At specialist insurer Hiscox, useful AI ideas are coming from the people who know the work first-hand. The company has trained colleagues to experiment with Copilot and build AI agents for everyday tasks, while strengthening oversight as their use grows.
Hiscox believes that the people who understand a business problem best are usually those dealing with it every day.
They know where information is hard to find, which repetitive processes cause delays, and which manual tasks stop them getting to more valuable work.
Microsoft Copilot and Copilot Studio is giving them the means to find solutions for these problems.
The company calls this approach “Citizen AI”.
“Citizen AI at Hiscox is all about putting the tools in the hands of the people who do these things day by day,” says James Jackson, Data Culture and AI Lead at Hiscox.
“This is about underwriters, claims professionals, finance and HR people doing their day job better by having AI augment what they do.”
Small agents, measurable gains
The approach has been highly successful. Hiscox employs around 3,500 people, who have collectively built more than 6,000 agents that don’t just assist with tasks; they complete them.
For example, Paul Lawrence, Chief Underwriting Officer for Hiscox London Market developed a ‘Tone of Voice’ agent to help prepare two board reports, saving six or seven hours, by his estimate. The agent ensures that language is on-brand and consistent in style and tone. He spent the time saved doing more valuable work with brokers and underwriting teams instead.
Jackson says the idea has travelled: versions of the agent are now used by other senior executives and their assistants to assist with preparing board presentations, material for external meetings, and everyday communications.
Another colleague, Jack Gargrave, who works as a trading manager, built his own agent workflow to automate a series of tasks that previously took around three hours. With the agent in place, the same work can now be completed in about 20 minutes.
Across the organisation, colleagues report saving an average of two to five hours a week, with some saving much more. Jackson estimates that productive time returned to the business and costs avoided have already passed seven figures.
Small fix, big impact
The approach has shown that a useful agent built by one colleague may reveal a problem shared across the business.
If several people build agents for the same task, Jackson’s team treats that as evidence of a wider need. It can then work with data science, technology, cyber and risk specialists to develop the idea into a more refined and powerful, shared tool.
Hiscox has already seen this with agents designed to extract contact information from emails. What began as several individual solutions is now developing into a larger business project.
“My job is to make it make sense for the entire business and make sure it gets the appropriate level of support,” Jackson says.
Adding guardrails
While rapid adoption of agent building can boost creativity and productivity, it also comes with challenges. Duplication is one; observability and governance is another. An organisation needs to know that the agents its people build are using proprietary data safely and securely, say, and that confidential data can only be accessed by employees with the right level of clearance.
Hiscox already uses its responsible AI process, called RAISE, to assess higher-risk uses of AI, and is now refining it so that scrutiny increases with the level of risk.
And Hiscox is now also introducing Microsoft Agent 365, which acts like an IT control centre for AI agents, helping organisations see, manage and secure all the AI assistants and automated agents being used across the business.
Jackson says it will help the company see what’s been built, identify duplication, and flag agents that potentially pose a risk to the business.
He sees this oversight as a critical part of making employee-led innovation sustainable, an approach he says won’t change for the foreseeable future.
“Let people build the capability themselves, with good guardrails and good control mechanisms,” he says. “Let the people lead on the work they know needs doing.”
Start with the work
Hiscox first built AI literacy across the organisation with Copilot, starting small and asking people to identify the irritations in their own jobs: the repeated tasks, frustrating searches or processes that take longer than they should. Then seeing what parts Copilot could complete more efficiently.
The most useful ideas aren’t always complicated. James Welford, a Learning and Development Manager, gives a good example: “At the start of a week, I ask Copilot to look through my team’s messages and Outlook emails, just to create a to-do list of things that I need to get completed that week. It helps me stay on top of everything and helps me make sure I don’t miss any of those smaller bits that might have fallen through the cracks.”
Paola Hristova, Business Readiness Lead in the UK, uses Copilot to challenge assumptions and test an approach from different angles. She describes it as “almost like a sounding board”, helping her consider governance and risk.
From access to embrace
Access to tools alone didn’t change how people worked, or spark the fierce adoption of agentic workflows. Earlier this year, Hiscox had around 200 active Copilot users. Six months later, usage was reaching about 3,200 people on busy days, or from 4% adoption to 74%.
The gap was closed through regular training and visible support from leaders. Hiscox has trained more than 3,000 colleagues. The company has also run executive workshops, employee events and a continuing programme focused on practical skills, from writing better prompts to using AI in meetings and building agents.
A former teacher, Jackson treated training as a crucial part of the roll-out rather than an optional extra.
“One of the things people worry about is that they won’t understand how to use the tools,” he says. “We have put on thousands of hours of training and built a dedicated AI hub with resources that are updated every week.”
The bottom-up approach ensures people become their own AI taskforce, Jackson believes.
“People know what’s broken in their job,” he says. “They can use AI to give themselves that time back, so they can do more of the job they signed up to do in the first place.”