Presentation (CTI)

OBJECTIVES: Patient journey is a vital exercise across the product lifecycle, used to identify unmet needs, gaps in care, and to define and size target populations. We had previously developed an optimized process, automating it to the extent possible with pre-agentic AI technologies. The workflow, however, still required substantial input from multiple teams, with hand-offs across time zones. To address these limitations, we developed a distributed multi-agent AI system in which autonomous agents collaborate through a robust agent-to-agent protocol, working alongside human team members to deliver the end-to-end workflow.
METHODS: Our architecture consists of autonomous AI agents, each with dedicated access to Model Context Protocol (MCP) servers and specialized toolsets. Each agent employs an intelligent planner to sequence tasks through MCP, while an agent‑to‑agent (A2A) communication layer facilitates seamless collaboration. Role‑specific agents handle workflow segments: extracting scope details from Statements of Work, constructing market definitions with standardized clinical codes, running high‑throughput data pipelines, and generating polished client deliverables. A dedicated reviewer agent, functioning as a clinical subject matter expert, applies LLM-as-a-Judge techniques to evaluate outputs and enhance auditability and trust. Finally, a human‑in‑the‑loop interface enables experts to review, validate, and intervene at any stage.
RESULTS: We have evaluated our multi-agent system in multiple real-world engagements, using IQVIA administrative claims data. Our solution orchestrated complex workflows among AI agents, existing tools, and human consultants. By removing bottlenecks, and by optimizing the consistency of engagements, we reduced the delivery time by at least 30%.
CONCLUSIONS: We have developed a trustworthy, scalable, and distributed multi-agent AI system that transforms the end-to-end delivery of Patient Journey projects. By integrating with collaboration platforms (e.g., SharePoint, Teams, Email) and eliminating delays caused by human-to-human handoffs, our system enhances efficiency, consistency, and throughput across teams.