01

The decision is relational

Range actions operate across SKU, branch, customer segment, supplier term and time. A local delist can release stock while destroying a mandatory range, substitution path or supplier-funded contribution elsewhere.

The evidence model must preserve those relationships instead of reducing the decision to velocity and stock cover.

02

Keep value components separate

Working-capital release, contribution, funding and service effects have different timing and recognition rules. Combining them too early creates double counting and hides trade-offs.

A decision ledger should show each effect, the shared exposure key and the outcome window before presenting an aggregate decision objective.

03

Optimise a feasible portfolio

Transfer lanes, pack sizes, pricing floors, mandatory lines, capacity and authority define the feasible set. Demand and sell-through forecasts inform the value of alternatives; they do not override those constraints.

The output is an atomic action against one named SKU-branch exposure, or an explicit hold with a review trigger—not a generic instruction to improve the range.

Sources

Further evidence

Google OR-Tools optimisation guidesMIT CTL research