The client — a €9 billion industrial group with 14 plants and 4,000 SKUs — ran planning the way most manufacturers do: a monthly cycle, a weekly exception meeting, and a monthly scramble when demand moved faster than the plan.
The diagnosis was not that forecasts were bad. It was that the plan was a snapshot in a world that had become a stream. By the time the monthly plan was approved, the assumptions underneath it had moved.
Over seven months we rebuilt planning as a continuous loop: a demand signal refreshed daily, a constraint model that understood plant changeover economics, and an agent layer that proposed replans whenever the cost of the current plan exceeded the cost of changing it. Planners moved from producing the plan to approving, amending, and owning the exceptions.
The results after four quarters: expedited freight down 31 per cent, inventory down 12 per cent without service loss, and planner time on plan production down by roughly two-thirds — most of which was reinvested in supplier collaboration.
Two things made the difference. The change was scoped as a planning redesign, not an AI deployment, so the operating cadence changed with the tool. And the team kept a human approval gate on any replan above a defined cost threshold for the first two quarters, which bought the credibility needed to widen the autonomy later.
