Repetition in a stable context makes behaviour more automatic.
More quotesNobody ever got promoted for announcing that last month’s average works quite well.
Give the same number a proprietary name, a pulsing dashboard and a diagram resembling the London Underground after a nervous breakdown, and suddenly it is transformation.
Model A beats the basic rule by 0.2 per cent. The case study calls this “a step-change in predictive intelligence”. It does not call it “possibly rounding”. That would spoil the launch energy.
George Box’s line about falling in love with models is funny because love rewrites the excuses. A success proves the model is brilliant. A failure proves the data was unusual, implementation was weak or reality behaved unprofessionally. Any method that cannot lose has stopped being analysis and started collecting tithes.
Complexity also changes how failure looks. A simple average looks stupid, so people check it. An opaque model looks clever, so its error can travel through three meetings wearing a lanyard. When an input silently becomes zero, authority buys the mistake time.
Of course complicated models sometimes matter. Aircraft, epidemics and supply networks contain interactions that “same as Tuesday” will miss rather badly. But complexity should pay rent: a material improvement, a necessary representation of the system, a better account of uncertainty. Awe is not legal tender.
I once heard a forecast defended with, “You’d need to understand the architecture.” No, mate. I need to understand why the expensive answer is worse than a ruler and a pencil.
When the specialist leaves, Model A goes into maintenance. The team quietly returns to last month’s average.
The dashboard still says strategic.