Insurance Service Agent

Overview

The Insurance Service Agent automates the four highest-volume workflows in property & casualty insurance using Salesforce Agentforce (ExternalCopilot, EinsteinServiceAgent planner). Built end-to-end in a single session — custom data model, Apex business logic, Flow wrappers, Agent Script, test data, deployment, and iterative testing — achieving 29/29 Apex tests passing on the first deploy and 5/5 conversation scenarios within two iterations (v1→v2). Three custom objects with 42 total fields model insurance-specific relationships across Policy, Claims, and COI Requests.

Problem

Insurance agencies handle thousands of repetitive policyholder requests daily. Certificate of Insurance (COI) requests are the #1 operational bottleneck in commercial insurance — each requires manual lookup, verification, and document generation with 2-5 business day turnaround; agencies lose clients over slow COI fulfillment. FNOL calls require an agent to manually record every detail under time pressure with frequent data entry errors. Policy inquiries (deductibles, premiums, coverage details) account for 40-60% of call center volume but require zero judgment to answer. And renewal management suffers because agents are overwhelmed by the above three — leading to lapsed policies and missed cross-sell opportunities.

Approach

Hub-and-spoke architecture: a central topic router with forceful instructions ('Your ONLY job is to route. NEVER answer yourself.') prevents the LLM from attempting to help without executing the proper action first. Priority-ordered routing with first-match-wins: Claims keywords always route first regardless of other terms in the message (a policyholder mentioning their coverage limit during an accident report routes to Claims, not Policy Inquiry). (1) FNOL/Claims Intake — two deterministic classification algorithms: priority tiers (Critical: total loss/fatality/structural collapse, High: accident/flood/theft, Medium: default, Low: cosmetic damage) and adjuster team assignment by claim category (Auto, Property, Liability, Workers Comp, Homeowners, General). Creates Claim__c record, queries back the auto-number. (2) COI Request — creates COI_Request__c with Status=Submitted, Date_Requested=today, reference number, and realistic turnaround timeline. (3) Renewal & Cross-Sell — queries ALL active policies for the policyholder (not just one), finds the soonest expiring policy, checks the 60-day renewal window, then runs portfolio-level coverage gap analysis: Auto without Umbrella → suggests Umbrella; Property without GL → suggests GL; Homeowners without Umbrella → suggests Umbrella; Workers Comp without GL or Property → suggests both. Returns renewal eligibility, days until expiry, gap analysis, and cross-sell recommendations in a single response. (4) Policy Inquiry — queries all 16 Policy__c fields including coverage limits, deductibles, premiums, payment schedule, effective/expiration dates, property/vehicle info, and assigned agent. Auto-lookup: Policy Inquiry and Renewal call actions immediately on topic entry (policyholder_email already bound from verification) — one-turn resolution, no questions asked. Hybrid reasoning: LLM interprets natural language ('I rear-ended someone' → claims scenario), all classification, adjuster assignment, and portfolio gap analysis are fully deterministic — no hallucination possible on critical business logic. Flow wrapper pattern and savepoint/rollback on DML operations consistent with PM Tenant architecture.

Tech Stack

Results

29/29 Apex tests passing on first deploy across 5 Apex classes. 5/5 conversation scenarios passing within two Agent Script iterations (v1→v2). Four P&C workflows fully automated: FNOL intake with deterministic adjuster assignment, COI request tracking with reference numbers and turnaround SLA, renewal management with cross-sell recommendations, and 16-field policy inquiry. Priority-ordered routing ensures claims are never misrouted even when messages contain overlapping keywords. 42 custom fields across 3 objects (Policy__c, Claim__c, COI_Request__c) with auto-number identifiers (POL-0001, CLM-0001, COI-0001). Ships with 3 pre-configured demo personas: Sarah Mitchell (Auto + GL — renewal-eligible, COI demo), Robert Johnson (Homeowners + Auto — claims history, policy lookup), Emily Chen (Workers Comp only — coverage gap analysis surfaces missing GL + Property with cross-sell recommendations). Sample end-to-end flows: 'I was in a car accident on Highway 101' → Claims routed, Priority=High, assigned to Auto Claims Team. 'I need a COI for Titan Corp' → COI_Request created, Status=Submitted, 1-2 day SLA. 'What coverage gaps do I have?' (Emily) → Workers Comp only detected → GL + Property both recommended. Designed for COVU and NYC insurtech prospects — directly addresses the #1 insurance operational bottleneck: COI turnaround from 2-5 days to instant self-service submission.