The challenge

Centrepoint's frontline support workers, the people working directly with homeless young people every day, were spending around 70% of their time on admin. Transcribing casework interviews, writing up case notes, answering emails. Only 30% of their time was going to the young people they were there to help.

For an organisation where two thirds of staff work in frontline services, and where Centrepoint's own data showed that direct time with young people was the single biggest driver of positive outcomes, efficiency had a direct impact on achieving their mission.

The solution

Rather than launching a large-scale AI strategy, Centrepoint worked with us to start small and start with people. Before any technology was considered, the team ran in-depth interviews with staff across the organisation to understand where time was really being lost and what people would do if they got it back.

From those conversations, over 150 potential use cases were identified, consolidated into seven pilots spanning frontline services, partnerships, volunteering, operations and innovation. Every pilot was staff-led. We provided support and playbooks, but the teams owned the process themselves.

For frontline caseworkers, an AI transcription and case note tool automated the most time-consuming admin. A custom co-pilot agent answered the 20-30 routine staff queries a week that were clogging up the operations team's inbox. Research pilots gave the partnerships and volunteering teams AI-powered tools to replace hours of manual desk research, with better results than before.

The impact

Frontline caseworkers saved an average of five to six hours per week, with some saving an entire day. Rolled out across Centrepoint's 250-plus caseworkers, that would be the equivalent of 40 full-time roles of time freed every year. That time went straight back to young people, more direct contact, more planned activities, more of the work staff had signed up to do.

Research pilots saved two to four hours per week, with quality improving alongside efficiency. Case notes became more complete and consistent. The operations chatbot, built in two hours, eliminated a stream of routine queries from an already-stretched team.

Across the pilots, ROI was significantly positive, with tool costs ranging from £15 to £55 per person per month delivering returns that comfortably justified the investment at pilot scale. For the research pilots in particular, where workflow changes are minimal and tools are trainable in a couple of hours, scaling across the organisation was an easy decision.

One caseworker's response captured what it meant in practice: "You simply cannot take this away from us."