Research that changed what was worth building

The situation

A large ports and logistics group wanted to understand where AI could help across maritime operations. There was an appetite to move quickly and a list of candidate use cases waiting to be built.

What everyone thought the problem was

That the operation contained obvious inefficiency, and that the job was to find enough AI use cases to attack it.

What I found

I went to the operators. That meant research with working container-ship captains — recruited through maritime communities when direct access proved difficult — alongside commercial stakeholders on shore.

What looked like inefficiency from the outside was frequently rational from the inside. Vessel speed, ETA behaviour and berth availability were shaped by incentives and information asymmetry between parties who did not share the same goals. Several proposed use cases would have automated a behaviour that existed for a commercial reason, and would have been ignored or worked around.

The decision

Score the candidate use cases honestly against desirability, viability, feasibility and adoption risk — and recommend cutting the ones that ignored the incentive structure, however technically feasible they were.

What we made

Current-state maps, a use-case catalogue with scoring, executive one-pagers, and prototypes for the themes that survived — voyage speed intelligence, container repositioning, and ETA and berthing decision support — including the interface for explaining a recommendation rather than merely issuing one.

What changed

The roadmap got shorter and more defensible. Weak use cases were stopped before they reached build, and the surviving ones were shaped around how decisions actually get made between parties with different incentives.

What this proves

A technically possible AI feature can still be a bad product if it ignores incentives.