AI Fuel Optimization in Shipping: A Real-World Maritime Case Study

One shipowner's AI-controlled propulsion system has already cut fuel use by up to 15 percent on live vessels — while most of the maritime industry is still running pilots that never make it past the trial stage. That gap between proven results and stalled adoption is the real story of maritime AI right now: it's not a question of whether the technology works on the water, but which operators are actually scaling it, and why so many others aren't.

Key takeaways

• A class-approved AI propulsion system is already delivering up to 15% fuel savings on commercial vessels, per the vendor.
• A 2025 industry survey found 81% of maritime organizations are running AI pilots, but only 11% have formal policies to scale one into production.
• The pattern in the working examples: a narrow, measurable problem, reused sensor data, and a human still making the final call.

A fuel-saving AI system already in commercial use

Frugal Technologies, a Danish maritime technology company, offers a class-approved AI control system called Frugal Propulsion, built to cut fuel consumption and emissions on large vessels. The company reports that three operators — Uni-Tankers, Christiania Shipping, and Royal Arctic Line — have reduced fuel consumption by up to 15% using the system (AJOT, June 2024).

In 2024, Frugal Technologies extended the platform with predictive maintenance. The upgraded system compares a vessel's current engine condition against its baseline performance using historical operating data, tracking temperature, pressure, and vibration deviations so engineering teams can catch developing problems before they cause a failure or an unplanned port call. The same sensor data that optimizes fuel burn is reused to flag anomalies — one data pipeline doing two jobs instead of two separate projects competing for budget and crew attention. The system has also been certified by Bureau Veritas and Lloyd's Register as an Overridable Power Limitation System, meaning it counts toward EEXI emissions compliance, turning a cost-saving tool into a regulatory one as well.

A similar pattern shows up in route planning. Ferry operator Stena Line piloted an AI model, built with Hitachi, on the Stena Scandinavica, simulating weather, currents, water depth, and speed-through-water scenarios to recommend the most fuel-efficient way to run a route, with the captain and officers still making the final call (BestPractice.ai case study). The crew adjusted the model's recommendations trip after trip, and it reportedly improved quickly — a detail worth sitting with, since the vessels getting the most value out of maritime AI tend to be the ones where crews stay in the loop and keep correcting it, rather than switching it on and walking away. No fuel-saving results had been published for this pilot at the time it was reported.

Why most maritime AI projects stall before they scale

These working examples sit against a less flattering industry backdrop. A 2025 study by maritime research firm Thetius, conducted with Marcura, surveyed more than 130 maritime professionals and found that 81% of maritime organizations are running AI pilot projects — but only 11% have formal policies in place to guide scaling a pilot into a real operational system. Just 17% say their AI decision-making processes are transparent to the people who rely on them (Cyprus Shipping News, August 2025).

The optimism in that survey is real: 82% of respondents said they're positive about AI's role in maritime operations, and 97% see clear value in using it to cut manual workflow inefficiencies. But 37% have personally witnessed an AI failure, and 38% named inadequate training as the single biggest barrier to scaling AI past the pilot stage, ahead of cost, data quality, or technology maturity. Nearly a quarter of respondents said vendor promises tend to outrun what the tools deliver in practice.

Perhaps the clearest signal in the data: 70% of maritime professionals want AI to recommend an action while a human makes the final decision, and 66% are specifically worried that leaning on AI too heavily will erode their teams' own judgment over time. That's not resistance to AI — it's a preference for the same model that Frugal Propulsion and the Stena Line pilot both use: AI that narrows the decision space and flags what matters, with a qualified person still in command.

What this means for maritime, logistics, and industrial operators

The operators getting real results from maritime AI right now aren't chasing the most ambitious pilot. They're picking a narrow, measurable problem — fuel burn, engine wear, route efficiency — reusing the data they already collect, keeping crews in the decision loop, and building the internal process to take a working pilot and scale it fleetwide. That last part is where the Thetius/Marcura survey says most of the industry is still falling short, and it's the focus of our own AI adoption roadmap for marine and offshore operators.

For operators evaluating where to start, three questions are worth asking: What single operational cost — fuel, maintenance, demurrage — would a working pilot need to move to justify the investment? Who on the crew or shoreside team owns reviewing and correcting the AI's recommendations? And what's the actual plan, not just the hope, for turning a successful pilot into a scaled, governed system? We've covered the broader pattern of what's working across the industry in From Pilot to Payoff: What's Actually Working in Maritime AI Adoption Right Now.

Next step

If your fleet is sitting on a fuel-optimization or predictive-maintenance pilot that hasn't scaled, that's a solvable process problem, not a technology one. See how we approach it on our AI adoption services page, or get in touch to talk through where AI could move the needle for your operation.