St. Luke’s Turns 400,000 Patient Interactions Into an Agentic AI Use Case

For St. Luke’s Hospital in Kansas City, Missouri, the case for agentic artificial intelligence started with a simple operational challenge: hundreds of thousands of patient interactions required thousands of hours of human labor every month. The solution put AI agents in the middle of those interactions, handling routine work and helping staff focus on patients who need more attention.

St. Luke’s Hospital was handling about 400,000 patient interactions each month across prescription requests, appointment scheduling and clinical consultations. Behind that volume was a labor-intensive operation that relied heavily on call-center agents, administrative employees and triage nurses. As demand grew, maintaining service levels required more staff, increasing the financial pressure on the organization.

The workload was particularly costly because many interactions involved administrative tasks that did not require clinical judgment. Based on an average handle time of 15 minutes per call, St. Luke’s monthly volume translated into roughly 100,000 labor hours. Direct labor expenses exceeded $3.5 million a month and approached $43 million annually.

The structure also affected how employees spent their time. Agents handled eligibility verification, prescription routing, appointment scheduling and documentation. Calls were frequently transferred between departments, extending interactions and adding steps to the patient experience. Triage nurses also spent time on clerical work, leaving less capacity for direct medical assessment. Longer queues and more complicated calls contributed to employee fatigue and longer waits for patients seeking routine assistance.

Putting Agentic AI to Work

NextGen Coding proposed an agentic AI call center designed to take over a portion of those routine workflows while preparing more complex interactions for clinicians. The system transcribes incoming calls, classifies their intent and routes them either to an autonomous agent or a human operator according to predetermined escalation policies. The system also includes dashboards tracking automation rates, handle times, abandonment risk and escalation frequency, with clinical oversight teams reviewing flagged interactions.

For calls that required human assistance, automated intake could gather information before the handoff. NextGen estimated that this would reduce the average handle time for those calls by about 1.5 minutes. The system was also designed to collect symptoms, relevant history and severity indicators before a patient reached a physician or nurse. A structured summary would then give the clinician useful context at the beginning of the consultation, allowing more of the interaction to focus on medical assessment.

The transformation was designed to happen gradually. Initial deployment would use conservative automation thresholds and staff onboarding, followed by expansion into scheduling and prescription workflows. The system could then be tuned to improve intent recognition, identify duplicate requests and determine when an interaction should be escalated to a person.

Labor Hours Reduced

At steady state, NextGen projected that human-handled interactions could fall from about 400,000 to roughly 120,000 per month. Monthly labor hours were projected to decline from 100,000 to nearly 27,000, while required staffing would fall from approximately 781 agents to nearly 211. Monthly workforce spending was projected to drop from roughly $3.57 million to about $965,536.

NextGen estimated that the resulting labor savings would exceed $2.6 million a month, or more than $31 million in annual gross efficiency gains. After accounting for projected AI infrastructure costs of about $212,250 per month, the case study estimated more than $28 million in annual net financial benefit.

St. Luke’s case study illustrates how agentic AI can move beyond answering questions to taking action across routine workflows. By handling administrative tasks, preparing information for clinicians and escalating more complex interactions to employees, the model gives the health system a way to use human expertise where it is most needed while allowing software to manage a larger share of the day-to-day workload.