The AI assistant ‘Bobbi’, built on Salesforce's Agentforce Public Sector platform, was rolled out across Thames Valley Police and Hampshire & Isle of Wight Constabulary in November 2025. Since that time, the forces estimate it has saved 3,290 operational hours, handling around 200 non-emergency conversations a day, and freeing the equivalent of 2.45 full-time roles for front-line policing. Customer satisfaction with Bobbi currently stands at 4.6 out of 5.
Extrapolated across a full year at the same rate, that would put annual hours saved at roughly 5,260, though this is an estimate based on partial-year data rather than a figure confirmed by Salesforce, and actual savings could shift as usage patterns mature. Bobbi is also adding value by surfacing high-harm cases and improving access to public services for citizens who might not have come forward through traditional channels. In fact, this may be an even more crucial contribution than its timesaving and resolution rates.
The headline metric has shifted
When Bobbi launched, Salesforce's original announcement reported it was resolving 82% of inbound enquiries without escalation in its first week. The latest figures put that figure at around 45%. Rather than pointing to weaker performance, however, this likely reflects wider adoption and more complex conversations reaching the assistant, including safeguarding and domestic abuse cases where successful escalation, not containment, is the better measure of success.
The problem behind the numbers
Before Bobbi existed, nearly half of all non-emergency calls to the two forces were not new crime reports at all. They were citizens chasing updates on cases already logged, amounting to roughly 400,000 enquiries a year.
Callers could wait up to 24 minutes for an answer that staff often could not give, since they lacked immediate access to the relevant case file. This was less a technology issue than a service-delivery problem due to volume, accessibility and limited resources.
What the assistant actually does
Bobbi works as a “digital front door” for non-emergency contact, grounded in 91 verified knowledge articles covering everything from parking offences to domestic abuse support. It cannot access the open internet, so every answer draws only from approved material.
When a conversation touches something sensitive or high-risk, the assistant hands off to a human operator rather than attempting to resolve it itself, an approach that mirrors the wider debate over where chatbots should give way to people.
Bobbi also supports multiple languages, widening access for communities that may face additional barriers to contacting police. Salesforce frames this as a deliberate design choice favouring governance and escalation over full automation, echoing how building proper oversight into AI systems has become a precondition for trust rather than an afterthought.
The outcome that matters most
Among these outcomes, the most significant may be what Bobbi catches rather than what it resolves. At least one high-harm offence is identified and escalated by Bobbi every day, and on average two cases of violence against women and girls are routed to human intervention daily. The CRM vendor also points to citizens who may never have approached police through traditional channels, including a teenager who used the assistant to seek help during a domestic incident when calling emergency services was not an option.
Lessons from Bobbi
Bobbi shows how AI agents in the public sector can deliver on efficiency while also creating real safeguarding value. The hours saved and roles freed up are significant, but the clearest evidence of success may be the technology's ability to widen access to support, reach vulnerable citizens, and direct human attention to where it is needed most. A higher automation rate does not necessarily mean a better outcome. More broadly, it is a reminder that the value of CX AI can be easy to miss if you are looking at the wrong metrics.