A US dental services company replaced fragile, hand-coded scraping scripts with an AI-driven automation that verifies patient insurance data across dozens of providers - with near-perfect accuracy.
Problem
The company's daily operations depended on verifying insurance eligibility for every patient - pulling coverage details, plan specifics, and medical history from dozens of different insurer systems.
The existing approach was brute force: engineers manually built and maintained custom scraping scripts for each insurance provider's portal. Every time a provider changed their UI, updated their login flow, or restructured their data - a script broke. Someone had to find the issue, rewrite the code, test it, and redeploy.
It was a never-ending maintenance treadmill. Slow, fragile, expensive to maintain, and impossible to scale as the company added new insurance partners.
Wyzwanie
Insurance provider portals are not built for automation. They vary wildly in structure, authentication flows, and data formats. They change without warning. Traditional RPA and scraping tools demand rigid, rule-based scripts that shatter the moment a button moves or a form field changes its label.
The company needed a system that could adapt - not one that required a developer every time a portal updated its CSS.
Rozwiązanie
We designed an intelligent automation pipeline combining AI-driven browser interaction with multi-stage data processing:
- Playwright as the core automation layer - an AI agent that navigates insurer portals the way a human would: logging in, finding the right screens, interpreting page layouts, and extracting the relevant data.
- Multi-stage LLM processing pipeline that takes raw extracted data - combining insurance plan details and patient medical history from multiple sources - and transforms it into clean, structured output.
- Direct JSON integration with the client's core system, delivering verified insurance data ready for immediate downstream processing and analysis.
The result is a system that handles the full verification workflow end-to-end: login, navigation, extraction, processing, and delivery - without human intervention.
Wyniki
| Metryczny | Wpływ |
|---|---|
| Accuracy | Nearly 100% on processed insurance records |
| Time to production | ~2–3 months from project start |
| Manual script maintenance | Eliminated - AI adapts to portal changes |
| Integration | Direct JSON output to client's core system |
| Status | In production, running daily since early 2025 |
What used to require a team of engineers maintaining a patchwork of fragile scripts is now handled by an AI system that runs autonomously, every day.
As the client noted on Clutch: the project delivered a production version within approximately 2–3 months, with almost 100% accuracy. The COGITA team adapted quickly to the client's workflows and responded fast when issues arose (see the full review here).
