The numbers, then the story.

The brief that arrived was a mobile app. The brief we agreed, four weeks later, was a tracking system with a mobile client, because discovery found that the scanning behaviour at the loading bay was the constraint. No interface improvement upstream of it would survive contact with a ramp agent wearing gloves in the rain.

So the first thing built was not a screen. It was the inventory model: one identity per device, one event log, and a reconciliation pass that could accept the same scan arriving twice from two operators and resolve it without a human. Everything visible rests on that, and getting it wrong would have been a rewrite rather than a fix.

The web platform handles inventory, inspections and reporting for the operations teams. The mobile client handles the bay: offline-first, because airside connectivity is not a thing you design around optimistically, and a scan that fails silently is worse than no scan at all. Inspections carry photographs and a signature, and both are attached to the device's log rather than to a report.

What moved the operation was not any single feature. It was that the turnaround clock became measurable at all. Once each device had an event log the delays were attributable, and most of them turned out to be concentrated at two handover points nobody had instrumented.