Specialist Insurer Automates 38% of
Service Volume with Agentforce

The Situation

A Lloyd's of London specialist insurer — providing professional indemnity, directors & officers, and cyber insurance — was experiencing a service operations challenge common to growth-stage insurers: a disproportionate volume of routine policy servicing enquiries (certificate requests, mid-term adjustments, coverage queries) was consuming the capacity of experienced account handlers who should have been focused on complex coverage structuring and renewal negotiations.

The insurer was running Salesforce Service Cloud for its broker and client service operations, with a substantial investment in Data Cloud for unified customer and policy data. The data infrastructure was in place. What was missing was the layer that could act on it without constant human intervention.

The leadership team had heard about Salesforce Agentforce but had concerns: the regulatory environment for insurance AI in the EU, the risk of agent errors on financial products, and the internal change management challenge of asking experienced underwriting support staff to work alongside AI agents. We were engaged to design and deliver the deployment addressing all three concerns simultaneously.

Scoping for Insurance-Specific Risk

The first four weeks of the engagement were not spent configuring Agentforce. They were spent defining precisely what agents were and were not permitted to do — and building the governance framework that would satisfy the insurer's Lloyd's regulatory obligations, their GDPR data processing requirements, and the EU AI Act's transparency and audit trail requirements for AI systems operating in financial services.

We defined three agent capability tiers: information retrieval (agents can retrieve and communicate policy data, document status, and coverage summaries — no action taken), guided resolution (agents can take pre-approved, rule-bound actions such as reissuing certificates, processing standard endorsements, and routing claims to the correct handler), and supervised escalation (agents handle initial triage and data gathering, then hand off to a human with a complete case summary and recommended action). Every agent action was mapped to one of these tiers, with corresponding audit trail, human oversight, and disclosure requirements.

This governance design took longer than the technology configuration that followed — but it was the work that allowed the insurer's compliance team to approve the deployment. Organisations that skip this step typically face post-deployment compliance remediation that costs multiples of what the governance design would have cost upfront.

Three Agents, Sixty Days

We deployed three agents in sequence over 60 days, each building on the data quality and integration work of the previous:

Agent 1 — Policy Enquiry Handler (day 1–20): The first agent was information-only — handling inbound queries about policy coverage, document status, and renewal dates by retrieving data from Data Cloud and Service Cloud. It disclosed its AI status to every broker and client interacting with it, logged every interaction for compliance audit, and escalated queries outside its permitted scope to human handlers within 30 seconds. Go-live was clean; the agent handled 62% of its initial interactions without escalation on day one.

Agent 2 — Certificate Issuance Agent (day 21–45): Certificate of insurance requests represent approximately 18% of the insurer's total service volume — high-frequency, highly standardised, zero-judgement-required. The second agent was configured to retrieve the relevant policy, validate the request against policy terms, generate the certificate document, and deliver it to the requester — with the full audit trail logged to Salesforce and the underwriting system of record. This agent eliminated approximately 280 hours per month of account handler time within its first two weeks of operation.

Agent 3 — Claims Triage Agent (day 46–60): The most complex agent handled first notification of loss interactions: gathering the structured data required for a claims handler to assess the notification (date of incident, coverage line, estimated quantum, relevant policy conditions), completing the Salesforce case record, and routing to the appropriate claims specialist with a pre-populated case summary. Claims handlers reported an average time saving of 22 minutes per new notification — time they reinvested in coverage analysis rather than data entry.

EU AI Act Compliance from Day One

The EU AI Act classifies AI systems used in financial services — including insurance servicing — as requiring specific documentation and transparency obligations. We implemented the full compliance framework concurrently with the agent deployment: mandatory AI disclosure in every agent interaction, complete interaction logging with 7-year retention aligned to the insurer's regulatory obligations, model cards for each agent documenting purpose, data inputs, action limits, and human oversight mechanisms, and a quarterly agent performance review process designed to satisfy supervisory requirements.

The compliance framework was not an afterthought or an add-on — it was built into the agent architecture from the first configuration session. The result was that the insurer's compliance team was able to sign off on each agent deployment in advance, without the programme delays that typically occur when compliance reviews happen after go-live.