AI initiatives often look convincing in pilot form but become much harder once they encounter real operational constraints: regulation, data quality, changing systems, privacy requirements, user adoption, and the need to demonstrate measurable value.

This session presents a practical case of introducing AI-based quality control in a medical diagnostics laboratory, where the implementation improved accuracy while reducing manual work.

The session examines the implementation strategy behind the result, including:
-- how the initiative was connected to a clear operational problem and measurable outcomes;
-- data-governance and compliance challenges, including GDPR and EU medical-device requirements;
-- architectural choices for working with data from a constantly evolving ecosystem of medical equipment;
-- the use of LLMs as part of the data-processing architecture;
-- phased learning and feedback loops rather than one-time deployment;
-- privacy-by-design and integration into existing workflows.

The broader lesson is that successful AI implementation depends less on choosing a single technology and more on designing an architecture that connects business goals, measurement, governance, data, controls, and continuous learning.

First presented: OOP 2026, Munich