Writings: Technology
Why route optimisation projects succeed
The ones that work don't have better solvers. They treat the first quarter as constraint discovery, and they scope against a number someone is already accountable for.
What a camera can and cannot see
Before any computer vision project, write down what the camera physically cannot observe — and check whether the thing you care about is on that list. It often is.
The last mile is a people problem
Most last-mile inefficiency isn't in the routing. It's in the twenty minutes before the van leaves and the things drivers know that the system doesn't.
Pilot purgatory
Enterprises don't struggle to start AI pilots. They struggle to end them. The pilot that neither fails nor scales is the default outcome, and it's a structural problem, not a technical one.
What P&L owners actually want from optimisation
They are not buying a percentage. They are buying the ability to explain a number to someone above them — which is a different product than the one most vendors pitch.
Counting things is not understanding them
A vision system that counts correctly and explains nothing is a more expensive clipboard. The interesting output isn't the count — it's the variance nobody could see before.
Notes on teaching executives
Senior people don't need the concept explained. They need permission to admit which part they never understood, and a worked example from their own P&L.