Capita’s Richard Holroyd on AI in Defence and Beyond

It was the second day of a very busy DSEI 2025 when Richard Holroyd, Chief Executive of Public Services at Capita, eased into a chair opposite me. After a brief exchange about my Glasgow accent, the conversation turned quickly to Capita’s position in defence.

Capita is one of the UK’s largest outsourcing and services providers, employing tens of thousands of staff and holding contracts across defence, government, transport and financial services.

Its reputation has been shaped by high-profile public sector deals, including Army recruiting and Royal Navy training, as well as criticism over delivery and missed targets. More recently, it has been building up a reputation as a technology-enabled services business, using data and AI to reshape how large contracts are delivered.

I asked Holroyd directly how he wants Capita to be seen in that light.

“A lot of our customer base will think of old Capita, which was outsourcers. We’ve shifted ourselves radically. We’re not a tech company, but we are a tech capable company.” Partnerships, he stressed, are central to that shift. “We’ve partnered with the hyperscalers, so with ServiceNow, with Salesforce, with Amazon Web Services, and we have a fantastic reputation for business integration with SMEs and specialist companies.”

Navy training at scale, built around the learner

Project Selborne has been one of Capita’s most high-profile defence contracts. Training was previously fragmented across dozens of arrangements, and the company has faced no shortage of questions about whether consolidation would deliver real gains. Holroyd was quick to point to outcomes Capita claims.

The firm says it has merged 27 contracts into a single framework and now supports 80 percent of shore-based training for sailors and marines across 14 sites. According to its own figures, Ofsted has rated establishments as good or outstanding, more than 1,000 educators have been upskilled to deliver over 650 courses, and apprentice retention sits at 92 percent compared with a national average of 59 percent. Capita also highlights the Bridge Trainer programme, which it says has doubled navigation training capacity.

Holroyd steered away from numbers and charts to describe what he sees on the ground. “If you were to visit the Clyde, Faslane is just over the hill,” he said. “You would see the integration of Kongsberg virtual reality simulators that now allow the Submarine Training Centre to train bridge crews to navigate above the surface. We actually have a whole physical conning tower, panoramic surround virtual reality stuff that allows leading cutting edge training capability.”

The bigger claim lies in how lessons are delivered. “When I was a soldier, if the lesson took 45 minutes, you sat there for 45 minutes. Even if you got it and could pass the test after five minutes, you sat there for 45 minutes,” he said. “We’ve used AI to build a schedule optimisation tool. We identify the best way to put the building blocks of the courses together. Take the training needs analysis, apply technology to it, and understand how you shift people through the lessons faster without compromising on quality.”

Pressed for a concrete example, Holroyd cited the submarine qualification course. “There is a particular course around qualification to go on a submarine, called SNQ, Submarine Qualification, brackets Wet. We were able to take 20 percent out of the length of time of that course,” he said. “The Navy has this concept of jetty bulge, where you get people through basic training and then everybody is sitting on the jetty waiting to do the right courses while the fleet is under resourced. This reduces that.”

He added that the new model allows sailors to progress between postings rather than waiting for classroom slots. “One of the problems the Navy had was you had to wait till you were on shore to do stuff,” he said. “Now you can do stuff while you are at sea or while you are at home. When sailors come home, they can be with their families and still progress.” The cultural shift is as important as the technology. “Young people were coming out of school and college and going back to 1960s ways of learning with chalk and talk,” he said. “We have modernised and brought that up to standards, so people are learning in the same way they did at college, and actually seeing it progress even further.”

For Holroyd, the operational summary is straightforward. “Our mission is to get more sailors to the front line faster,” he said. The published statistics back that up to a point, though his emphasis was on how AI and data are used to change the order and tempo of training blocks rather than on any single technological fix.

Army recruiting, from paper to digital and back to people

Few contracts have brought Capita more criticism than Army recruiting. Delays of more than 200 days between application and enlistment were once the norm, and missed targets drew parliamentary attention. Holroyd acknowledged the scale of the challenge but argued that the process is now moving faster.

When Capita took over, the average journey from application to basic training was about 240 days. Holroyd said that figure has since fallen to around 150. One early step was reducing the mountain of paper medical records that slowed everything down. At one point the company was handling 12 metric tons of physical documents a year. Large language models now scan those records, flagging what clinicians need to look at and even reading doctors’ handwriting. Holroyd stressed that the final call always rests with a human.

Postal delays were another choke point. Records could sit in GP inboxes for months, so Capita pressed the NHS to send digital healthcare files instead. Holroyd said records that once took up to three months to arrive now appear in less than a week, sometimes within 24 hours. The company claims this change has boosted throughput by 30 percent and cut average wait times from 30 days to 8.

The wider funnel, he added, is now busier. Capita reported the highest application inflow in seven years at 132,000, with the strongest active candidate pipeline in three years at more than 10,000 in July. It also points to officer loading consistently at 95 to 100 percent in recent years, and junior entry averaging above 97 percent over the past seven years.

Holroyd was keen to stress that efficiency savings are not being used to cut staff. Instead, recruiters now have more time to keep candidates engaged, something Capita believes is vital given the high dropout rate early in training. One measure is an app that provides a structured fitness regime, designed to stop recruits injuring themselves or quitting after a few weeks. He also linked the reforms to generational expectations. In his view, today’s candidates, who live on their phones, see paper forms and long waits as an anachronism. “There are not any analogue kids anymore. They are all digital,” he remarked.

Policy detail often determines how modernisation works in practice. Holroyd offered the example of tattoos, which can still disqualify applicants if they are above the neck. Until recently an applicant with cosmetic eyebrow tattoos could be automatically rejected simply by ticking the wrong box on a form. That process, he said, is being redesigned. “We are implementing a recognition tool. You upload a photograph of the tattoo. It will understand where it is, what it looks like, and create a report that pushes you to look at the ones that are offensive or marginal. The human still stays in the loop.”

Beyond defence, the same pattern in public services

I asked how far Capita’s AI-driven approach extended beyond defence. Holroyd pointed to Transport for London, where residents applying for tunnel discounts once faced a 400-person call centre. “The original solution involved a 400 person call centre in order to process documentation,” he said. “We put together an AI solution that was able to verify the document. It could find the address, read it, cross reference it, verify it against other publicly available databases, confirm that 63 people with different names have not used that address already, and then process the discount.”

The company says the system now carries out 29 authenticity checks, from x-ray analysis to pixel-level error detection, and is better at spotting fraud than the human eye. According to Capita, transaction costs have fallen from about £2 per resident to nine pence, wait times from four days to 11 seconds, and staff requirements by around 75 percent. It estimates that the change saves TfL roughly £2.6 million a year.

Turning to welfare, Holroyd highlighted Capita’s role in Personal Independence Payment assessments. “We run about 50 percent of PIP,” he said. “AI based transcription services can listen to the entire conversation and then write that up in evidential quality. It saves a huge amount of time and we can progress significantly more assessments per day. The machine does not make the decision. The decision stays with the human.” He said the same pattern was being applied in student disability assessments for the Student Loans Company.

Local authorities are another test case. Holroyd described how a single system could handle multiple services in one step. “When a citizen calls in and says I am moving, cancel my parking permit, the system will identify that you are moving,” he explained. “It will surface that you are also a member of the local authority gym, you have a library card, you are registered at this service, and send you hyperlinks by text to make all the changes.” At Lambeth Council, Capita says this model recovered £3 million in historic council tax debt last year and increased payment commitments by £1.2 million, often without resorting to debt collectors.

Holroyd also pointed to Capita’s pensions business as an example of automation inside the company itself. By tracking keystrokes, he said, the firm identified where staff were spending time on repetitive cut-and-paste tasks. “You can now build an agent to do that. It goes to that screen, that database, puts it into the model, produces the letter. Now the agent checks that it is all in the right place and that the numbers are right.” His view of AI remained pragmatic: “AI is brilliant at doing any task that would take about 15 minutes. It creates more time for beneficial things.”

Where AI sits, and why the boundary matters

AI is the thread running through much of Capita’s defence and public service work. But it is also the area that attracts the most scepticism. In recruiting and training, the risks of error are obvious: a misread medical record could wrongly exclude a candidate, while over-reliance on automation could weaken safety checks. I asked Holroyd directly whether AI in these processes could really be considered safe and reliable. He pushed back firmly. “It is perfectly safe because it is not making the decision,” he said. “The human has to make that decision. What AI can do is take away and summarise.”

That principle, he argued, holds whether it is a scan of a medical record or a photograph of a tattoo. “We do not get the core data. We get a scan of the medical record that we can read. It arrives digitally instead of through the post,” he said. “You upload a photograph of your tattoo. You are not giving us the data to control. You are giving us, of your own volition, a photograph of a tattoo.”

Holroyd also acknowledged the limits. “We are working in the process layer, not in the data layer. Real drive in AI comes when you can access it in the data layer, because once you are in the data layer you can pull all sorts of data out and use it to speed processes. That is very carefully guarded. Questions about AI and data are key questions for government that have not yet been resolved.”

Compliance was another point I pressed. Here too he drew a hard line. “Anything related to compliance will always require a human in the loop,” he said. “At the moment, we know Lisa processed that and Ben took the decision. AI will take away a whole load of the work that allows Lisa to spend more time and Ben to spend more time. You get better quality decisions and more of the backlog moved faster.”

Contracts, outcomes and speed

Discussion of AI quickly led to the question of contracts. Defence readers will know that the Army and Navy are not short of experience with long-term outsourcing deals, many of which have been criticised for locking in outdated technology. I asked Holroyd how that reality squares with his pitch for faster digital change.

“Moore’s law is a thing of the past,” he replied. “AI can now move so fast that if you give me a six year contract and specify the technology on day one, it will be out of date within a year. Give me an outcome based contract that says we need you to run this process, and then we can use current technology to improve it and keep improving as we go.” He conceded that this is difficult within current commercial frameworks.

Holroyd drew a contrast between defence and some civilian departments. The Royal Navy and Army are already demanding quicker results: more soldiers trained, more sailors available, less time wasted on repetitive tasks.

By comparison, he argued, parts of government still specify inputs rather than outcomes. One example he cited was a department with a fraud problem worth billions, where the tender was designed around call centre staff and headsets rather than analytical tools. The point he wanted to drive home was that AI’s value depends not just on the technology itself, but on whether contracts allow organisations to update processes at speed. For Capita, that argument underpins both its defence and public sector work.

Training the workforce to use the tools

If AI is to deliver at scale, staff need to know how to use it. Holroyd said Capita has built training modules for its own employees to integrate the technology into daily work. One example he offered was Microsoft Copilot. While on holiday in Corfu, he asked it to summarise the meetings he had missed.

Because his team had used transcription services, the system produced a list of decisions and linked to the underlying presentations. The point, he said, was not the novelty of the tool but the productivity gain: fewer hours spent catching up, more time on useful work.

The same pattern is visible inside Capita’s pensions business. By monitoring keystrokes, the firm identified how much time staff were spending on repetitive cut-and-paste tasks. Automation now performs those steps, while employees check that the information is correct. Holroyd presented this as a microcosm of Capita’s wider philosophy: AI handling the routine tasks so that humans can focus on judgment and quality control.

The close

The interview finished where much of it had circled back to: the promise that technology can speed processes without displacing the human role. Holroyd ended with an example from training. “Imagine in training, we have managed to get more sailors to the front line faster,” he said. “There are more sailors, better trained, better equipped to resource our ships. That is because they are spending less time doing less valuable tasks.”

The same argument runs across Capita’s portfolio. The company says Army recruiting timelines have been cut, Navy training capacity and retention improved, transport discounts in London cleared in seconds rather than days, and local councils recovering millions in unpaid tax.

Holroyd was careful to frame AI as a support, not a substitute. “AI is not here to replace the human,” he said. “It is here to free them up for what matters.” He also insisted that Capita itself has shifted, distancing the present company from its past reputation. “We’ve shifted ourselves radically. We’re not a tech company, but we are a tech capable company.”

That claim to reinvention is now central to Capita’s pitch. For defence, the test will be whether these efficiencies and new methods translate into the hard outcomes ministers and the services demand.