HM Revenue & Customs has put Microsoft Copilot at the center of a broader modernization program that now reaches more than 28,000 staff, with plans to expand to 50,000 licences during 2026. The scale matters less as a government AI adoption headline than as a sign that HMRC is wiring generative AI, cloud hosting, automated customer support and consolidated data into everyday tax administration—not isolating them in pilots.
The department’s first annual Transformation Roadmap progress update, published July 2, lays out what has changed since its five-year roadmap arrived in July 2025. As detailed by HMRC and highlighted by Global Government Finance, the immediate targets are familiar to anyone who has dealt with a large public-sector IT estate: fewer legacy systems, faster customer service, more digital self-service, better fraud and compliance tooling, and a technology foundation that can survive peaks in demand.
For Windows administrators and enterprise IT teams, HMRC’s update is a useful case study in where Microsoft’s workplace AI tooling is actually being used in a large regulated organization. The department says Copilot is helping employees draft and summarize documents and emails, summarize meetings, and support other daily work. But it is pairing that rollout with role-specific training, synthetic-data testing environments, human review, and new operational systems rather than presenting Copilot as an autonomous decision-maker.
HMRC calls its Microsoft Copilot deployment one of the largest internal AI rollouts in UK government. Its 28,000 current licences are intended to grow to 50,000 this year, a significant proportion of an organization whose annual report records more than 66,000 full-time-equivalent staff in HMRC itself.
That deployment is accompanied by an unusually direct productivity claim. In a supplementary note to its 2025–26 annual report, HMRC says evaluation of its 2024 Copilot pilot estimated an average saving of about one hour a week per employee, translating into a claimed net productivity benefit of £50 million a year. That is an estimate, not a realized cash saving: the practical benefit is additional capacity, and whether it becomes better service or simply more output will depend on management and process design.
The department is also investing in the prerequisites that often determine whether these programs produce useful results. HMRC says tens of thousands of staff have completed AI-focused learning, while its roadmap describes internal “accelerator” initiatives combining policy, operational and digital specialists. It has also created synthetic-data environments intended to let teams and private-sector partners test AI tools without exposing live taxpayer information.
That framing is important. A desktop copilot can make an individual faster at turning notes into prose. It cannot by itself reconcile fragmented records, establish a defensible fraud decision, or ensure that an automated answer is appropriate for a taxpayer’s circumstances. HMRC appears to be treating Copilot as an employee productivity layer over a larger agenda of data consolidation, system replacement and workflow redesign.
HMRC has long faced criticism for phone-line performance, so the program is being introduced against measurable service pressure. Its annual report says the average telephone wait fell to 12 minutes and 35 seconds in 2025–26, from 18 minutes and 38 seconds a year earlier, while 85.1% of adviser-contact attempts were handled. HMRC says the wait time dropped below 10 minutes in March 2026 for the first time in more than four years.
The new platform is meant to add AI incrementally to call-queue management, wait-time estimates and digital-assistant support. Separately, HMRC is piloting AI call summarization that drafts notes from taxpayer conversations, with an adviser reviewing the output. It is also using interactive AI simulators to let customer-service staff rehearse difficult conversations without putting real customers at risk.
That is a much narrower, more defensible use of generative AI than handing customer decisions to a chatbot. Call summarization may reduce after-call work and improve consistency, but a human remains responsible for the final record. The same boundary should apply to the CCaaS rollout: operational gains are plausible, while accuracy, explainability and escalation paths will determine whether it improves trust.
Voice biometrics is already further along. HMRC says all individual customers calling speech-enabled lines are now offered the chance to use a unique voice print for identification. The technology could eliminate repetitive security questions, but it also puts biometric data and consent handling squarely into the service design. For a tax authority, the security case must be matched by clear alternatives for customers who cannot or do not want to use voice identification.
That is the less glamorous but more consequential part of the update. Public-sector AI initiatives often stumble on technical debt, fragmented identity systems, uneven data quality and brittle integrations. HMRC is explicitly tying its cloud work to a push for common platforms, stronger data capabilities and improved technical health.
The department currently rates its technical-health maturity at 3.3 out of five using Gartner’s PAID model, with a 2030 target of 4.0. Its annual report says it blocked about 23.35 billion potential IT security threats during 2025–26, with 99.95% handled automatically by its defenses. Those figures demonstrate the volume of hostile activity around a high-value government target, but they do not by themselves measure the security posture of the wider transformation.
For IT professionals, the operational lesson is straightforward: moving a service to cloud hosting and issuing Copilot licences are separate migrations with a shared dependency on identity, logging, data classification, access control and incident response. HMRC’s roadmap recognizes that its transformation is as much about secure foundations as it is about new interfaces.
The roadmap describes pilots using AI, data science and private-sector technology to address deliberate evasion and improve compliance. It also points to a Secure Digital Exchange Communications program for secure communications and file exchange, a planned Digital Disclosure Service targeted for 2027–28, and AI-powered guidance for caseworkers.
HMRC’s annual report goes further, claiming that AI and advanced analytics helped protect and recover £10 billion in tax during 2025–26. It also records 97 million unique records in a new Central Customer Registry, intended to build a more coherent view of taxpayers across systems. The figure is significant because model quality depends heavily on the completeness and relationship mapping of the underlying data.
But compliance automation is where governance becomes most acute. Risk-scoring can help direct investigators toward suspicious cases; it should not make opaque, unchallengeable judgments about individuals or businesses. HMRC’s references to human review, responsible AI leadership and controlled testing are welcome, yet the department will ultimately be judged on false positives, appeal outcomes, service accessibility and the clarity of its decisions.
Making Tax Digital for Income Tax, introduced in April 2026 for sole traders and landlords with gross income above £50,000, is the flagship service change. HMRC says more than 350,000 businesses have signed up. It is also moving new individual users toward GOV.UK One Login, with a plan to start onboarding existing individual customers in 2027.
The next meaningful checkpoint is not the licence count or another pilot announcement. It is whether phased CCaaS and CRM deployments in 2026–27 improve service without excluding customers, and whether HMRC can translate its large Microsoft Copilot rollout into controlled, auditable gains for staff rather than simply adding AI to an already complex estate.
The department’s first annual Transformation Roadmap progress update, published July 2, lays out what has changed since its five-year roadmap arrived in July 2025. As detailed by HMRC and highlighted by Global Government Finance, the immediate targets are familiar to anyone who has dealt with a large public-sector IT estate: fewer legacy systems, faster customer service, more digital self-service, better fraud and compliance tooling, and a technology foundation that can survive peaks in demand.
For Windows administrators and enterprise IT teams, HMRC’s update is a useful case study in where Microsoft’s workplace AI tooling is actually being used in a large regulated organization. The department says Copilot is helping employees draft and summarize documents and emails, summarize meetings, and support other daily work. But it is pairing that rollout with role-specific training, synthetic-data testing environments, human review, and new operational systems rather than presenting Copilot as an autonomous decision-maker.
Copilot Is Becoming Infrastructure, Not a Side Project
HMRC calls its Microsoft Copilot deployment one of the largest internal AI rollouts in UK government. Its 28,000 current licences are intended to grow to 50,000 this year, a significant proportion of an organization whose annual report records more than 66,000 full-time-equivalent staff in HMRC itself.That deployment is accompanied by an unusually direct productivity claim. In a supplementary note to its 2025–26 annual report, HMRC says evaluation of its 2024 Copilot pilot estimated an average saving of about one hour a week per employee, translating into a claimed net productivity benefit of £50 million a year. That is an estimate, not a realized cash saving: the practical benefit is additional capacity, and whether it becomes better service or simply more output will depend on management and process design.
The department is also investing in the prerequisites that often determine whether these programs produce useful results. HMRC says tens of thousands of staff have completed AI-focused learning, while its roadmap describes internal “accelerator” initiatives combining policy, operational and digital specialists. It has also created synthetic-data environments intended to let teams and private-sector partners test AI tools without exposing live taxpayer information.
That framing is important. A desktop copilot can make an individual faster at turning notes into prose. It cannot by itself reconcile fragmented records, establish a defensible fraud decision, or ensure that an automated answer is appropriate for a taxpayer’s circumstances. HMRC appears to be treating Copilot as an employee productivity layer over a larger agenda of data consolidation, system replacement and workflow redesign.
The Contact Center Will Be the More Visible AI Test
The most tangible customer-facing technology change is likely to come through HMRC’s planned Contact Centre as a Service, or CCaaS, platform and its planned enterprise CRM system. The department says both will go live in phases starting in 2026–27.HMRC has long faced criticism for phone-line performance, so the program is being introduced against measurable service pressure. Its annual report says the average telephone wait fell to 12 minutes and 35 seconds in 2025–26, from 18 minutes and 38 seconds a year earlier, while 85.1% of adviser-contact attempts were handled. HMRC says the wait time dropped below 10 minutes in March 2026 for the first time in more than four years.
The new platform is meant to add AI incrementally to call-queue management, wait-time estimates and digital-assistant support. Separately, HMRC is piloting AI call summarization that drafts notes from taxpayer conversations, with an adviser reviewing the output. It is also using interactive AI simulators to let customer-service staff rehearse difficult conversations without putting real customers at risk.
That is a much narrower, more defensible use of generative AI than handing customer decisions to a chatbot. Call summarization may reduce after-call work and improve consistency, but a human remains responsible for the final record. The same boundary should apply to the CCaaS rollout: operational gains are plausible, while accuracy, explainability and escalation paths will determine whether it improves trust.
Voice biometrics is already further along. HMRC says all individual customers calling speech-enabled lines are now offered the chance to use a unique voice print for identification. The technology could eliminate repetitive security questions, but it also puts biometric data and consent handling squarely into the service design. For a tax authority, the security case must be matched by clear alternatives for customers who cannot or do not want to use voice identification.
Cloud Migration Is the Quiet Enabler
HMRC’s roadmap says more than half of its services and underlying IT have either been decommissioned or moved from on-premises servers to secure cloud hosting. The stated goal is an estate that is more resilient, easier to update, and easier to scale during demand spikes.That is the less glamorous but more consequential part of the update. Public-sector AI initiatives often stumble on technical debt, fragmented identity systems, uneven data quality and brittle integrations. HMRC is explicitly tying its cloud work to a push for common platforms, stronger data capabilities and improved technical health.
The department currently rates its technical-health maturity at 3.3 out of five using Gartner’s PAID model, with a 2030 target of 4.0. Its annual report says it blocked about 23.35 billion potential IT security threats during 2025–26, with 99.95% handled automatically by its defenses. Those figures demonstrate the volume of hostile activity around a high-value government target, but they do not by themselves measure the security posture of the wider transformation.
For IT professionals, the operational lesson is straightforward: moving a service to cloud hosting and issuing Copilot licences are separate migrations with a shared dependency on identity, logging, data classification, access control and incident response. HMRC’s roadmap recognizes that its transformation is as much about secure foundations as it is about new interfaces.
Tax Compliance Is the Commercial Case for Data and AI
The strongest financial justification for the program sits in compliance. HMRC’s latest official estimate puts the 2024–25 tax gap—the difference between tax theoretically owed and tax actually paid—at 6.4%, or £59.2 billion. The department cautions that the number is backward-looking and therefore does not capture the most recent policy or investment changes.The roadmap describes pilots using AI, data science and private-sector technology to address deliberate evasion and improve compliance. It also points to a Secure Digital Exchange Communications program for secure communications and file exchange, a planned Digital Disclosure Service targeted for 2027–28, and AI-powered guidance for caseworkers.
HMRC’s annual report goes further, claiming that AI and advanced analytics helped protect and recover £10 billion in tax during 2025–26. It also records 97 million unique records in a new Central Customer Registry, intended to build a more coherent view of taxpayers across systems. The figure is significant because model quality depends heavily on the completeness and relationship mapping of the underlying data.
But compliance automation is where governance becomes most acute. Risk-scoring can help direct investigators toward suspicious cases; it should not make opaque, unchallengeable judgments about individuals or businesses. HMRC’s references to human review, responsible AI leadership and controlled testing are welcome, yet the department will ultimately be judged on false positives, appeal outcomes, service accessibility and the clarity of its decisions.
Digital-First Still Has to Mean Inclusive
The roadmap’s customer-service ambition is for at least 90% of interactions to be digital by 2030. It is already at 78%, according to HMRC, and its app had 7.6 million unique users in 2025–26, up from 5.9 million the prior year. The department says it aims to reduce postal outputs by about 75% and save £50 million a year by 2028–29, while retaining paper communication for people who need it.Making Tax Digital for Income Tax, introduced in April 2026 for sole traders and landlords with gross income above £50,000, is the flagship service change. HMRC says more than 350,000 businesses have signed up. It is also moving new individual users toward GOV.UK One Login, with a plan to start onboarding existing individual customers in 2027.
The next meaningful checkpoint is not the licence count or another pilot announcement. It is whether phased CCaaS and CRM deployments in 2026–27 improve service without excluding customers, and whether HMRC can translate its large Microsoft Copilot rollout into controlled, auditable gains for staff rather than simply adding AI to an already complex estate.
References
- Primary source: Global Government Finance
Published: 2026-07-20T10:36:27+00:00
- Related coverage: gov.uk
HMRC's Transformation Roadmap - GOV.UK
The government's plan to transform the tax and customs system.www.gov.uk
- Related coverage: assets.publishing.service.gov.uk
HMRC_Circle_Element_Light-Green_RGB
Mature businessman presenting new project ideas to the team in meeting room. Businessman standing by large screen and brainstorming project segments in boardroom.assets.publishing.service.gov.uk
