4-month POC with a Top 10 Global Carrier — what we measured.
The Fuelture Tech engine was tested against the live procurement workflow of a leading global container carrier, across six vessels and a four-month evaluation window. The objective: prove that an autonomous decision system could outperform an experienced manual procurement team on real voyages.
A controlled, like-for-like comparison.
For each vessel and each procurement event during the window, the carrier's manual decision was logged. In parallel, the Fuelture Tech engine generated its own recommendation using only the data available at the moment of decision.
Outcomes were compared on three axes — realized procurement cost, forecast accuracy across multiple horizons, and the operational feasibility of the recommended action. Results were reviewed jointly with the carrier's bunker team at the end of each month.
From data ingestion to validated outcome.
Four months, six vessels, dozens of refueling events. A live, side-by-side test of human and machine decision-making.
Data ingestion & engine calibration
Vessel speed-consumption profiles, historical bunker prices, voyage schedules, and noon reports were ingested. The forecast and decision engines were calibrated against the carrier's reference dataset.
Shadow mode
For every real procurement event, the engine produced a parallel recommendation. The carrier's team continued to execute their own decisions; recommendations were logged for retrospective comparison.
Joint review & blind test
Mid-pilot review: results presented to the carrier's bunker team. Selected upcoming procurement events were structured as blind tests against the engine.
Validation & reporting
Outcomes consolidated across all six vessels. Final validation report covered cost saving, forecast accuracy by horizon, and qualitative feedback from the carrier's procurement leadership.
Measured against the manual baseline.
Three results stood out to the carrier's leadership: the magnitude of per-voyage savings, the precision held across long horizons, and the consistency of the engine's recommendations across vessels.
Forecast accuracy by horizon
| Horizon | Accuracy (1 − MAPE) |
|---|---|
| 1-day ahead | 98.94% |
| 2-day ahead | 98.04% |
| 4-day ahead | 97.51% |
| 6-day ahead | 97.01% |
Accuracy is reported as 1 − MAPE (Mean Absolute Percentage Error) on observed VLSFO prices.
The engine's recommendations were consistently within the band of what an experienced trader would have considered — and frequently outside the band of what they actually executed.
— Internal Pilot Validation Notes
Implications for an SME fleet.
The pilot was conducted on the workflow of a global carrier — but the same economics scale down. For a mid-sized fleet running 50 voyages a year, a $40K-per-voyage uplift translates directly into seven-figure annual savings, before any downstream carbon liability is accounted for.
Cost
Direct, recurring procurement savings — captured every refueling event, not once a year.
Risk
Forecast precision at long horizons reduces the price-at-risk on long voyages and forward charters.
Carbon
Every saved tonne of fuel is also a saved tonne of CO₂ — and a smaller carbon liability under the IMO trajectory.
The carrier in this pilot has asked to remain unnamed. We are an early-stage venture — recently incorporated, with the technology stack validated and our first commercial pilots in development. If you operate a fleet and would like to run a comparable validation on your own routes, we would welcome the conversation.