Stop betting blindly on inventory

In consumer goods, inventory going wrong is often not a bad market call. It is a design flaw in how you count stock and how the feedback loop works. When a brand is out of the hits and buried in the long tail at the same time, a spreadsheet and a static formula will usually make the swing worse. That is the bullwhip effect. A precise replenishment system has to deal with four physical facts.

1. Look through the static book: effective inventory

A warehouse number does not describe what can actually sell. Replenishment should use what will be sellable in the near future:

Effective Inventory = (on-hand - locked orders) + inbound purchase + inbound returns + estimated unsettled returns

Available is physical stock minus paid orders still waiting to ship. The inbound pipeline is purchase in transit plus goods already sent out and now coming back. Estimated unsettled returns are the share of orders still inside the return window that, by probability, will come back. That is future supply. Add it back now, or you chase a fake stockout. The return rate has to be checked against SPU history, category baselines, and the brand’s longer record.

2. Filter noise: return latency

Using short-window net sales to get average daily sales is how replenishment gets the story wrong. Returns lag. Returns you see in seven days belong to orders from an earlier cycle. When sales slow down, the short-term return rate can even print above 100% and wreck net sales on paper.

Use a 30-day window to smooth that lag and get a stable ADS_30d. Use a 7-day window only to catch a real spike. Only when ADS_7d is more than 1.2 × ADS_30d do you mix them in. The weights need testing:

ADS = 0.7 × ADS_30d + 0.3 × ADS_7d

3. SPU totals, SKU shares

Forecasting at SKU level falls into small-sample noise. If one size barely sold, its velocity is meaningless. Replenish from that and you stock out of smalls while large sits.

Do it in two steps. Forecast demand for the whole style from net daily sales and the target cover period. Then split that total by the category’s long-run size mix, and subtract each SKU’s own effective inventory. Decide volume at SPU. Allocate at SKU with large-number proportions.

4. MOQ is a real constraint

The forecast is a continuous number. The factory ships in discrete minimum order quantities. The hard case is the mid-tail style whose forecast sits under MOQ.

If the suggestion is already at or above MOQ, place that quantity. If it is under MOQ but at least half, round up to one MOQ and check how long it will take to sell through, plus whether you can cover cost. If it is under half, skip this round or wait to combine with another order.

After rounding up, if the sell-through would run past a safe selling window, do not produce it on its own. Draw a cost floor: at one MOQ, the share you can expect to sell (from history) should cover tooling and materials. The rest is profit.

5. Close the loop

Every replenishment, snapshot the suggestion, any manual change, and the inputs (daily net sales, the return-rate baseline). Fifteen days after arrival, check whether you stocked out in the cover window. If the math was too conservative, raise the safety stock a little. At forty-five days, compare actual turns with the forecast. If stock piled up, lower the safety stock and recalibrate the return rate.

Supply-chain quality in physical goods is whether the system can see lag, statistics, and factory minimums as they actually are. See real available stock. Strip return noise. Split the details with large numbers. Know the MOQ floor. Then replenishment stops being a daily scramble and becomes a rule that runs.