Problem

Global routing models are built for coverage, not accuracy

Different vessel types require separate models because these details matter to how vessels move.

The same principle applies to geography. Every waterway has its own navigational patterns and practical constraints that shape vessel behaviour.

However, to achieve global coverage, most models must simplify or generalise this regional detail. This reflects a fundamental modelling trade-off: prioritising coverage means sacrificing detail.

That trade-off becomes critical as vessels approach their destination, where small deviations can quickly compound. Routes stop reflecting reality, arrival times drift, and planners lose confidence in predictions precisely when operations require reliability.

In the final approach, local detail matters. And it’s exactly where global models struggle.

Solution

Region-specific AI routing models built on real behaviour

We build AI models trained on AIS data which learn how different vessel types move in specific waterways. These models are not only specialised by vessel type and operational context, but by geography, ensuring predictions are finely tuned to the reality of how vessels actually move.

This approach allows our models to perform a variety of important prediction tasks:

• Trajectory - predict the most likely route a vessel will take between two locations.
• ETA - estimate arrival times with data-driven confidence bounds by vessel type.
• Plan - generate fuel-efficient passage plans aligned with a target transit time to meet schedules.

Our approach delivers clear advantages over existing models:

Use Cases

Our models support any application that requires reliable vessel movement predictions during the last 12-24 hours of a journey towards a destination.

Port operator

• Optimise berth scheduling and reduce operational disruption with continuously updated trajectory predictions, ETAs, and confidence bounds.

Fleet manager

• Improve fuel efficiency and schedule adherence by generating data-driven passage plans and monitoring route deviations.

Commodity trader

• Reduce scheduling uncertainty with data-driven ETAs and confidence bounds during the critical final hours of a voyage.

Autonomous vessel

• Enable safe navigation in complex environments by embedding localised navigational practices and validating navigation systems against modelled vessel behaviour.

Workflow

Insights without changing how you work

Access reliable predictions without changing the way you work. Our models are designed to integrate cleanly into existing workflows and systems, making it easy to move from data to decision.

Generate predictions via:

• Viewer - a no-code, browser-based interface for generating and visualising individual predictions powered by our APIs.
• APIs - low-code integration into planning, monitoring, or decision-support systems. See the Docs for details.

Start by supplying:

• Vessel type and length category - predictions account for specific vessel characteristics.
• Locations - a live or historical vessel position, plus a current or intended destination, supplied as simple coordinates. Vessels positions can be easily obtained from MarineTraffic.

Models

Tested against reality

Our models undergo comprehensive testing before deployment, using thousands of randomly selected real vessel transits with results aggregated for robust evaluation. To ensure reliability, each model is benchmarked on two key tasks:

• Trajectory reconstruction - assessing how accurately the model can recreate vessel tracks from start to finish - this forms the core foundation of our predictions.
• ETA prediction - evaluating how accurately the model estimates vessel arrival times by building on the predicted route and capturing vessel behaviour throughout the journey.