A German healthcare billing service provider deployed doXray inside its own infrastructure to classify, validate and extract data from payment reminders, invoices and handwritten forms. Turnaround has dropped from days to minutes, touchless processing has doubled to more than 80%, and field-level accuracy is above 95%. The system has been in continuous production since 2023, and not a single document has left the building.
This BPO processes billing documents for statutory health insurers, physicians’ associations and hospitals in Germany. Every month, 1.5 million documents arrive in every form a German healthcare billing operation produces: printed payment reminders, co-payment notices, supplier invoices and handwritten forms.
Keying this work by hand put a hard ceiling on how much volume the business could take on, and it stretched turnaround into days. Automation was the obvious answer, but the obvious tools were off the table. Patient-related data could not be sent to a cloud API.
The documents this BPO handles are not its own. They belong to its clients, and behind those clients are patients. Sending that material to a cloud service was not a matter of preference or price. It was something the business could not offer its clients and expect to keep them. That requirement alone ruled out most of the document AI market.
Within the remaining options, handwriting was the hardest problem. Classification and invoice processing are well understood. Handwritten forms at production volume, filled in by people who never expected a machine to read them, had resisted every earlier attempt at automation.
The goal was to automate classification, validation and extraction at full volume without any document leaving the customer’s infrastructure.
doXray was installed entirely inside the BPO’s own environment. The models run locally and the data stays where it is. A single deployment covers three workflows:
Payment reminders and co-payments: documents are classified on arrival, and the relevant fields are extracted, validated and routed straight into downstream systems.
Invoice validation: invoices are classified, extracted and checked against the customer’s own rules and reference data. Anomalies go to a human reviewer, and everything else passes through untouched.
Handwritten forms: structured data is extracted from handwritten input, at the same volume and inside the same secure perimeter as the other two workflows.
For an organisation that processes documents on behalf of health insurers, running on-premise is not a technical detail. It is what makes the service offerable to its clients at all.
Three years after go-live, the system processes the full monthly volume in production:
Touchless processing doubled: the share of documents processed with no human intervention rose from 40% to more than 80%.
Turnaround cut from days to minutes: documents move from arrival to downstream systems almost immediately.
Handwriting solved at production volume: forms that used to be read and keyed by staff are now extracted automatically at over 95% field-level accuracy.
Processing costs down: with most documents flowing through untouched, the cost per document has dropped sharply.
Growth without extra headcount: the business can take on more volume and more clients without hiring additional FTEs.
| Evaluation criteria | Before doXray | After doXray |
|---|---|---|
| Touchless processing | 40% | >80% |
| Field-level accuracy | 60% | >95% |
| Monthly volume | Capped by keying capacity and human in the loop | 1.5M documents |
| Turnaround | Days | Minutes |
Book a live demonstration tailored to your document processes.
To book a demo, reach out to us directly:
contact@doxray.comdoXray's document recognition built into the Exportdocuments.com webshop: upload the commercial invoice, get the export declaration.
Insurance applications compared against medical reports automatically, in a secure UK cloud, so underwriters see only what differs.