Billing errors are silently draining clinic revenues
A finance director at a French post-acute care network runs her monthly close, but the numbers don’t add up: €2.3M in rejected claims from supplementary insurers, incomplete patient files flagged too late, and a rehab coding backlog that has pushed their Information Systems Medicalization Program (PMSI) reporting into the following quarter. Not one of these issues stems from clinical failure. All are administrative – and all are preventable.
This scenario is playing out across hundreds of healthcare facilities in France. Since 2024, the country’s Medical and Rehabilitation Services (SMR) sector has operated under a new funding model where half of each facility’s revenue is tied to the accuracy and timeliness of its clinical coding and billing. The margin for administrative error has effectively been eliminated.
The problem is not unique to France. In the United States, hospitals now spend over 40% of their total expenses on administrative functions, and an estimated $43 billion per year trying to collect payments owed by insurers for care already delivered. Across OECD countries, administrative costs represent 15 to 30% of total healthcare spending .
How AI is impacting healthcare billing today
The cost savings AI can bring to the healthcare industry are phenomenal. In one deployment focused on operational optimization and another on generative AI, the combined financial recovery exceeded €35 million for a single care network.
In post-acute care facilities, revenue losses occur at every stage of the patient journey: at admission, during the stay, at discharge, and in receivables management. Until recently, each stage relied on manual processes, fragmented IT systems, and retrospective controls that only caught errors weeks after the money was lost.
Two complementary approaches bringing value
1) Advanced business intelligence applied to clinical operations
A major European post-acute care network recently deployed a unified data platform that broke down the silos between its clinical, operational, and financial data. The platform moved coding from a retrospective exercise (where clinicians filed reports weeks after care delivery) to a real-time, predictive process. Care acts were tracked against the patient’s tariff group as they happened, and the system modeled optimal discharge timing to maximize funding and avoid costly overruns. Result: €20M recovered through better alignment between actual care delivery and billing accuracy, simply by making existing data usable and actionable.
2) Generative AI applied to administrative workflows
In another deployment, AI agents were mapped to four critical steps of the billing cycle:
- Pre-admission: AI verifies insurance documents and coverage authorizations, flagging data inconsistencies before the patient arrives, eliminating errors that would cascade during their stay.
- During the stay: The system continuously verifies the completeness and compliance of the facility’s ERP documents, flagging missing items in real time rather than at discharge.
- Invoicing: A final AI-driven check ensures that the invoice (including external services and high-cost drugs) matches the initial coverage authorization exactly.
- Receivables: When supplementary insurers reject claims (a growing and increasingly algorithmic phenomenon), AI analyzes the denial, validates applicable exemptions, and automatically drafts evidence-based appeals.
Result: An 80% reduction in insurance contract errors at a pilot facility, and €15M recovered in previously unpaid invoices across the network.
Addressing the gaps in healthcare billing
Most AI initiatives in healthcare billing fail not because of technology, but because of three execution gaps.
Gap 1 – Missing data foundations: AI cannot optimize billing if the underlying data is fragmented, inconsistent, or delayed. The framework deployment described above required months of work to unify clinical, operational, and financial data into a single model before any analytics could run. Facilities that skip this step and jump straight to AI end up automating broken processes.
Gap 2 – Compliance treated as an afterthought: In France, the March 2026 CNIL and HAS AI in healthcare guidelines impose strict constraints: models must not memorize patient data, and retrieval-augmented generation (RAG) architectures must be localized and GDPR-compliant by design. Facilities that built AI workflows with compliance embedded from day one deployed faster than those that tried to retrofit it.
Gap 3 – The project is owned by IT alone: Every successful deployment shares one trait: joint ownership between the CIO and the operations or finance leadership. When AI in billing is framed as a tech project, clinical and administrative staff resist adoption, but when framed as a shared effort to eliminate the daily frustration of rejected claims and manual rework, adoption accelerates.
For healthcare executives considering this path, sequencing matters:
- Start with data: Unify clinical and financial systems, establish real-time reporting, measure your current leakage.
- Deploy AI: Begin with workflow steps where errors are most frequent and most costly – typically pre-admission checks and denial management.
- Scale from there.
The €35M recovered in these two deployments is not an outlier, but an indication of how much value is currently lost due to administrative inefficiency. As funding models tighten and insurers automate their denial processes, facilities that integrate data and AI into their billing cycle will protect their margins. Those that haven’t will keep hemorrhaging revenue they have already earned.
