Predictions and visibility are inputs. Decisions are the output.

Supply chain technology stacks into three layers. Predictions — what do we think the future looks like? Visibility — what's actually happening right now? Decisions — what should we do about it? The first two have moved forward dramatically. The third has barely moved at all.

ML-driven hierarchical forecasting beats the naive baselines we shipped in 2016 by 25 to 40 percent. Demand sensing captures granular signal that didn't exist a decade ago — point-of-sale telemetry, weather, search trends, social chatter. Control towers stream events in near-real-time across networks that used to refresh nightly. Credit where credit is due: the prediction layer and the visibility layer have earned the investment they've absorbed.

The decision layer is the layer that converts those better inputs into a committed action — release this PO, allocate this customer, hold this safety stock, reserve this lane. It is built on the same architecture it was built on twenty years ago: sequential five-stage planning, single-future master-data assumptions, single-objective optimization, no memory of yesterday's commits, no measurement of today's policy.

Better forecasts do not produce better decisions when the decision architecture is sequential, single-future, master-data-driven, and amnesiac. That sentence is the diagnosis the rest of this book is built on.