The Protected Constraint

The adoption claims and the measured outcomes are irreconcilable. The regulatory ground is firmer than most operators realise, and it is about to get firmer still. Three things are true at once in this sector, and holding all three is the beginning of a sensible plan. The technology genuinely predicts better than what came before. The operational benefit of that prediction is undemonstrated outside vendor case studies. The constraints that govern what an operator may do with a prediction are legally enforceable, physically real and, in one important respect, weaker in Great Britain than in the European Union. The Protected Constraint · Operations Artificial intelligence is good at predicting what a logistics operation will face and poor at knowing what it is forbidden to do about it. The strongest deployments therefore keep three things apart, which are the forecast, the operational fact and the instruction, and they hold the hard limits on driving hours, load security, site safety and vehicle condition outside the model where a person owns them and a test proves they hold. The recommendation in this paper is to build the constraint library before buying the optimiser, because a recommendation that breaches a limit is worse than no recommendation, and because the limits are the part a regulator will ask about.