Decoupling Points and a 48-Hour Promise at a Configurable Pump Maker

One policy cannot serve a catalogue whose demand shape ranges from weekly to less than quarterly. Build-to-order across 1,400 SKUs forces the factory to carry the variability of the tail and the volume of the head in the same queue - that is where the WIP and the expediting come from.

Three policies:

1. Top 60 runners → make to stock / configure to order from stocked platform subassemblies. This is the only way a 48-hour promise is credible. Stocking the configurable module rather than the finished SKU gives risk pooling across variants: one subassembly buffer covers several orderable SKUs.
2. The ~12% of SKUs carrying 68% of revenue but with option variability → assemble/configure to order from stocked subassemblies, quoted in days not weeks. The decoupling point sits at the subassembly, not the finished unit.
3. The tail (<3 orders/year) → make to order, quoted honestly at 6–10 weeks. Trying to hold this demand in stock produces the 62-day-type inventory that never turns; slow movers should carry the long lead time, not the buffer.

Why: the decoupling point should sit where forecast error is lowest and pooling is highest relative to how much differentiation the customer pays for. For this product that is the platform subassembly. Segment by demand variability and margin, not by revenue alone - a high-revenue, high-variability SKU behaves like a tail item for planning purposes.

Buffer the highest common level the customer will tolerate, then differentiate service levels.

• Content: platform castings/motors dressed to a common configuration, plus a small finished-goods set for the 60 distributor SKUs. Buffering at the platform level means a single unit of safety stock serves multiple orderable variants; buffering at finished-SKU level multiplies it.
• Sizing: use demand variability and replenishment lead time per segment, not a flat weeks-of-supply target. The top 60 get a service level around 97–98% because the contractual promise depends on it; the tail gets 85% and a quoted lead time that carries the risk. Allocating inventory proportional to revenue would over-stock the mid-tail and under-protect the promise.
• The 26-week component is a capacity problem, not an ordering problem. At 26 weeks the forecast is close to useless - reserve supplier capacity at platform level, consider consignment or supplier-held stock, and keep a strategic buffer of the common casting. Do not attempt to forecast 1,400 SKUs half a year out.
• Working capital check: the platform buffer should reduce total inventory versus today's WIP, because it replaces order-by-order accumulation with a planned, pooled position. If it does not, the buffer is being built at the wrong level.

A service promise is a liability; the contract has to price it.

• Forecast commitment: a rolling 12-month forecast with a frozen 8-week window and a stated accuracy obligation. Without it, the distributor gets 98% service on a forecast the manufacturer carries alone.
• Volume commitment or take-or-pay: minimum annual volume, and liability for inventory bought against their forecast if they walk.
• Terms that reflect the inventory: the 48-hour promise is funded by stock sitting in the manufacturer's warehouse; either the price reflects it or the stock is consigned/VMI at the distributor with agreed holding cost.
• Exclusions and recovery: what happens on allocation, how expedites are charged, and the service-credit ceiling.
• Internal prerequisite: confirm the platform buffer can actually deliver 48 hours from order entry through pick-pack-ship - measure it on the top 60 for four weeks before signing. Committing to a promise that has not been demonstrated is how OTIF 71% becomes a penalty clause.