AI Revenue Intelligence for Hospital Claims AccuracyAI Revenue Intelligence for Hospital Claims Accuracy

The client is a US-based healthcare technology company that helps hospitals and health systems improve safety, efficiency, and accountability through real-time visibility. Its platform combines real-time location systems (RTLS), clinical workflow data, and operational analytics to give providers objective insight into staff activity, asset utilization, and care delivery across their facilities.

In US hospitals, the care that gets delivered is not always the care that gets billed. Despite widespread EHR adoption, real-world staff activity and equipment use remain under-documented. The gap between what happens at the bedside and what reaches the claim is an execution gap, not a data-hygiene failure, and it drives revenue leakage, compliance exposure, and audit risk that manual review cannot close at scale.

Key Challenges

1. Time-based billing leakage: Bedside clinical time and equipment deployment went under-documented, leading to time-based reimbursement leaks.
2. Presence is not usage: RTLS shows an asset is in the room, not whether it is in active clinical use, making billing justification and audit defense difficult.
3. Audit exposure: Without timestamped, objective records, claims and internal audits rest on manual logs, raising HIPAA and CMS risk.
4. Bundled-payment under-capture: Under BPCI and CJR, high-acuity cases risk underpayment when coders lack the telemetry to justify the right DRG.

The client needed a partner fluent in both hospital billing operations and the engineering required to build a production-grade intelligence layer over live RTLS and EHR data. Coditas brought AI platform engineering, healthcare data integration experience, and working knowledge of claims reconciliation, compliance, and clinical documentation. Rather than treat this as a reporting problem, Coditas approached it as a multi-agent system, one that reasons over telemetry, clinical notes, and claims together and keeps clinical and billing teams in control of every decision.

Coditas unified RTLS telemetry, ADT feeds, and claims into one governed system, then built a multi-agent reconciliation layer on top.

A resource-utilization engine mapped badge and asset movement against patient occupancy using geofenced room polygons, turning entry and exit events into timestamped presence logs per physician, nurse, and tagged asset. Accelerometer signal separated active clinical use from passive storage. The reconciliation layer then combined a deterministic rules engine (time-based CPT/E&M and equipment-runtime rules), a supervised model trained on the client’s historically adjudicated adjustments, and a retrieval-grounded agent that matches free-text clinical notes to presence data to test whether a flagged charge is documented and justified.

No agent moves a claim. Each discrepancy is surfaced with its evidence to a certified coder who accepts, edits, or rejects it, with every decision logged. Coditas rolled out in phases from single-unit pilot to enterprise integration, validating flag accuracy against coder adjudication at each stage.

Technologies

RTLS; ADT/EHR via HL7/FHIR (Redox); Kafka; Python ETL; deterministic rules engine; supervised pattern model; retrieval-grounded LLM for narrative correlation; time-series database; Kubernetes, S3, BigQuery; React dashboards; HIPAA-compliant architecture with RBAC, encryption, and audit logging.

Across the pilot units, the system converted raw telemetry into defensible revenue and compliance intelligence.

• $1.2M+ in previously unbilled revenue identified for coder review, including missed CPT codes, underbilled nursing time, and overlooked equipment charges.
• 28% reduction in audit-related claim denials on reconciled claims, through timestamped logs and documentation-matched justification.
• 3x faster discrepancy resolution, from 12 days to under 4.
• 15% improvement in DRG assignment accuracy on high-acuity encounters flagged for coding review.
• Staff-utilization benchmarks across 10+ units, surfacing both over- and under-allocation.

“For the first time, our revenue-cycle and compliance teams had an objective, timestamped account of what actually happened at the bedside captured automatically, evidenced against clinical documentation, and reviewed by a certified coder before any claim moved.”

— VP of Product, US Hospital Operational Intelligence Platform

The gap between care delivered and care billed is structural: the record is assembled by people mid-shift who cannot capture everything, while an objective account of presence already sits in the hospital’s RTLS data. Coditas turned that existing data into an active reconciliation layer, one that flags what was missed, evidences it, and keeps coders in control of the claim. That is what it means to put AI where the execution gap actually is.