CSRs Needed Instant Policy Answers. Legacy Tools Delivered Search Friction.
A top-three life insurance carrier's customer service organization handles thousands of daily policyholder inquiries: life insurance policy servicing, beneficiary change requests, policy loan eligibility, coverage term questions, premium payment options. Every question requires accurate answers, compliant with state insurance regulations and New York State Department of Financial Services (NYDFS) oversight.
The carrier's proprietary insurance knowledge corpus was large and complex — including policy documents, underwriting guidelines, and state-specific regulatory requirements. This made it difficult for junior CSRs to find answers as quickly as experienced staff. The result: longer handle times, higher escalation rates, inconsistent service quality.
The organization had already deployed an AI-powered chat assistant, but its underlying host service lacked the managed LLM inference capabilities, retrieval‑augmented generation (RAG) pipeline architecture, and VPC-native security controls required to ingest and reason over a large proprietary insurance knowledge base in a regulated environment.
With no cost‑effective alternative comparable to AWS Bedrock for secure, private LLM inference, the carrier engaged Perficient to migrate and re‑architect the solution on AWS.
The Right Information. At the Exact Moment It's Needed.
We migrated the existing assistant to AWS, re-architected it with AWS Bedrock at its core, and integrated it directly into the CSR workflow, enabling real-time access to the carrier's knowledge corpus without system switching.
What the solution delivers:
- A RAG pipeline built on Amazon OpenSearch Vector DB, enabling semantic retrieval across policy documents, underwriting guidelines, and state-specific regulatory content
- Amazon Titan Multimodal Embeddings to vectorize the carrier’s proprietary insurance corpus for accurate retrieval at inference time
- AWS Bedrock for managed LLM inference with prompt guardrails that ensure responses are grounded exclusively in verified insurance content — not general model knowledge
- VPC Endpoints ensure all policyholder data and insurance knowledge content stays within the carrier's private network
- Deployment on Amazon EKS across two Availability Zones, integrated with Ping Identity SSO
- Full alignment with NYDFS data governance and security requirements
This architecture allows CSRs to receive immediate, compliant answers during live customer interactions — reducing search friction while maintaining regulatory integrity.
110,000 Questions. 13% Faster. Junior Reps Performing Like Senior Staff.
In the first three months of production use, the assistant answered more than 110,000 insurance policy questions across the CSR organization, covering policy servicing, beneficiary inquiries, loan eligibility, and coverage questions. Average handle and hold times on policy servicing calls dropped 13%. Through real-time access to the carrier's insurance knowledge corpus, junior CSRs now resolve complex policy questions on pace with senior staff.
CSR satisfaction hit 87%, with users citing faster and more accurate retrieval of insurance policy answers as the primary productivity driver. The platform scaled from a 50-user pilot to 1,100+ CSRs in enterprise-wide deployment — demonstrating production-grade GenAI adoption at scale within a regulated insurance institution.
That’s the difference between searching for answers and delivering them instantly.
AWS Premier Tier Services Partner
We migrate and modernize workloads that perform under pressure and extract intelligence from your data to drive real business decisions.