Operational AI inherits every flaw already in the data

Predictive maintenance, crew scheduling, and route optimisation are some of the most promising AI use cases in aviation and transport, and also some of the most exposed to bad underlying data.

These datasets were built for operations: keeping aircraft flying, crews legal, and networks running. They were not built with analytics or model training in mind, and it shows the moment you try to use them for either.

A predictive maintenance model trained on maintenance logs with inconsistent fault coding, or a crew scheduling optimisation built on rosters with undocumented manual overrides, will produce confident recommendations that quietly degrade operational reliability rather than improving it.

Readiness has to be assessed before the AI use case is chosen, not after the model is already in production and something goes wrong.