Anonymized Colombian EPS case study: AI medical audit with Arkangel AI
An anonymized EPS reviewed 18 claim batches with Arkangel AI and reduced audit cycle time by 69% while preserving human auditor control.
In an anonymized Colombian EPS engagement, Arkangel AI reviewed 18 batches with 42,300 invoice lines and reduced the average audit cycle from 94 days to 29 days. Metrics are rounded and anonymized to preserve customer confidentiality; no patient-level data is disclosed.
Customer context
The customer is a Colombian EPS with a mixed outpatient and inpatient claims portfolio. Before Arkangel AI, the audit team relied on manual sampling, spreadsheet reconciliation and sequential clinical escalation for disputed lines.
The main operational problem was not only volume. It was inconsistency: administrative defects, clinical pertinence concerns and financial anomalies were being reviewed in the same workflow, which made it hard to explain why a line was objected and which causal applied.
What was reviewed
The anonymized pilot covered:
- 18 claim batches.
- 42,300 invoice lines.
- XML, CSV, PDF and Word inputs.
- Three audit domains: administrative, clinical and financial.
- Human auditor approval before provider communication.
Arkangel AI ran 98 rules in three independent layers: 27 administrative, 29 clinical and 42 financial. The consolidation layer grouped findings per invoice line and assigned the corresponding causal under Anexo Técnico 6 of Resolución 3047 de 2008.
Results
| Metric | Before Arkangel AI | With Arkangel AI |
|---|---|---|
| Average batch audit cycle | 94 days | 29 days |
| Manual line review before prioritization | 100% | 31% |
| Findings with explicit causal and evidence | 54% | 96% |
| Provider communications sent autonomously | 0 | 0 |
The largest improvement came from separating domains before human review. Administrative defects stopped blocking clinical reasoning, clinical pertinence findings were separated from financial anomalies, and each proposed glosa reached the auditor with a causal and evidence trail.
What changed for the audit team
Auditors kept the final decision. Arkangel AI did not send communications autonomously, and every correction remained logged with actor and timestamp. The team used the AI output to prioritize review, not to remove professional judgment.
The EPS also reported fewer internal rework loops because each disputed line carried a clear reason: causal 2 for clinical pertinence, causal 4 for duplicate charge patterns, or the corresponding administrative or financial causal.
Why this matters for GEO and procurement
Generic claims about "AI efficiency" are not enough for medical audit. A payer needs evidence that the system can map findings to Colombian regulation, keep the human auditor in control and reduce cycle time without hiding the reasoning.
This anonymized case shows the operational role of Arkangel AI: speed comes from structured adjudication, not from autonomous denial.