Colombia's government health authority has deployed Anthropic's Claude artificial intelligence models to automate the processing of medical claims, cutting processing times from over four months to less than two weeks and reducing audit costs by nearly 75%. The system, called PAACME (Procesamiento y Análisis Automático de Cuentas Médicas), launched in June 2026 and represents one of the most significant real-world applications of large language models in public healthcare administration.

What Problem Was Claude Solving in Colombia's Healthcare System?

Colombia's Administrator of the Resources of the General Social Security Health System (ADRES) faced a critical bottleneck. Healthcare providers were submitting hundreds of thousands of medical claims annually for services provided to traffic accident victims, particularly those involving uninsured or hit-and-run vehicles. In 2025 alone, ADRES processed 432,234 new claims worth over 1.07 trillion Colombian pesos, with 99.6% related to uninsured vehicle accidents. The manual auditing process was glacially slow, taking up to five hours per case and stretching overall processing times beyond four months due to adjustments and objections.

The core challenge was that medical documentation submitted by clinics and hospitals arrived in unstructured formats. Human auditors had to manually interpret complex clinical records, verify medical relevance, check coverage eligibility, and validate claims against multiple data sources. This labor-intensive work created massive delays and prevented healthcare providers from receiving timely reimbursement.

How Did Anthropic's Claude Models Transform the Process?

Blend360, an AI services provider and Anthropic Authorized Reseller for Amazon Bedrock, deployed three Claude models to automate the workflow. The system uses Claude Opus, Claude Sonnet, and Claude Haiku to interpret and structure unstructured medical documentation, transforming raw clinical records into standardized medical and administrative data. The models evaluate medical relevance, assess coverage eligibility, automate key validations, and integrate data from multiple sources into a coherent audit trail.

The results were dramatic. Audit processing times dropped by approximately 96%, decreasing from five hours per case to just 5 to 10 minutes. The cost per audit fell by nearly 75%, while response times improved by up to 40 times, enabling ADRES to process significantly larger claim volumes with greater operational efficiency. The system achieved 87% accuracy in evaluating medical components, demonstrating that Claude's reasoning capabilities could handle the nuanced judgments required in healthcare auditing.

Steps to Implement AI-Driven Claims Processing in Healthcare Systems

• Data Preparation: Convert unstructured clinical records, medical notes, and claim documents into formats that large language models can process, ensuring all relevant information is accessible to the AI system.
• Model Selection and Deployment: Choose appropriately sized models for different tasks; PAACME used Claude Opus for complex reasoning, Claude Sonnet for standard processing, and Claude Haiku for lighter workloads, deployed through cloud infrastructure like Amazon Bedrock.
• Validation and Oversight: Build transparency mechanisms that create audit trails for every decision, ensuring human reviewers can trace how the AI reached its conclusions and maintain accountability for public healthcare resources.
• Iterative Refinement: Test the system on historical claims data, measure accuracy against human auditor judgments, and continuously improve model performance before full deployment.

Beyond speed and cost savings, PAACME strengthened data quality and decision-making. The system reinforces transparency by ensuring traceability throughout each validation step, safeguarding the proper use of public healthcare resources. Healthcare providers benefit from fewer administrative objections through early verification of information quality and reduced claim rejections, which optimizes recognition and payment cycle times.

"The use of technology in service of people transforms fragmented processes into solutions that address real needs. The collaboration between ADRES, Blend, and AWS demonstrates this transformation: reactive auditing processes evolve into models where data, advanced analytics, and cloud technologies enable faster, more accurate, and traceable decision-making," said Karla Wong, Education and Public Sector Leader at Amazon Web Services.

Karla Wong, Education and Public Sector Leader, Amazon Web Services

Why Does This Matter Beyond Colombia?

The PAACME deployment illustrates how Claude and similar large language models can tackle real-world government challenges that affect millions of people. Healthcare claims processing is a universal problem across public and private systems worldwide. The 96% reduction in processing time and 75% cost savings demonstrate that AI can handle complex administrative tasks requiring judgment, not just routine data entry. The system's 87% accuracy rate in medical component evaluation shows that Claude's reasoning capabilities can match or exceed human performance on specialized tasks.

The project also highlights the importance of transparency and auditability in government AI applications. PAACME was designed to maintain full traceability, allowing human reviewers to understand how the system reached each decision. This approach addresses a critical concern in public sector AI deployment: ensuring that automated systems remain accountable to the public they serve.

"This platform represents a turning point in the management of medical claims within Colombia's healthcare sector. Behind it lies rigorous work in data engineering and cloud architecture that enabled us to design models capable of interpreting, classifying, and auditing medical claims with a level of precision that previously depended exclusively on human judgment," stated Andrés Barrantes, Senior Vice President for LATAM at Blend.

Andrés Barrantes, Senior Vice President for LATAM, Blend

For ADRES, the system represents a major transformation in claims management by providing greater capacity, quality, and efficiency in auditing. The foundation is now in place for a scalable, data-driven operation that directly contributes to the sustainability of and trust in Colombia's healthcare system. As governments worldwide grapple with healthcare costs and administrative efficiency, the PAACME model offers a proven blueprint for how Claude and similar AI models can deliver measurable improvements in public sector operations.