Putting residents back at the heart of Buckinghamshire Council's website

We have a huge amount of content so it's very difficult for residents to find everything they need. We prioritise the high volume, popular tasks on the homepage itself, but there's a long tail of things people might only ever need once in their life, and while those pages exist, we've never really had a good way of surfacing them beyond search.

We'd learnt a lot from the improvements we'd already made on our subsites, particularly around surfacing popular tasks, and we wanted to bring that thinking onto the main site where the majority of our residents land.

We'd also procured Zoom's AI virtual agent, and rather than tucking it away in the bottom right corner of the page, which is where webchat conventionally sits, we wanted to bring it right to the forefront, because it can take someone straight to the obscure page they need, however low the volume. Those two ambitions, a refreshed homepage and use of AI chat, became the focus of the project.

User research and usability testing shaped a lot of the details. Rather than simply checking whether people could complete a task, the sessions were designed to understand the why behind their behaviour, looking at what made residents feel confident they were in the right place, where they hesitated in the navigation, and what it would take for them to actually trust an AI chatbot over browsing for themselves. We tested across the homepage, the navigation menu, real waste and recycling scenarios, and the AI chatbot itself, so we could see how residents think and talk about their tasks, rather than just how our own service teams describe them.

That was particularly important when it came to the chatbot, where we tested whether people could find the 'Ask Us' button and whether it felt like the right place for it, and it became clear there's still hesitancy around AI, with people wanting to know clearly that they were talking to an AI rather than a person.

One of the more surprising findings came from search, where we'd initially tucked the search bar away inside a menu. People didn't necessarily use it, but they were uncomfortable not being able to see it, and the message was, essentially, "I'm probably not going to use it, but I want to know it's there." So we brought search back into a permanent, visible spot in the top right, while still giving the centre of the homepage over to the chatbot.

The services list was the toughest problem to crack. The old homepage had around 17 top level services with sub links underneath, roughly 69 links in total, and we got that down to single links per service area, 17 in total, but usability testing showed that was still overwhelming. We were told the number that would actually work, based on what fits comfortably on a page and within a reasonable number of scrolls on mobile, was nine. I wasn't sure we could get there, but the team took it as a personal challenge and found a way to consolidate those 17 service areas down to nine.

We also leaned heavily on GDS patterns for the navigation and information architecture, largely because GDS has the time and resource to do exhaustive user research that we simply don't, so it makes sense to follow their lead except where we're doing something different, like the AI chat, where the extra research investment pays off. In practice, that meant moving from a homepage and a single index page to a homepage, a first level index, and a second level index, so residents always have a clearer sense of where they are on the site. The result is that most of our roughly 3,000 pages are now reachable within three clicks.

The Zoom virtual agent was a separate procurement with its own dedicated team, but the ambition to lean into AI ran right through this project too. It's channel agnostic, so however a resident chooses to contact the council, by web chat or by phone, it now comes into the same virtual agent, which tries to resolve the query itself and hands it over to our contact centre if it can't.

Getting the two projects to work smoothly together took a bit of back and forth, since we'd assumed the virtual agent's floating chat icon would work in the same way as the buttons we'd built into the new homepage, but it turned out everything needed to run off the same trigger, so the team had to build a bespoke floating icon to match. It's seamless now, and I'm pleased with how it's turned out.

The launch went really smoothly. We ran staging and production in parallel for a couple of weeks so we could see what was happening before fully committing, and by far the biggest challenge on our side was migrating all the content from 17 service pages down to nine ahead of go live.

On the AI chat side, we saw an almost immediate uplift in use, around 30% at first, settling to a steady increase of 20 to 25% on where we were before, which is exactly the outcome we wanted, getting people using the tool and seeing that use keep climbing as more residents discover it and see that it works.

We're already applying the same ambition internally, using ServiceNow to build AI search across our internal systems, so staff can find what they need just as easily as residents can.

Now that the new templates and shorter page structures are live, there's a piece of work ahead on content itself, since the titles and structure need to reflect the new design, and we'll spend the next couple of months working through analytics to see where residents are still struggling.

The virtual agent reports back on what people are asking and where it can't find an answer, which is often because we simply haven't written that information down or have wrongly assumed it's obvious. A good example was on our contact us pages, where we'd typically list a phone number much lower than our online form. To a resident it was obvious what option we prefer them to choose first, but an AI will take a request to find a phone number literally, even when it isn't the best option to contact us. We're now working through our contact pages to make our preferred route more explicit rather than implied, as well as adding a dedicated AI chat button that services can place wherever makes sense for them. If AI is struggling to understand our content, it's a fair bet residents are too, so it's turning into a useful, ongoing discipline of keeping our content honest and current.