The business challenge
At a scale of 6.5 million claims a year, manual data gathering and exception handling created operational friction. First-pass auto-adjudication stood at 76%, while pended claims took 16 days to resolve. The architecture needed to improve throughput while maintaining clear decision ownership.
The architecture behind the outcome
I designed an event-driven, multi-agent pipeline on AWS covering submission, back-end data gathering through APIs, and policy and coding checks. Exceptions are routed to evaluators with a full reasoning package so that the human review starts with the relevant evidence and context.
Delivery & decision ownership
- Coordinate submission, data collection, and policy and coding checks through an event-driven pipeline on AWS.
- Package reasoning and supporting context for exceptions routed to qualified evaluators.
- Keep the rules engine as the system of record and a qualified human on every denial.
- Begin rollout in shadow mode to build compliance confidence before operational adoption.
Results that matter
| Measure | Before | After / outcome |
|---|---|---|
| First-pass auto-adjudication | 76% | 89% |
| Pended claim turnaround | 16 days | 5 days |
| Payment accuracy | — | 99.6% |
| Evaluator handle time | Baseline | 59% reduction |
| Annual claims volume | — | 6.5 million |
The program delivered $14.3M in annual benefit. Evaluator handle time fell by 59%, while first-pass automation increased and pended claims moved through the process substantially faster.