officer before any formal investigation notice was issued to the taxpayer, and that individuals would retain the full existing suite of appeal and tribunal rights available under current tax legislation. The agency said it was also implementing an internal monitoring regime to track demographic patterns in flagged cases and compare them against the broader filing population on a quarterly basis.
Tax professionals broadly welcomed the investment but warned that the agency would need to manage an anticipated surge in formal compliance correspondence with care. “The risk is that a system optimised to surface errors will inevitably also generate a meaningful volume of false positives that consume officer time and cause considerable distress to taxpayers who have filed correctly,” said Marcus Webb, a partner at accountancy firm Caldwell Grant. “How HMRC manages the communication, the appeals process, and the human review quality will ultimately determine whether this becomes a genuine compliance success story or an expensive public relations problem.” HMRC said it was expanding telephony capacity and digital self-service support options in anticipation of increased contact volumes from September onward.
The contract also includes a performance-linked payment structure under which a portion of Claribel’s fees over years three through five will be tied to independently verified tax recovery outcomes, giving the firm a direct financial stake in the accuracy and effectiveness of its platform. Officials said this represented an unusual degree of accountability for a government technology procurement and was designed to align the supplier’s incentives with public-interest outcomes. Claribel confirmed the arrangement and said it was confident the system’s performance would meet or exceed the thresholds required.