Practical, evidence-led writing on decisioning, production machine learning and what actually happens when models meet operations.
UK payment fraud losses reached £1.28 billion in 2025. The systems built to stop it are, in many organisations, destroying more value than the fraud they prevent.
Some decisions cannot be A/B tested into existence. Simulation is what turns "we think this will work" into something a risk function can authorise.
Most customer scoring systems answer one question: who is most valuable. They skip the prior question, which is whether this is worth doing at all.
Every adaptive decisioning system depends on knowing what worked. Get attribution wrong and the optimiser learns confidently from a biased signal.
Most contact strategies are set by judgement and reviewed quarterly. There is a better way, and the mathematics behind it was published in 1933.
Gartner found that fewer than half of AI prototypes reach production. The reasons are rarely modelling failures. They are questions nobody asked early enough.