PMSI, CCAM, T2A: How AI Is Automating Hospital Billing in 2026

Key Takeaways at a Glance

Introduction

In France, every hospital stay must be translated into codes (PMSI, CCAM, GHM) to trigger the T2A funding that keeps hospitals financially afloat. This medical coding process, carried out by Medical Information Technicians (TIM), is time-consuming, complex and vulnerable to human error. As French hospitals face mounting budget pressure, incomplete or incorrect coding can represent millions of euros in lost annual revenue for a single institution.

Artificial intelligence is changing this equation profoundly. In 2026, NLP (Natural Language Processing) models can read a consultation report, identify the procedures performed, and automatically suggest the corresponding CCAM or GHM codes, with accuracy exceeding 90% on common procedures. This is no longer prospective thinking: it is an operational reality already deployed in pioneer institutions across France.

Galeon, active in 19 hospitals and processing more than 3 million patient records, built its intelligent EHR around precisely this logic: structuring medical data at the source so that it becomes immediately usable, including for automating billing. An EHR that produces clean data at the point of clinical entry is the most powerful billing tool available.

This article details how PMSI, CCAM and T2A work, the concrete mechanisms by which AI automates medical coding, the measurable financial gains for hospitals, and the real limitations to understand before deploying.

What Are PMSI, CCAM and T2A, and Why Are They Interconnected?

The PMSI (Programme de Médicalisation des Systèmes d'Information) is the national system that describes and classifies hospital activity in France. Introduced in the 1980s and progressively extended, it works on the principle that every hospital stay can be summarised by standardised codes, grouped into GHMs (Homogeneous Patient Groups).

The CCAM (Classification Commune des Actes Médicaux) is the national reference that lists and codes all technical procedures performed by healthcare professionals in France. It contains more than 18,000 codes, each associated with a precise tariff value. A procedure that is not coded, or is incorrectly coded, is a procedure that is not reimbursed.

The T2A (Tarification à l'Activité, or Activity-Based Financing) is the funding model that directly links a hospital's revenue to its coded and reported activity. Introduced in 2004 for MCO facilities (Medicine, Surgery, Obstetrics), it creates a direct dependency between the quality of PMSI/CCAM coding and the level of hospital funding.

The conclusion is simple: in a T2A system, a hospital that codes poorly, funds itself poorly.

Why Is Hospital Billing Still a Major Problem in 2026?

Despite three decades of digitisation, medical coding in most French hospitals remains predominantly manual, fragmented and sub-optimal. Several persistent reasons explain this.

First, the complexity of reference frameworks is exponential. CCAM is regularly updated (several revisions per year), GHM grouping rules are highly technical, and ensuring consistency between the principal diagnosis (DP), significant associated diagnoses (DAS) and CCAM codes demands expertise that TIMs alone cannot apply exhaustively across high volumes.

Second, TIMs are under permanent strain. According to an ATIH survey, medical information departments face a growing shortage of qualified professionals, compounded by increasing workloads. An experienced TIM can process between 30 and 60 complex files per day, a volume that is insufficient in large institutions.

Third, clinical data is often unstructured. Medical reports are written in free text, in heterogeneous formats, sometimes dictated or handwritten. Extracting codable information from these documents requires time and expert reading.

The result: according to a study by the CHU de Bordeaux published in 2023, between 5% and 12% of hospital activity may be subject to under-billing or incomplete coding, depending on the institution and the specialties involved.

How Does AI Concretely Automate PMSI and CCAM Coding?

AI-based billing automation rests on three complementary technological building blocks, operating in real time within the EHR.

NLP: Reading and Understanding Clinical Language

Natural Language Processing models are trained on millions of clinical documents. They are capable of identifying, within free-text reports, the clinically relevant entities: diagnoses, procedures, medications, comorbidities. Once these entities are extracted, they are automatically mapped to the corresponding PMSI, CCAM or ICD-10 codes.

In 2026, leading systems achieve concordance rates exceeding 92% on common procedures (outpatient surgery, standardised consultations), and between 75% and 85% on complex multi-pathology cases.

Assisted Coding: Suggest, Don't Impose

The most widely deployed approach in 2026 is not fully autonomous coding, but assisted coding: AI proposes a ranked list of relevant codes with a confidence score, which the TIM then validates, modifies or rejects. This model preserves human accountability while reducing per-file processing time by 40% to 60% according to pilot studies.

Anomaly and Omission Detection

Beyond initial coding, AI plays a crucial quality control role: it identifies inconsistencies between the principal diagnosis and coded procedures, flags procedures likely performed but not documented, and alerts on files at risk of rejection by the Health Insurance system during T2A audits.

An AI billing engine does not replace the TIM: it gives them additional eyes across every single file.

What Is the Real Financial Impact of T2A Automation on Hospital Revenue?

The financial gains from coding automation are documented and significant. An analysis conducted across several institutions that deployed AI-assisted coding tools between 2022 and 2025 (ANAP report, 2024) shows consistent results.

Pilot institutions report on average:

• A T2A revenue recovery of between 3% and 8% through the detection of uncoded or under-coded procedures
• A reduction in the rejection rate of files by the Health Insurance system of 30% to 50% on average
• A reduction in the average PMSI production delay from several weeks to a few days post-discharge

For a mid-sized hospital with a T2A budget of €100 million, a 5% improvement in recovery rate represents €5 million in additional annual revenue, without any increase in actual activity. This is money the hospital had already earned, but was simply not billing.

The impact is even more pronounced in high procedure-density specialties: surgery, intensive care and oncology, where the complexity of stays multiplies the risk of omissions.

How Does Galeon Integrate AI Billing Into Its Intelligent EHR?

Galeon's proposition goes further than simple coding assistance at end of stay. The Galeon EHR is designed from the ground up to structure medical data at source, at the moment of clinical entry by the healthcare professional, which makes downstream coding mechanically simpler, faster and more reliable.

In practice, this means:

• Procedures are entered in a standardised, interoperable format, compatible with CCAM and GHM reference frameworks
• Diagnoses are documented with comorbidities from the point of consultation, without subsequent re-entry
• AI integrated into the EHR can suggest PMSI coding in near real time, without waiting for patient discharge

This clean data at source model is fundamentally different from the classic approach, where TIMs re-code a posteriori from information imperfectly entered into heterogeneous systems.

Galeon deploys this intelligent EHR in 19 hospitals, including 2 university hospital centres (CHU), with over 3 million patient records and several thousand active healthcare professionals. This structured, shared dataset, secured by Blockchain Swarm Learning®, also constitutes a unique training ground for refining AI coding models on real French medical data.

To better understand how AI integrates into the care pathway beyond billing, see our article.

Manual Coding vs AI-Assisted Coding: The Comparison

What Are the Real Limits and Challenges of Hospital Billing Automation?

It would be inaccurate to present AI medical coding as a frictionless solution. Here are the concrete limitations to integrate into any serious deployment strategy.

Input Data Quality Remains the Determining Factor

An AI model can only code what is documented. If medical reports are incomplete, rushed or written in a non-structured format, the performance of automated coding drops significantly. AI amplifies the quality of existing data: it does not create it.

Medical and Regulatory Responsibility Remains Human

In France, responsibility for PMSI coding and T2A transmission to ATIH rests legally with the institution's Medical Information Department (DIM) physician. No AI tool can transfer this responsibility. Every algorithmic suggestion must be validated by a qualified professional.

The Risk of Over-Coding Is Real

A poorly calibrated AI tool, or one used without supervision, can generate over-coding: procedures suggested but not actually performed. This risk is particularly problematic as it exposes the institution to enhanced Health Insurance audits and potential financial penalties. Full traceability of coding decisions is therefore non-negotiable.

Interoperability Remains a Systemic Challenge

Most French hospitals operate with heterogeneous EHR systems, legacy infrastructure and non-standardised data formats. Integrating an AI coding tool into this environment without rebuilding data flows is complex, costly and often underwhelming in the short term.

The Human Factor and Evolving Roles

TIMs are experiencing a profound transformation of their role. Moving from a data-entry function to a supervision and strategy function requires training, support and carefully managed change management, factors that are consistently underestimated in deployment projects.

FAQ: PMSI, CCAM, T2A and AI in 2026

What is PMSI and why is it crucial for hospital funding? PMSI (Programme de Médicalisation des Systèmes d'Information) is the national system that classifies each hospital stay into a Homogeneous Patient Group (GHM). This GHM code determines the T2A tariff paid by the French Health Insurance to the hospital for that stay. Poorly documented or incomplete PMSI directly results in funding below actual activity: it is the primary source of hospital under-revenue in France.

How does AI automatically code CCAM procedures? AI coding engines use NLP models trained on millions of clinical documents. They read medical reports, identify performed procedures and map them to the corresponding CCAM codes. In 2026, leading systems achieve over 92% concordance on common procedures. Coding remains validated by a TIM or DIM physician before transmission to ATIH.

Is AI coding compliant with French Health Insurance and ATIH requirements? An AI coding tool is compliant when used as a decision support tool, not as an autonomous decision-maker. French regulation requires a qualified professional to validate each transmitted code. Well-designed tools produce full traceability (who coded what, when, on what basis), which facilitates audits and reduces the risk of disputes.

What concrete financial gain can be expected from T2A automation? Institutions deploying AI-assisted coding tools report between 3% and 8% of additional T2A revenue recovered, primarily through the detection of uncoded or under-coded procedures. For a hospital with a T2A budget of €100M, that represents between €3 and €8 million in additional annual revenue, without increasing actual activity.

How does Galeon secure billing data with blockchain? Galeon uses Blockchain Swarm Learning® to allow hospitals to share AI learnings without ever moving patient data outside their local server. Every training or usage action is indelibly recorded on the inter-hospital blockchain. Billing data remains under the full sovereignty of the hospital at all times.

Will AI replace Medical Information Technicians (TIM)? No. TIMs are not replaced: they are repositioned. With AI handling code suggestions and consistency checking, TIMs focus on qualitative supervision, complex case management, relationships with medical teams, and revenue optimisation strategy. This is an evolution of the profession, not its elimination.

Which hospitals are already using AI for PMSI/CCAM billing? In France, several university hospitals and hospital groups (APHP, CHU de Bordeaux, private groups such as Ramsay Santé) have initiated AI coding assistance pilots or deployments between 2022 and 2025. Galeon, active in 19 hospitals including 2 CHUs, integrates this logic directly into its intelligent EHR, with the advantage of medical data structured at source from day one.

In Summary

Hospital billing (PMSI, CCAM, T2A) is one of the most underexploited levers in French healthcare finance. In 2026, artificial intelligence enables medical coding to be automated with accuracy exceeding 90% on common procedures, reducing omission errors that cost hospitals between 3% and 12% of T2A revenue depending on the institution. This automation does not replace TIMs: it frees them from repetitive tasks so they can focus on supervision and strategy. Galeon builds this logic into the core of its intelligent EHR, active in 19 hospitals and structuring over 3 million patient records: clean medical data at the point of entry is the most powerful billing lever that exists. Real limitations (input data quality, human regulatory responsibility, over-coding risk) must be factored into any serious deployment project.

Sources

1. ATIH - Agence Technique de l'Information sur l'Hospitalisation. Methodological Guide for MCO Stay Summaries. 2024.
2. ANAP - Agence Nationale d'Appui à la Performance. Feedback on AI Use in Medical Information Departments. 2024.
3. ameli, le site de l’Assurance Maladie en ligne ‍
4. HAS - Haute Autorité de Santé. Artificial Intelligence in Healthcare: Ethical and Regulatory Issues. 2023.
5. French Ministry of Health. Report on Hospital Digital Transformation — Ségur du Numérique. 2023.  
6. CHU de Bordeaux / Medical Information Department. Study on Under-Coding Rates by Specialty. 2023. (Internal data, available on request from the DIM.)
7. Galeon. Blockchain Swarm Learning® — Technical Architecture and Hospital Use Cases.