Thames Valley Police's deployment of Bobbi, a Salesforce Agentforce-powered AI agent, demonstrates that agentic AI has moved decisively beyond pilot territory into operational environments where the stakes are genuinely high. Handling 200 conversations daily with a 45 percent full-resolution rate across 14,000 annual citizen contacts, Bobbi has freed 3,266 operational hours in six months—a figure that translates to meaningful capacity reallocation rather than headcount reduction. The implementation reveals a critical insight for CX teams: the real value of agentic AI lies not in contact deflection alone, but in creating new pathways for vulnerable populations to access services they previously avoided. The forces discovered within days of launch that citizens unable or unwilling to make phone calls—particularly those experiencing domestic abuse or sexual violence—would engage with a text-based agent to "test the water" before escalating to human handlers. This reframes the agent's role from cost-centre to safeguarding tool, raising an important question for teams evaluating Agentforce or similar platforms: are you measuring success purely on automation rates, or on the quality and safety of interactions your agent enables?
The guardrails Thames Valley implemented—restricting Bobbi to 91 verified internal knowledge sources, capping responses at 350 words, and maintaining continuous human oversight—reflect a maturity in agentic deployment that many CX teams have yet to adopt. Rather than rushing to launch, the forces spent three to four weeks implementing and the remainder testing, a ratio that inverts typical project timelines. This approach directly contradicts the speed-first mentality that dominates much AI adoption in contact centres. For Zendesk administrators and support leads considering similar deployments, the implication is stark: governance and observability are not post-launch luxuries but pre-launch necessities. The fact that Bobbi identifies at least one high-harm offence daily and routes two violence-against-women cases for human intervention on average demonstrates that agents operating in sensitive domains require different calibration than those handling billing queries. What this means for teams already running Agentforce or planning implementations is that the competitive advantage will accrue not to those who automate fastest, but to those who build the most trustworthy escalation pathways and maintain the clearest human-AI boundaries.
The reinvestment of freed capacity into higher-value crime and safeguarding work, rather than cost reduction, signals a philosophical shift in how organisations should approach agentic AI. Thames Valley's explicit rejection of using Bobbi to "save money" in favour of service quality improvement suggests that CX leaders should reframe agent ROI conversations with stakeholders away from FTE reduction and toward impact multiplication. The 4.6 out of 5 citizen satisfaction rating and multilingual capability indicate that agents can simultaneously reduce friction and expand access—a dual outcome that traditional automation rarely achieves. For contact centre leaders, this case study provides both a template and a challenge: the technology is proven, the guardrails are documented, and the business case is clear, but success requires treating agentic AI as a service design problem rather than a labour arbitrage opportunity.
Six months after its launch, Thames Valley Police and Hampshire & Isle of Wight Constabulary say that Bobbi, the UK’s first police AI agent, is changing how communities access policing and how contact center teams manage non-emergency demand. Built on Salesforce Agentforce, Bobbi was introduced