Hospital billing system with digitally sovereign AI support
Coding suggestions, plausibility checks and analytics for the Austrian LKF model. How our hospital billing system supports coding and controlling while patient data stays in Austria.

Support for coding and controlling
In Austrian hospitals, the quality of documentation determines points, and therefore money. Every forgotten secondary diagnosis, every incorrectly assigned procedure and every case that goes to the regional health fund with errors costs revenue or working time. At the same time, many hospitals are short of specialists in clinical coding and medical controlling. A hospital billing system with AI support can make a noticeable difference here, provided the hospital's most sensitive data does not end up in the wrong hands. Anexia Digital Engineering has developed its own product for this, which we adapt to each hospital and introduce on site.
Why hospital billing ties up so much effort
Since 1997, inpatient stays in Austria have been billed under performance-oriented hospital financing, known as LKF (Leistungsorientierte Krankenanstaltenfinanzierung). Each stay is assigned to a case group based on diagnoses and medical procedures, and the model sets a number of points for each group. On top of this come rules for length of stay, supplements for intensive care units and special cases for multiple procedures. Hospital outpatient care has its own points model.
The basis is two catalogues. Diagnoses are coded according to the Austrian version of ICD-10, procedures according to the ministry's procedure catalogue. Both are revised every year, as is the model itself. The LKF model 2027 was adopted on 3 June 2026, and the ministry provides the complete documentation at the end of September each year.
For the people working in coding, this means new rules year after year, while the number of cases and the length of medical reports keep growing. A discharge letter after a complex stay quickly runs to several pages. Extracting all the billing-relevant information from it is demanding work that requires concentration and is rarely done without interruption.
Where time and points get lost today
There are four places where coding and controlling regularly run into trouble. Secondary diagnoses appear in the free text of a report but are not coded. Procedures are documented but missing from the record or entered with the wrong code. Errors only surface when the fund reviews the case, when correcting them takes the most effort. And reports are compiled painstakingly by hand from spreadsheets and arrive too late to steer anything.
None of these problems comes down to a lack of care. They arise because volume and complexity grow faster than the teams. A hospital billing system with AI support can help at exactly these four points.
What our hospital billing system does
Our system supports billing under the LKF model with artificial intelligence in three places. It replaces neither the hospital information system nor the specialists. It takes reading, searching and routine checks off their hands.
Coding suggestions from the documentation
The system reads medical reports, discharge letters and surgical reports and suggests diagnoses and procedures based on them. For each suggestion, it shows the passage of text it relies on. The coder can therefore see at a glance why a code is being suggested, and can accept, change or reject it.
This evidence is not a convenience feature, it is the core. A suggestion without a traceable source would be worthless in everyday billing, because nobody can take responsibility for information whose origin is unclear.
Plausibility checks before billing
Before a case leaves the hospital, the hospital billing system checks it for contradictions and gaps. Does the main diagnosis match the procedure performed? Is information missing that the model requires for this case group? Is the length of stay unusual relative to the diagnosis? Do age, sex and code fit together?
Cases that stand out land in a worklist sorted by urgency. Corrections happen while the records are still fresh and the treating department can answer questions without having to search.
Analytics for controlling and management
The third component turns billing data into knowledge for steering the hospital. Figures on points, case numbers, lengths of stay and coding quality are available daily, broken down by department, case group and period. Forecasts show where the year is heading, and reports for management and heads of department are produced without manual work.
For controlling, traceability is what counts. Every figure can be traced back to the individual cases it is built on.
People make the decision
One point matters to us, and we raise it in every first conversation. Our hospital billing system does not code on its own and does not bill on its own. Every suggestion is reviewed and approved by a qualified person. Responsibility for billing lies with the hospital and cannot be handed over to any software.
This leads to a few properties we have built in deliberately. Every acceptance, change and rejection of a suggestion is logged. No suggestion enters billing without approval. Where documentation is unclear, the system should report the gap instead of guessing. And if the system fails, coding continues in the usual way.
Language models can be wrong, even good ones. Anyone who claims otherwise is promising more than the technology can deliver. What matters is that mistakes are noticed before they have consequences.
What digitally sovereign means in practice
Medical reports and discharge letters are among the most sensitive data of all. Many AI services on the market process text in data centres outside Europe or with providers subject to access by foreign authorities. For a hospital, that is a risk that can hardly be justified.
That is why our hospital billing system is built differently from the ground up.
The language models run either in your hospital's own data centre or in an Anexia Group data centre in Austria. No medical report leaves this setting. There is no interface to a model provider in a third country.
We use open models whose weights are available. This means the hospital does not depend on any single vendor, and a model can be replaced when a better one appears.
Patient data from your hospital is not used to train models for other clients. Adaptations to your documentation stay in your installation.
All access is logged, and roles and permissions follow the requirements of your IT department. Anexia is certified to ISO 27001.
For a hospital billing system, digital sovereignty is not a political buzzword. It means the simple ability to say at any time where the data is, who can see it and what happens to it.
Our case study on the video analysis app shows how we applied the same principle in a completely different setting. You can read about the role digitally sovereign AI plays in our plans for 2027 in our article on robotics development.
The legal framework in brief
We are not lawyers, and this section does not replace legal advice. Still, there are a few key points you should know, because they come up during internal approval of a hospital billing system with AI.
Under the GDPR, health data belongs to the special categories of personal data. Before going live, a data protection impact assessment is usually required, the processing has to be entered in the record of processing activities, and a data processing agreement is needed with external service providers. We supply the technical documentation for this.
The EU AI Act applies in stages. The obligation to ensure sufficient AI literacy among staff has applied since February 2025. The Digital Omnibus postponed the requirements for stand-alone high-risk systems to 2 December 2027. In our assessment, software that suggests codes based on existing documentation and makes no medical decisions does not usually fall into this category. Even so, the classification should be checked in writing for each hospital.
Regardless of the classification, we recommend documenting the use internally, monitoring error rates and taking user training seriously. These steps are part of introducing our system anyway.
Adapting to your hospital
No two hospitals document in the same way. Departments have their own templates, abbreviations and habits, and the range of services at a specialist hospital differs fundamentally from that of a standard hospital. An off-the-shelf hospital billing system quickly reaches its limits here.
That is why we adapt our product on several levels. Specialties and range of services determine which rules and checks take priority. The system is taught in-house abbreviations and phrasing. Worklists and approval workflows reflect your actual organisation, reports and figures follow the definitions of your controlling department, and roles and permissions match your authorisation concept.
This effort pays off, because whether a system like this is accepted depends almost entirely on whether it speaks its users' language and fits into their workflow.
Implementation on site
A hospital billing system cannot simply be imposed on a hospital from afar. Our teams come to you, talk to coding, controlling, IT and data protection, and accompany the introduction in person. The process has five steps.
It begins with analysis. We look at how cases are documented, coded, checked and reported today, and where the greatest friction lies.
Next comes the technical connection. The system is installed in your environment and connected to the hospital information system using the standards your system offers.
Then a pilot phase begins in one or two departments. At first, the suggestions run in parallel with the usual work, so their quality can be assessed without risk.
At the same time, we train the users. This is less about operating the software and more about judgement. When can you trust a suggestion, how do you recognise a wrong one, and how do you report it back?
Only once the pilot has proven itself is the system rolled out to further departments. You set the pace.
Fitting into the existing system landscape
A new tool that demands an extra screen and duplicate entries will not be used. Our hospital billing system therefore fits into the existing landscape. It pulls documents and case data from the leading system and returns checked codes to it. For data exchange, we use established healthcare standards such as HL7 version 2 and FHIR where your system provides them, and otherwise the manufacturer's existing interfaces.
Official scoring stays where it belongs. Our system calculates points alongside for orientation and shows how a change would affect them. The result of the designated procedure remains authoritative.
The annual model change
Every January, a new LKF model comes into force. Codes are added or dropped, case groups are redrawn, points are adjusted. For coders this means a changeover every year, and the same goes for software.
We maintain the rules and catalogues in our hospital billing system and apply the changes in good time before the cut-off date. Each case is checked against the model year that applies to it. Analyses across the turn of the year take the break into account, so that comparisons with the previous year are not misleading.
How the benefit can be measured
You will deliberately find no promises with percentages in this text. How much a hospital billing system with AI achieves depends on a hospital's starting point, and that can only be determined reliably in the pilot. We therefore agree on figures in advance and measure them before and after the introduction.
These include the time per case in coding and the share of cases closed without queries. They also include the number of corrections after reporting to the fund, the share of suggestions accepted, changed and rejected, and the time between discharge and completed coding.
If the figures after the pilot show no clear advantage, we advise against rolling out further. That is part of the agreement too.
Which hospitals the system suits
Hospitals with a high case volume, a broad range of services and short-staffed coding teams benefit most from a hospital billing system with AI support. This applies to fund hospitals as well as to private hospitals that bill through the Austrian financing fund for private hospitals. Operators with several sites gain additionally, because rules and analyses become consistent across all their hospitals.
Introduction makes less sense where documentation exists mainly on paper or as scanned images. In that case, the first step would be a different one, and we say so openly.
Frequently asked questions about the hospital billing system
Some questions come up in almost every conversation. Here are our answers.
Will the AI replace our coders?
No. It takes reading and routine checks off their hands. The professional decision and the responsibility remain with your staff.
Does patient data leave our hospital?
Only if you want it to. The hospital billing system can be operated entirely in your data centre. Alternatively, it runs in an Austrian data centre of the Anexia Group. There is no processing by model providers outside Europe.
Which hospital information systems does the product work with?
We clarify this during the analysis. The prerequisite is that your system provides documents and case data via an interface. The connection is part of the implementation project.
How long does the introduction take?
That depends on size, system landscape and internal approvals. You receive a reliable schedule after the on-site analysis.
What happens with an incorrect suggestion?
The coder rejects or corrects it. The feedback is stored and goes into improving your installation.
The next step
In a no-obligation first conversation, we show you the hospital billing system using examples and discuss how analysis, adaptation and introduction would work in your hospital. Afterwards, you can judge whether a pilot in one of your departments makes sense. Use our contact page to arrange a date.


