How to prepare S&OP for AI: the foundations that make demand sensing work

Why most S&OP AI pilots stall

The typical pilot connects a forecasting model to historical orders and reports a better accuracy figure. Then nothing changes: planners keep overriding, Finance keeps its own plan, and the executive meeting still argues about which number is right. The model was not the problem. The cycle was.

Where forecasts are built from distributor orders rather than demand, where excess and obsolete inventory absorbs working capital while availability suffers, and where decision rights between commercial ambition and supply reality are unclear, no model fixes the outcome. These are operating-model failures, and they are where OREDJA starts.

The four foundations

Designed and run in this order, they make every later AI use case measurable.

• One S&OP cycle: demand review, supply review, reconciliation with Finance, executive S&OP, on a fixed monthly calendar with named owners.
• One set of numbers: a statistical baseline enriched with market intelligence, reconciled with the financial plan, so that the executive meeting decides rather than debates.
• Decision rights and inventory policy: who arbitrates between service and cost, targets by product family and market, modal choices set deliberately.
• A planning organisation with the competence to sustain it: profiles, training, a KPI framework for planners, and planning parameters maintained in the ERP.

The AI use cases that follow

Once the cycle runs, AI earns its place in three places. Forecast exception and override management: the model proposes, flags the items where human overrides have historically destroyed accuracy, and leaves the planner the rest. Demand sensing on short horizons: point-of-sale, orders and promotions read daily to adjust the near-term plan. Scenario preparation for S&OP: the supply and financial consequences of two or three demand scenarios prepared before the meeting, so the executives decide in the meeting rather than ask for another analysis.

Each use case has an owner (the demand planner, the supply planner, the S&OP lead), a threshold above which a person decides, and a KPI the business already tracks: forecast accuracy by family, service level, days of inventory, working capital.

What a documented transformation looks like

For a global luxury group, OREDJA redesigned the regional network with a supply chain modelling tool, then ran the planning routines that followed. Twelve months after go-live, delivery lead time to boutiques and wholesale partners had fallen by three days and logistics cost by seven per cent on the group’s own baseline. The AI layer came after the routines, not instead of them.