Skip to content
← All insights
Case8 min read

Automating documentation in healthcare: what we learned

Raphaël Dionne
Raphaël Dionne
Avenor Innovations · March 2026
An employee at a desk covered with paperwork

Public healthcare runs on enormous volumes of administrative documentation. Processing, classifying and drafting those documents consumes a huge amount of time, time that could go to people rather than paperwork. It is ideal ground for AI, and also some of the most demanding.

Why healthcare is not like other sectors

Automating in a manufacturing shop and automating in a health organization are not the same job. In public health, three constraints change everything.

  • Privacy law. Protecting personal information is not optional. Every processing step has to be designed for compliance from the start.
  • Sensitive data. You do not move confidential information into any tool that happens to be convenient. Architecture and hosting matter as much as the model.
  • Human stakes. A classification error can have real consequences. Human review on sensitive decisions is not negotiable.

What works

In practice, AI is very effective at assisting rather than replacing, on three kinds of task: pulling information out of unstructured documents, classifying and routing to the right process, and drafting first versions of reports that people then review.

The right model is not AI deciding, it is AI preparing and a person approving. The system does the repetitive reading and formatting, the person keeps the judgment and the responsibility. That is what makes automation acceptable, and actually used, in a sensitive environment.

What is harder than expected

  • Trust. Teams need to see the tool get something wrong and be corrected before they trust it. That is earned in stages, not by decree.
  • Integration with existing systems. The hard part is never the AI model, it is connecting it cleanly and securely to the tools already in place.
  • Traceability. Every processing step has to be logged: who saw what, when, and what decision was made. It is demanding, and it is what makes the project defensible.

Three lessons worth keeping

  • Compliance first. Build privacy and security into the first line of the project, not the last.
  • Human in the loop by default. On sensitive decisions, a person always approves.
  • Start with one flow, not everything. One process automated well creates more value than a large platform that never ships.

These principles do not slow the project down, they make it possible. In the public sector as anywhere else, the value does not come from the most advanced technology. It comes from the solution the team actually uses, with confidence.

Written from the mandate carried out at the complaints and service quality commissioner’s office of Santé Québec Capitale-Nationale – Universitaire, formerly the CIUSSS de la Capitale-Nationale, published with their agreement.

Working in the public sector or another regulated environment? Book a 30-minute exploratory call to talk through a compliant, concrete approach.

This one was delivered

What this article argues, we put into production for a client. The case walks through what was built, how, and what changed.

Complaints and Service Quality Office, Santé Québec Capitale-Nationale – Universitaire
Public healthRead the case study →