Who is RxBenefits?
RxBenefits is the employee benefits industry’s first and only technology-enabled Pharmacy Benefits Optimizer (PBO). Founded in 1995 and headquartered in Birmingham, Alabama, RxBenefits serves over 3.9 million members and has processed more than 117 million prescription drug claims. With over 700 pharmacy pricing, data, and clinical experts, RxBenefits works independently of PBMs and carriers to deliver prescription benefit savings to employee benefit consultants and their self-funded employer clients.
At the core of RxBenefits’ operations is a member-first approach: in-house call centers staffed with representatives empowered to resolve member issues — from prescription coverage questions to co-pay inquiries — as well as handling pre-authorization requests from healthcare providers.
The Challenge: Scaling Member Support Without Scaling Costs
RxBenefits’ member support operations handle five-digit weekly call volumes. Members call about prescription coverage, co-pays, and formulary questions. Healthcare providers submit pre-authorization requests. Each interaction demands accuracy, compliance, and speed.
But the operational reality was costly. High call volumes drove significant staffing needs, while the repetitive nature of many inquiries contributed to elevated employee turnover — a persistent and expensive challenge in customer service operations. Every departing employee meant weeks of onboarding and training before a new hire could handle calls independently.
At the same time, RxBenefits’ technology landscape included legacy systems that required agents to navigate multiple, often redundant platforms to retrieve member information — slowing resolution times and adding friction to every interaction.
RxBenefits wanted to use AI to achieve two goals: increase member satisfaction through faster resolution times, and reduce the operational burden of repetitive, administrative work on their employees — within a highly secure, compliant healthcare infrastructure.
Our Approach: Understanding Before Building
Before writing a single line of code, we invested significant effort into understanding the problem space. We analyzed thousands of call transcripts to map the operational landscape:
- Why are members calling? What are the most frequent issue types?
- Which call types drive the longest handle times?
- Where do agents spend the most time on administrative tasks vs. actual member support?
This analysis revealed that a small subset of issue types accounted for a disproportionate share of total call time — and that much of the agent’s effort was spent retrieving data from disparate systems rather than solving member problems. This rigorous problem definition shaped every design decision that followed.
The Solution: Three AI Workflows
We partnered with RxBenefits to build three AI-powered solutions, each targeting a distinct part of the member support operation.
Integrated Voice Assistant (Member-Facing)
An AI voice agent that handles select inbound use cases — such as member verification — directly. In many cases, the member's issue is fully resolved without ever reaching a human agent, enabling meaningful call deflection while maintaining a high-quality member experience.
Agent Assist Toolkit (Employee-Facing)
A suite of AI agents that support human agents during live calls: pulling data from legacy systems in real time to eliminate manual cross-referencing, pre-populating faxes and emails with AI-generated content for one-click send, and reducing administrative overhead so agents can focus on the member.
Call Intelligence & Tracking
AI-driven monitoring that tracks performance metrics across the full call operation in real time. Beyond dashboards, the system allows managers and ops teams to interrogate actual call transcripts in natural language — asking questions like "what are the most common reasons members are escalating?" or "which agent workflows are taking longest?" — without writing a single query. This turns every call into a source of operational intelligence, and closes the loop between deployment and continuous improvement.
Impact at a Glance
Looking Under the Hood
The voice assistant processes inbound calls, classifies intent, and either resolves the inquiry autonomously or routes to a human agent with full context. The agent assist toolkit runs alongside the agent’s workflow, surfacing relevant data and pre-populating communication templates in real time.
Real-time audio pipeline via gRPC and Kinesis. Five9 VoiceStream pushes raw audio over gRPC into an ECS Fargate proxy, which feeds directly into AWS Kinesis Data Streams. This low-latency path enables real-time transcription — the system is processing and interpreting speech as the call is happening, not after it ends.
Three-pronged RAG with LLM-based retrieval routing. Rather than a single vector search, the system uses Amazon Bedrock to classify each query at runtime and route it to the right retrieval method: purpose-built APIs for live data lookups, SQL against a structured Postgres layer for aggregation queries (e.g. claims counts, eligibility status), and Amazon Kendra for unstructured document search across internal knowledge sources including Confluence, SharePoint, and historical call data. The LLM decides which path — or combination — fits the question.
Per-agent real-time delivery with human-in-the-loop validation. AI-generated guidance is delivered to each agent’s interface via AWS AppSync GraphQL subscriptions scoped by agent ID — so every agent only sees what’s relevant to their active call. Agents review and confirm suggestions before acting, keeping humans in control while dramatically cutting the time spent manually retrieving information.
“We evaluated several approaches to introducing AI into our operations. What set nexamind apart was their insistence on understanding our problem deeply before proposing solutions. The data analysis phase alone changed how we think about our call operations — and the AI agents they built are now a core part of how we serve our members.”
Facing similar challenges in your call center?
We'd love to understand your situation before proposing anything. Tell us what you're working on.