How are you choosing the order decoupling point when you’ve promised 2‑day delivery across the U.S? Using SCOR and postponement, I’m inclined to push to 4 regional DCs and pull from there, but SKU‑level forecast error >35% is crushing turns and inflating safety stock. Is there a practical threshold (ABC class, CV, or service level target) you use to flip between make‑to‑stock and make‑to‑order for long‑tail SKUs?
I’d segment by CV and node hit rate: keep A SKUs with CV≤0.8 and ≥3 hits/week/node forward-deployed to 4 DCs, push the rest to 1–2 DCs with DC kitting and 2‑day air as the safety valve — stocking the long tail everywhere is like trying to carry every ice cream flavor at every stand. For borderliners, use a simple rule like “flip at CV≈0.8 or when 95% service explodes safety stock,” then true-up monthly. What’s your per‑node hit rate on the C‑class?
One thing that helped us was a simple “days-of-cover gate” per SKU family: if the inventory to hit about 96% fill at a regional node exceeded about 8 days, we kept it in 1–2 hubs with light postponement (late labeling/kitting) and used 2-day air only for exceptions; otherwise we staged to all four DCs. We tune that cutoff by cube and zone mix. Do you have late pickup windows to make the hub path work for the long tail?
Given ‘forecast error >35%’, I use a breakeven: if the extra safety stock from four DCs (square‑root effect) times your carrying rate exceeds the expected cost to expedite a small tail from one or two hubs to meet the 2‑day SLA, keep it centralized and postpone at the hub. I’d rerun OWD by ZIP quarterly and only push forward the SKUs/families where that flips — does that change the case for 4 regional DCs?