GovernmentCustomer Service

How UK Police Forces Built AI Agent Bobbi to Resolve 75% of Non-Emergency Calls Autonomously

Thames Valley Police and Hampshire & Isle of Wight Constabulary (TVP and HIOWC) are two of England's largest police forces, together serving over 4 million people across 3,800 square miles. Receiving more than 400,000 non-emergency calls per year — nearly half from citizens simply seeking case updates — the forces built a digital front door for policing: a secure citizen portal powered by MuleSoft and Agentforce Experience, and an AI agent named Bobbi that handles 200+ non-emergency conversations per day, resolving 75% autonomously. The result is £1.4M in cost avoidance and a 97% reduction in cost per interaction.

Outcomes

£1.4MCost avoidance from self-service case updates
97%Reduction in cost per interaction
200+Non-emergency conversations handled by Bobbi per day
75%Autonomous resolution rate for Bobbi
10%Reduction in call volumes

Tools & Technologies

1M
MuleSoft
Integration platform for connecting applications, data, and APIs across cloud and on-premise systems.
2SE
Salesforce Experience Cloud
Salesforce platform for building branded digital experiences, portals, and communities connected to CRM data.
3SA
Salesforce Agentforce
Platform deploying autonomous AI agents for customer service, sales, and employee tasks across Salesforce.

AI Categories

Challenge

Nearly half of 400,000+ annual non-emergency calls to TVP and HIOWC were citizens asking for case status updates — tying up operators who could be handling higher-priority matters, while victims waited on hold for information they were entitled to receive.

Solution

TVP and HIOWC built a MuleSoft-integrated secure citizen portal via Salesforce Experience Cloud for self-serve case tracking and officer messaging, then layered Agentforce to power AI agent Bobbi, which handles 200+ non-emergency conversations per day and autonomously resolves 75% without escalation.

Full Story

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.

Access 430+ AI use cases, 415+ tools, and adoption signal rankings.

Source

SALESFORCE
June 2026
Original case study

Similar Cases

1E
How Engine Uses Salesforce Agentforce to Cut Handle Time 15% and Save $2M
Engine
15%Average handle time reduction
2C
How CACI's DarkBlue Uses Elasticsearch and Claude to Accelerate Dark Web Criminal Investigations
CACI
Seconds per query regardless of data age or volumeCriminal investigation acceleration
3GF
How Grupo Falabella Uses Agentforce on WhatsApp to Resolve 60% of Requests
Grupo Falabella
60%Service requests resolved autonomously on WhatsApp
4NM
How New Mexico County Uses C3 AI to Boost Property Appraisal Accuracy by 50%
New Mexico County Assessor
+50 percentage pointsImprovement in model accuracy
5M
How ManTech Cut IT Tier 1 Support Workload by 50% with Moveworks AI
ManTech
50%IT Tier 1 FTE reduction
6O
How OpenTable Uses Agentforce to Resolve 70% of Customer Inquiries
OpenTable
70%Diner and restaurant inquiries resolved autonomously
7LC
How LY Corporation Uses Agentforce to Resolve 80% of Customer Cases
LY Corporation
80%Customer cases resolved by Agentforce
8R
How Reddit Uses Agentforce to Deflect 46% of Advertiser Support Cases
Reddit
46%Support cases deflected by AI
9NH
How NYC Health + Hospitals Uses Snowflake to Cut Data Delivery from Days to Minutes
NYC Health + Hospitals
5 days → 5 minutesData delivery time for membership and claims data
10ME
How Massachusetts Education Office Saves $1.5M Annually with Snowflake
Massachusetts Executive Office of Education
$1.5MAnnual cost savings from Oracle to Snowflake migration
See all use cases →