Rail
91% accurate delay prediction for Eurostar.
91% accuracy
The problem
Transport operators are under pressure to do more with less: a smaller post-pandemic ridership, the same fixed network, and riders who increasingly avoid crowded services. Eurostar needed a way to anticipate how trains would actually arrive — minute by minute, station by station — so frontline operations could proactively manage staff and rolling stock instead of reacting to delays after the fact.
What we built
We built a time-based prediction layer that runs at station, train and carriage level. The engine uses our Physics-Informed AI methodology to model the spatial movement of the entire system, and re-runs frequently — triggered by impact factors like weather, upstream delays and holidays — so the forecast adapts as conditions change. The output: a live view of when trains will actually arrive, surfaced where the operations team makes decisions.
91%
accuracy predicting arrival delays
Live
rail environment, in production
Station · train · carriage
prediction granularity
“We worked with TUBR on an exciting new project: could we predict the arrival of our trains at their destination? We chose TUBR for many reasons. We wanted an innovative start-up. We wanted a partner who could match our drive and ambition. We wanted a company who could clearly add value to Eurostar. TUBR delivered all of this. The team were able to understand our needs quickly. They delivered a great product, really quickly. They were great at communicating, and at reassuring our frontline. They're a great partner. They absolutely delivered what we needed!”
Neil Powling, Eurostar Data Operations Manager · 23 October 2025