Bringing hidden billing anomalies to the surface.
Saint-Luc, the largest hospital in Brussels, handles hundreds of thousands of patient consultations and tens of thousands of surgical interventions a year. Every one of them has to be billed correctly for the hospital to get paid, and at that scale, small billing errors were adding up into real, ongoing revenue loss.
The challenge
Saint-Luc is the largest hospital in Brussels. In 2025 alone, it handled 470,832 consultations and 20,744 surgical interventions. Each one fed into a billing process that relied on healthcare professionals to manually encode every procedure and the material used with it, using the right INAMI codes.
That left room for small omissions. A blood draw could be correctly encoded, for example, while the blood pouch equipment used alongside it was missed.
One missing item did not look like much. Repeated across Saint-Luc’s volume of activity, those omissions became a persistent source of revenue loss. They were also difficult to find. Small, frequent and dispersed across the billing data, many remained undetected until the month-end reconciliation.
The solution
We proposed two AI models and built the web application to put their findings to use. One checks every invoice for anything missing or inconsistent. The other looks across the hospital's billing data over time, catching patterns, code bugs, process delays, and system gaps that no single invoice would ever reveal on its own.
Saint-Luc's financial team no longer combs through the full billing volume by hand. The application surfaces a focused list of flagged anomalies, letting the team investigate the root cause and resolve each one directly, shifting their role from manually catching errors to reviewing what the models have already found.
The model accurately flagged 7 out of every 10 anomalies, confirming real billing errors worth the team's time to investigate. That precision changed how the team works: less time spent searching, faster resolution once an issue is found, and revenue that used to slip away quietly now gets caught and recovered.
The approach
We started with a proof of concept. Before committing to a production build, we tested whether patterns in Saint-Luc's INAMI codes were predictable enough to identify patient visits with potentially missing codes. The test confirmed that they were.
From there, we worked with Saint-Luc's invoicing team to understand how they handled anomalies day-to-day, what information they needed to investigate an error quickly, and how the application had to fit into their way of working. Those conversations shaped the two models behind it.
Development ran in short loops. Each step was reviewed with the invoicing team, and its feedback fed directly into what came next. Only validated choices made their way into the application.
That loop continued after delivery. As Saint-Luc used the application in real conditions, feedback from the invoicing team shaped each new version. We retrained the models on new data so they continued to reflect how the invoicing process evolved.
