Case study · Healthcare · Revenue cycle
How GeBBS Healthcare Solutions processes 50,000 charts daily using GenAI.
GeBBS · Revenue cycle management
Medical coding depended on certified coders reading unstructured notes. We built a Bedrock pipeline that codes 40-50% of charts without human intervention.
50,000
Charts processed daily
5-6x
Reduction in time per chart
40-50%
Charts coded without human intervention
95%+
Coding accuracy maintained
The opportunity
Coding accuracy that depended on reading every note by hand.
GeBBS converts clinical documentation into billing codes for 100+ healthcare organisations, at millions of charts a year and 95%+ accuracy. That accuracy came from certified coders interpreting unstructured physician notes-which is exactly the part that does not scale.
Rule-based automation could not capture clinical nuance, so volume growth landed on headcount. Turnaround time grew with it, delaying billing cycles and leaking revenue through codes nobody had time to catch.
Unstructured clinical notes at volume
Every chart needed a human to interpret free-text physician documentation.
Rule-based automation missed nuance
Deterministic rules could not capture the clinical context a coder reads for.
Heavy reliance on certified coders
Throughput was bounded by how many credentialled people were available.
Turnaround time grew with volume
Spikes became operational bottlenecks, delaying billing and leaking revenue.
The impact
Half the charts now code themselves, at the same accuracy.
20,000 charts fully automated a day
Across multiple specialties, 40-50% of charts are coded end to end without a coder touching them.
5-6x less time per chart
Multi level LLM processing turns unstructured notes into billing-ready entities before review begins.
Humans on the complex cases
Human-in-the-loop validation is reserved for charts that need judgment, not applied to all of them.
95%+ accuracy held
An audit and compliance layer in PostgreSQL keeps the evidence trail the accuracy claim rests on.
The stack
What it runs on.
- Model layer
- Amazon Bedrock, large language models
- Ingestion
- Automated chart ingestion from Amazon S3
- Code mapping
- Embedding-based similarity search against ICD-10 and CPT
- Audit and compliance
- PostgreSQL audit layer, human-in-the-loop validation
The road ahead
The pipeline is in. Next is widening what it covers.
With ingestion, coding and audit automated, the work moves outward: more specialties through the automated path, and the same accuracy measurement extended as coverage grows.
- 01
More specialties on the automated path
- 02
Broader code-set coverage
- 03
Tighter revenue-leakage measurement
Talk to us
Bring one process. Leave with an outcome architecture.
30 minutes on one function, mapped to the architecture behind it. No slides.