Cómo las fuerzas policiales del Reino Unido crearon el agente de IA Bobbi para resolver el 75 % de las llamadas no urgentes de forma autónoma

Thames Valley Police and Hampshire & Isle of Wight Constabulary serve a combined population of over 4 million people across 3,800 square miles of England, handling everything from neighbourhood crime to major investigations. Together they receive over 400,000 non-emergency calls per year, and the data revealed a striking pattern: nearly half those calls were from victims and witnesses simply asking for a case update — people who had already reported an incident and needed to know what was happening.

This created a compounding problem. Citizens waited on hold to ask a simple question, while operators who could be handling more urgent matters were fielding case-status queries. The forces needed to give the public a faster, self-serve way to access information they were entitled to — without compromising the security standards required for sensitive police data.

TVP and HIOWC built a secure citizen portal as a digital front door for policing, using MuleSoft to integrate data from across the forces' backend systems and Salesforce Experience Cloud to deliver a structured, authenticated interface. Citizens can now log in, view their case, send messages directly to the officer handling their matter, and receive automated status updates at every step of an investigation — without picking up the phone.

On top of the portal, the forces deployed Bobbi, an AI agent built with Salesforce Agentforce to handle inbound non-emergency conversations. Bobbi handles over 200 conversations per day and resolves 75% of them without escalating to a human operator — detecting when a case falls into a high-risk category and routing those to a human immediately, while autonomously addressing the majority of status enquiries and routine guidance requests.

The financial impact is pronounced given the public-sector context: £1.4 million in cost avoidance from self-service case updates, and a 97% reduction in cost per interaction compared to the previous phone-only model. Call volumes dropped 10% as citizens shifted to the digital channel. For policing, where every officer-minute has a direct opportunity cost, these numbers translate directly into capacity for frontline work.