How Copenhagen Airport built a new AI framework in 18 months and eliminated all technical debt
Copenhagen Airport has spent 18 months building a new data platform on Databricks, meant to serve as the framework for future AI solutions. The project also removed all technical debt and legacy technology relating to data. The driver was physical: traffic and passenger numbers keep rising, but the airport's infrastructure on eastern Amager cannot be expanded quickly, so it has to use existing space more wisely and make better decisions based on data.
Tenna Ørum-Kristensen, director of Data & AI, says that when she joined nearly four years ago the data landscape was built on SSIS packages and extensive SQL, with performance and reliability problems and no clear data ownership. Management could get different figures depending on whether they checked a data warehouse, source tables or a spreadsheet. A review two years ago found heavy technical debt and little internal buy-in, since external consultants had built much of the infrastructure. The airport chose Databricks to avoid excessive vendor lock-in, with infrastructure managed as code on open-source technologies, and it expanded its in-house team.
Senior management gave the team a mandate to spend 90 percent of its energy and resources on migration and decommissioning legacy systems, and put that ahead of new business requests. The project was completed in August 2026, and the old data warehouses are being decommissioned. The platform now hosts 1,800 Power BI reports and scales up or down in the cloud. Developers reuse code components, and for some of them 90 percent of code is AI-generated using Databricks' built-in language models.
Next steps include a trial of ZeroOps, where AI agents monitor operations, spot errors and propose fixes with a human in the loop, and a "chat with data" tool that turns natural-language questions into SQL. Departments such as Sustainability will get their own mini-environments to test hypotheses. A key use case is predictive maintenance: sensor data across 12 kilometres of conveyor belts is used to predict disruptions, so maintenance can happen outside peak periods and avoid baggage delays and compensation claims.
According to Ørum-Kristensen, the hardest part was not coding but working out who owned the old solutions and what the business actually needed. In most cases the team rebuilt from scratch with users.