Performance, Scalability and Maintenance: An Experimental Study of RPA vs. Autonomous Agents in the Healthcare Domain
| Source: arXiv AI
Tags: RPA, Playwright, browser-use, LangGraph, Gemini 2.5 Flash-lite, Healthcare Automation
In a healthcare portal test, classic Playwright RPA beat an LLM browser agent built on browser-use, LangGraph and Gemini 2.5 Flash-lite: 98.90% vs 96.49% accuracy, about 13 times faster and six times cheaper per report.
Details
Healthcare data often sits in fragmented legacy web portals. Robotic Process Automation is deterministic but breaks on interface changes, while LLM agents promise adaptability at higher compute cost. The authors built both to retrieve mammography and breast ultrasound reports from a lab portal: an RPA solution in Playwright and an agent solution using browser-use, LangGraph and Gemini 2.5 Flash-lite. Across two rounds a week apart, RPA reached 98.90% accuracy and 98.27% F1 versus 96.49% and 94.48% for the agent. RPA took 16.07 s per patient versus 207.21 s and cost USD 0.030 per report versus 0.187. Both completed every step; the agent varied more over time. For a stable interface, scripted RPA remains the better choice, a useful data point for teams weighing agents against existing automation.