Retail replenishment guidance that stores follow

One target does not fit 34,000 items

Most retail chains run replenishment off a single weeks-of-supply target, and most retail replenishment guidance stops right there: three weeks, chain-wide, sometimes with a department-level override, applied to 34,000 active items across 62 stores. It is simple to explain, simple to audit, and wrong at both ends of the range simultaneously.

Three weeks of supply on a produce item that turns in four days is spoilage. Three weeks on a slow center-store item that sells two units a month is a nine-month buy dressed up as a policy. The average is fine. The tails are where the money goes, and a single target guarantees you are wrong in both tails at once.

Here is what the departments actually run at our 62-store operator.

Produce at 1.2 weeks and center store at 6.3 are not two departments with different discipline. They are two departments with genuinely different demand shapes, and the replenishment guidance should say so explicitly instead of letting store teams discover it by trial.

Segment by demand shape, not by department

Department is a convenient proxy and a poor one, because every department contains both fast predictable items and slow erratic ones. Segment on the two properties that actually determine the right policy: rate of sale, and demand variability.

The last row is the honest one and it is usually left off these tables. When a supplier's minimum order is 40 cases and a store sells 12 a cycle, the target is not a decision anybody at the store is making. Labelling those items explicitly stops store teams from being graded against a number they cannot hit, and it routes the problem to purchasing where it belongs.

The variability split matters more than most guidance admits. Two items selling 20 units a week, one steady and one swinging between 5 and 40, need different safety stock even though their rate of sale is identical. Setting one target for both means the steady item carries excess and the erratic one goes out of stock, which is the worst available combination.

Safety stock, sized to the actual risk

Safety stock is where replenishment guidance either earns its keep or quietly becomes overstock policy. The usual practice is a flat percentage uplift, which is the same error as a flat weeks-of-supply target.

Size it against lead time and variability:

safety stock = Z x sigma_demand x sqrt(lead time in weeks)

sigma_demand is the standard deviation of weekly demand for that item at that store. Z is the service-level factor: about 1.65 for 95% availability, 2.05 for 98%. The square root of lead time is the part people drop, and dropping it makes long-lead items dangerously under-covered.

Two practical adjustments to the textbook formula.

Set Z by item importance, not uniformly. Chasing 98% availability on the entire tail is how safety stock becomes the overstock in the weeks-of-supply chart above. Reserve the high service levels for items where the out-of-stock actually costs a trip: the traffic drivers and the destination items. A slow tail item at 95% is fine.

Floor it at minimum presentation. Some items need a facing's worth of stock regardless of the math, because an item with two units on a shelf built for eight reads as out of stock to the shopper even though the system says it is in stock. This is the one case where the number should lose to merchandising judgment.

Review cycle is a lever, and it is free

Increasing review frequency does the same job as safety stock and costs inventory instead of ordering nothing. An item reviewed daily needs materially less cover than the same item reviewed weekly, because the exposure window between review points is shorter.

That makes review cycle the first lever to pull when a category is carrying too much. Moving fast items from twice-weekly to daily review typically removes more inventory than any target change, and it does it without any increased out-of-stock risk. The constraint is delivery schedule: reviewing daily is pointless if the supplier delivers twice a week, which is why review cycle should be set against the delivery calendar rather than uniformly.

Where a supplier delivers weekly, the review cycle is weekly no matter what the policy says, and the target has to absorb the full week plus lead time plus variability. This is a real cost of a low-frequency delivery schedule and it belongs in the landed-cost comparison when choosing a supplier, where a cheaper per-unit price on a weekly delivery often loses to a slightly higher price on twice-weekly.

Store clusters, because 62 stores are not one store

A single set of targets across every store assumes the stores are interchangeable, and they are not. A small-format store with 8 linear feet of dairy and a full-size store with 40 have different presentation minimums, different rates of sale, and different sensitivity to a supplier minimum.

Cluster on the two properties that change the replenishment answer:

The small-format cluster is where uniform targets do the most damage. Those 12 stores hit supplier minimums constantly, so a large share of their range is structurally over-covered and no store-level discipline changes it. Flagging those items as minimum-bound rather than as over-ordered is the difference between an actionable list and a list the store learns to ignore.

The seasonal cluster breaks flat targets in the other direction. Five stores whose demand triples between seasons will be over-covered for eight months and out of stock for four if a constant weeks-of-supply target is applied, because the target is calibrated to an average that never occurs.

Retail replenishment guidance stores will actually follow

The best-specified replenishment policy in the world fails if it arrives as a sixty-page document. Three properties separate guidance that gets followed from guidance that gets ignored.

It arrives as an exception list, not a policy. Store teams should not be computing weeks of supply. They should receive the twenty lines that are outside band this cycle, in order of how far outside.

It explains the override path. Every policy meets a case it did not anticipate. If the only options are comply or work around it, teams work around it and the data quietly stops meaning anything. A documented override with a reason code preserves both the exception and the signal.

It is graded on outcomes the store controls. Grading a store on weeks of supply for minimum-bound items is grading them on the supplier's minimum. Split those out before anyone's performance is measured.

Lead time is the input everyone stales

Every formula above takes lead time as a parameter, and lead time is almost always a static field entered when the supplier was set up and never revisited. That single stale number silently corrupts safety stock across the entire range a supplier serves.

Measure it instead of storing it. The realized lead time is the gap between order transmission and receipt, and it has a distribution rather than a value. What matters for safety stock is not the mean but the upper tail, because the stockout happens on the slow deliveries, not the average ones.

Harbor's stated three days against an eleven-day 90th percentile is the whole story of their service problem expressed as a replenishment input. Every safety stock calculation on their items was sized for a supplier that does not exist. Recomputing against realized lead time raised cover on their six core items and removed most of the out-of-stocks, which is a replenishment fix for what looked like a purely supplier problem.

The general rule: use the realized 90th percentile, and refresh it quarterly. A supplier whose tail is widening is degrading months before their OTIF crosses a threshold.

Doing this in Scout

The replenishment math above is not difficult. What is difficult is keeping it current across 34,000 items and 62 stores, when rate of sale and variability move continuously and the item classes shift under them. Most chains set the classes once, at implementation, and never revisit them, which is why the guidance drifts away from reality over about eighteen months.

Scout recomputes the class assignment from the trailing demand rather than holding it as a static field. An item that moves from medium to fast-stable changes its target and review cadence without anyone rebuilding a spreadsheet, and the minimum-bound items are identified automatically by comparing the supplier minimum against the store's rate of sale rather than being tagged by hand.

The output is the exception list: the lines outside band this cycle, per store, with the reason (over target, under target, or minimum-bound) attached. That is the same view the over-ordering review reads, which means the over-ordered lines and the under-covered lines surface in one place rather than in two systems that disagree.

Scout computes and monitors the targets. It does not generate purchase orders, hold the order guide, or send anything to a supplier: the ordering system stays where it is.

Reviewing the guidance itself

Retail replenishment guidance decays. Rate of sale drifts, suppliers change minimums and delivery schedules, store clusters shift as formats are remodelled, and the class assignments quietly stop describing the range.

Put a quarterly review on the parameters, not just on the exceptions. Three questions cover it: which items changed class this quarter, which suppliers' realized lead-time distributions moved, and which store-items are newly minimum-bound. All three are computable, none require judgment to produce, and together they catch the drift that otherwise shows up eighteen months later as an inventory problem nobody can explain.

Summary

• A single chain-wide weeks-of-supply target is wrong in both tails at once. Segment on rate of sale and demand variability, not department.
• Size safety stock against lead time and variability with the square-root term intact, set the service factor by item importance rather than uniformly, and floor it at minimum presentation.
• Review frequency substitutes for safety stock and costs nothing, so it is the first lever when a category is carrying too much, bounded by the delivery schedule.

Further reading: preventing over-ordering covers the review that catches orders exceeding these targets, and reduce overstock covers clearing what has already accumulated.