AI weather routing moves from promise to numbers: 5.3% average fuel cut in 2024 trials

The debate over whether AI voyage optimisation actually saves fuel is being settled by data. As Digital Ship reports, trials of the OSR-OW routing tool — built by Weathernews and HD Hyundai Marine Solution — across eight vessels in 2024 delivered an average fuel reduction of 5.3%. The developers now guarantee a floor of at least 3% savings, which they put at roughly $150,000 for every 1,000 tonnes of fuel a ship consumes.

The mechanism is straightforward in principle and hard in practice. Modern systems blend weather forecasts, ocean currents, port traffic and each vessel’s own performance model to choose and continuously adjust a route — trading a slightly longer track for calmer water, or shaving speed at the right moment to arrive on time with less fuel burned. The gains are largest, the OSR-OW developers note, on long voyages through open water where weather has the biggest influence on the routing choice. Adoption is already broad: more than 3,000 vessels use Weathernews’ OSR system and over 600 run HD Hyundai’s OceanWise platform.

None of this makes weather routing a magic lever. The IMO’s own GreenVoyage2050 guidance frames the efficiency gain from weather routing as vessel- and route-dependent, typically in the low single digits — meaningful, but not transformative on its own. The value comes from doing it on every voyage rather than occasionally.

What sharpens the incentive is regulation. Under the Carbon Intensity Indicator (CII), a ship’s emissions rating is tied directly to how efficiently it moves cargo, so fuel saved and rating improved are now two names for the same task. That is why routing vendors increasingly wire their optimisation into CII planning: the route that burns less fuel is also the route that protects the vessel’s rating.

Apeks view — The story here is not that AI found a shortcut across the ocean; it is that a soft claim became a measured, contractable number. A 3% guaranteed floor changes the conversation from “does it work?” to “prove it did.” That, in turn, raises the real question: can the system show its working — the inputs, the assumptions, the counterfactual voyage — so an owner or an auditor can trust the saving? Optimisation you cannot verify is just a lower number on a slide.