Netcompany

Three named AI systems are documented as running rather than piloted: DSB's train-delay swarm, a model per railway line orchestrated by the firm's PULSE platform, which predicts arrival to within three minutes at 95 percent accuracy; Topdanmark's TopGPT, which masks personal data on locally hosted models before a retrieval step reads the customer's own policy, live around the clock since April 2024 and past 80,000 conversations by that November; and Copenhagen Airport's luggage-timestamp model, in production since 2016 and retrained whenever it drifts.

The AI work is a minority of what the firm publishes—the case index is dominated by large public-sector and enterprise IT delivery for the European Commission, NHS England and the UK Ministry of Defence—and the European Commission's FERMI disinformation tool has completed three police pilots on live data rather than a full deployment.

What raises confidence is what the write-ups admit: the Copenhagen Airport study records the queue-prediction model that never reached production beside the two that did, so ask for the same candor about your own use case.

The weaker fit is a buyer wanting one scoped model from a small team—this is a listed group of more than 9,500 people that builds on its own platforms and expects to operate what it builds, so settle minimum engagement size and account staffing before scoping.