The future of public service delivery will be increasingly shaped by artificial intelligence (AI). That message is at the heart of the government’s newly published Artificial Intelligence Playbook, which provides practical guidance for departments to embed AI in ways that are safe, ethical and grounded in real-world outcomes.
But while the playbook offers a roadmap, the real test lies in how departments implement it amid operational constraints, complex legacy systems and rising public expectations. It is the lived experience of frontline teams that reveals what successful AI deployment truly requires: clarity of purpose, confidence to experiment and a strong culture of collaboration.
Those themes were front and centre in the latest episode of Changemakers, Civil Service World’s new podcast series, which showcases innovation across government. In a discussion sponsored by Capita, CSW editor-in-chief Suzannah Brecknell spoke with Ben Lovatt, deputy director of the Contact Engagement Platform at HMRC; Duncan Brown, head of software engineering at the Department for Science, Innovation and Technology’s AI incubator; and Owen Barry, Capita’s managing director for central government, about how automation, robotics and agentic AI are reshaping public sector operations – and what it will take to go further.
Vision, purpose and risk appetite
Lovatt explained that the starting point of his team’s journey to implement robotic process automation wasn’t a new product but a pressing operational challenge. “HMRC’s architecture is legacy. A lot of it is very, very old. Tying those systems together is difficult.”
Tasks that required navigating 10, 20 or 30 different systems created delays, inconsistency, and inefficiencies. As HMRC completed a transformation programme which involved moving into fewer regional centres and faces pressure to do more with less, it needed to find new ways to scale its capacity without compromising quality.
The solution, developed in partnership with Capita, was DODG - called ‘dodge’ and standing for the digitisation of data gathering. Lovett explained his team approached the problem as an engineering challenge rather than just a process problem, and what began as a handful of robotic process automations in 2016 evolved by 2023 into a modular, integrated system that gathers, checks and validates data for caseworkers in the background, freeing them up to focus on complex, human-based decisions.
The results are significant. DODG has enabled HMRC to save the equivalent of more than 100 full-time staff, and compliance activities supported by the system now bring in an estimated £100 million to £1 billion each year. Lovett said the solution hasn’t replaced people but shifted them into more interesting, productive work.
That balance between automation and meaningful human work is crucial. As Capita’s Barry put it, effective automation “frees people up to do high-value work” and often improves the overall quality of services.
Real-world use cases
Several examples shared during the episode proved that AI and automation are already delivering measurable benefits across government. Barry described how Capita recently supported Transport for London in implementing a new toll discount system for residents and small businesses near the Silvertown and Blackwall tunnels.
Historically, such schemes would have required large teams to manually process applications. Instead, TfL deployed an automated solution that verifies documentation – such as proof of residence and identity – in just 11 seconds instead of four days. The system also incorporates advanced fraud detection, flagging manipulated documents that might previously have passed a human review process.
Meanwhile, DSIT’s Brown highlighted the work of i.AI in unlocking inaccessible government data. “We can use AI as a reading machine”, he said, highlighting a project called Extract, which processes handwritten and scanned planning documents to support the government’s housebuilding ambitions. “It gets that stuff into a state where it can be computed upon,” he said, transforming old paper records into usable, searchable data.
The next frontier: agentic AI
The conversation also explored agentic AI – systems capable of working towards goals independently, refining and adapting processes in real time. To Barry, the future of work will increasingly involve managing these intelligent agents and supported his vision highlighting Capita's own exploration of the technology. “We've done a scan of our own business and found 250 use cases where we could deploy this across a 40,000-person organisation,” he noted.
Brown offered a note of caution about the current state of agentic AI and highlighted two critical aspects of the technology. One is autonomous decision making, which he described as a really fraught and complicated area that has threads going out into public trust, accountability, quality of service, and responsibility. The other aspect focuses on decentralisation, where “you can have agents which tie [systems] together” more effectively, potentially enabling more flexible data sharing and integration across different platforms. He argued the near-term potential lies in the latter.
From experimentation to execution
The speakers were unanimous in their belief that successful AI implementation requires more than technology – it takes leadership, trust, and a willingness to test and iterate.
“Test your hypothesis with people who’ve done it before,” Barry advised, and suggested to start small, prioritise the biggest opportunities, and then focus on execution. Lovett agreed and added that the best way to overcome fears is to involve the users. “We want to know what our [staff] want to use,” he said, noting this approach ensures they feel engaged rather than threatened by the technology.
The key takeaway is that the future isn't about replacing humans but creating powerful partnerships between human creativity and technological capability. As Barry succinctly put it, the goal is to “free people up to do high-value work”. The emerging vision is of a more agile, efficient government powered by intelligent systems that augment rather than replace human potential.
Listen now to the full conversation on the Changemakers podcast