Store Replenishment

(s, S) policy · Continuous review

A retail store faces stochastic daily demand and is replenished from a regional DC after a lead time \(L\). A continuous-review (s, S) policy triggers an order whenever on-hand inventory drops to the reorder point \(s\), bringing inventory back up to the order-up-to level \(S\). The policy balances stockout cost against carrying cost and ordering frequency. Foundational papers: Scarf (1960) optimality of (s, S); Veinott & Wagner (1965) computational scheme; Federgruen & Zipkin (1984) efficient algorithm.

Why it matters

Capital tied up in store inventory · documented optimisation lift

$870B
US retail inventory outstanding at month-end 2024 — the working capital that store-level replenishment policies govern.
Source: US Census MTIS, end-2024.
5–15%
Documented inventory-cost reduction when replacing reorder-point heuristics with optimised (s, S) parameters across SKUs.
Source: Industry benchmarks, retail OR practice.
Scarf
Scarf (1960) proved (s, S) is optimal for the many-period inventory problem with K fixed ordering cost — the foundational result that justifies the policy class.
Scarf (1960), Math. Methods in Social Sciences.
1–2 / wk
Typical retail SKU sees one to two deliveries per week from the DC; the (s, S) parameters determine exactly when an order fires and how much arrives.
Industry practice; observed delivery cadences.

Where the decision sits

Store fulfillment loop · demand forecast feeds policy parameters

Store replenishment is the daily heartbeat of retail inventory: the forecast tells us how fast the SKU sells, the safety stock buffers forecast error and lead-time variability, the (s, S) policy converts those numbers into a clean automated trigger, and the order eventually arrives at the back room. Upstream the multi-echelon network coordinates DC inventory with the store; here we focus on the single store-DC link.

Forecastdemand \(\mu, \sigma\)
Safety stockservice level
Compute (s, S)policy parameters
Trigger ordersDC ships to store

Problem & formulation

Continuous-review (s, S) under Normal demand and constant lead time

OR family
Stochastic Inventory Control
Policy class
Continuous-review (s, S)
Solver realism
★★ Closed-form heuristic
Reference
Hadley & Whitin (1963)

Parameters

SymbolMeaningUnit
\(\mu\)Mean per-period (daily) demandunits / day
\(\sigma\)Standard deviation of per-period demandunits / day
\(L\)Constant replenishment lead timedays
\(K\)Fixed ordering cost per order placed$ / order
\(h\)Holding (carrying) cost per unit per period$ / unit-day
\(p\)Stockout penalty per unit short$ / unit
\(c\)Unit purchase cost$ / unit

Demand model

Per-period demand and lead-time demand are Normal:

$$D \sim \mathcal{N}(\mu,\, \sigma^2) \qquad D_L \sim \mathcal{N}\bigl(\mu L,\; \sigma^2 L\bigr)$$

Independence of daily demands gives lead-time variance \(\sigma^2 L\); standard deviation of lead-time demand is \(\sigma\sqrt{L}\).

Decision variables

SymbolMeaningDomain
\(s\)Reorder point: place order when on-hand drops to \(s\)units
\(S\)Order-up-to level: each order brings inventory back to \(S\)units, \(S > s\)

Order quantity at each replenishment is \(Q = S - I_{\text{on-hand}}\), where \(I_{\text{on-hand}}\) is the inventory at the trigger instant (typically \(\approx s\)).

Reorder point (cycle-service-level form)

Set \(s\) so that the probability of a stockout during the lead time is at most \(1 - \alpha\), where \(\alpha = p / (p + h)\) is the cost-balance service level (Hadley & Whitin, 1963):

$$s \;=\; \mu L \;+\; z_\alpha \,\sigma\,\sqrt{L} \qquad \text{where} \qquad z_\alpha \;=\; \Phi^{-1}\!\left(\frac{p}{p + h}\right)$$

First term is expected lead-time demand; second term is the safety stock that buffers lead-time variability.

Order-up-to level via EOQ

A practical heuristic is \(S = s + Q_{\text{EOQ}}\) where the economic order quantity balances fixed cost \(K\) against per-period holding cost \(h\):

$$Q_{\text{EOQ}} \;=\; \sqrt{\frac{2\,K\,\mu}{h}} \qquad S \;=\; s \;+\; Q_{\text{EOQ}}$$

Approximate cost per unit time

A standard cost decomposition (Hadley & Whitin, 1963; Silver, Pyke & Peterson, 1998):

$$C(s, S) \;=\; \underbrace{\frac{K\,\mu}{S - s}}_{\text{ordering}} \;+\; \underbrace{h\!\left(\frac{S-s}{2} + s - \mu L\right)}_{\text{holding}} \;+\; \underbrace{p\,\mathbb{E}\!\left[(D_L - s)^+\right]}_{\text{stockout}}$$

The expected shortfall \(\mathbb{E}[(D_L - s)^+] = \sigma\sqrt{L}\,L(z)\), where \(L(z) = \phi(z) - z(1 - \Phi(z))\) is the standard Normal loss function and \(z = (s - \mu L)/(\sigma\sqrt{L})\).

Interactive solver

Closed-form (s, S) heuristic + Monte Carlo simulation of the inventory trajectory

(s, S) replenishment solver
Cycle-service-level reorder point · EOQ-based order-up-to · 60-day simulation
★★ Closed-form heuristic
Daily sell-through
Per-day std. dev.
DC to store
Order placement
Carrying cost
Per unit short
Purchase price
—
Reorder point \(s\)
—
Order-up-to \(S\)
—
\(Q_{\text{EOQ}}\)
—
Exp. stockouts / cycle
—
Avg. inventory (units)
—
Avg. cost / period ($)
On-hand inventory \(I_t\) Reorder point \(s\) Order-up-to \(S\) Order arrival

Under the hood

We compute \(z_\alpha = \Phi^{-1}(p / (p+h))\) via the Beasley-Springer-Moro rational approximation, then \(s = \mu L + z_\alpha \sigma \sqrt{L}\) and \(Q_{\text{EOQ}} = \sqrt{2K\mu/h}\), yielding \(S = s + Q_{\text{EOQ}}\). To visualise the policy in action we run a 60-day Monte Carlo simulation: each day demand is drawn from \(\mathcal{N}(\mu, \sigma^2)\) (clipped at zero), inventory is depleted, and an order of size \(S - I_{\text{on-hand}}\) fires when \(I_{\text{on-hand}} \leq s\). Outstanding orders arrive \(L\) days later. The 6 KPIs report the policy parameters plus the observed average inventory, expected stockouts per cycle (closed-form via the loss function), and total cost per period from the \(C(s, S)\) decomposition.

Reading the solution

What an inventory planner does with the (s, S) parameters

How the safety-stock cushion behaves

The reorder point \(s\) decomposes into expected lead-time demand \(\mu L\) plus a safety-stock buffer \(z_\alpha \sigma \sqrt{L}\). Two structural properties matter:

  • Square-root law in lead time. Doubling \(L\) raises the safety-stock term by only \(\sqrt{2} \approx 1.41\), not 2 — longer leads are punished sub-linearly.
  • Service-level sensitivity. Moving from a 95% to a 99% cycle service level shifts \(z_\alpha\) from 1.65 to 2.33; the last few percentage points of service are expensive in carried units.
  • Demand variability dominates. If \(\sigma\) doubles, safety stock doubles — but \(s\) itself rises by less because the \(\mu L\) term is unchanged. Forecast accuracy is the cheapest lever.

Sensitivity questions the model answers instantly

  • Cut lead time from 4 to 2 days? Safety stock falls by \(1 - 1/\sqrt{2} \approx 29\%\); average inventory drops, ordering frequency unchanged.
  • Halve fixed ordering cost \(K\)? EOQ falls by \(\sqrt{2}\), so each cycle is shorter and average cycle stock \((S - s)/2\) drops — orders fire more often but each is smaller.
  • Stockout penalty \(p\) doubles? The cost-balance service level \(p/(p+h)\) rises, pushing \(z_\alpha\) and the safety stock up; stockouts per cycle fall.

Model extensions

From single-SKU baseline to retail-OR variants that matter

Stochastic lead time

Replace constant \(L\) with random \(L \sim F_L\). Lead-time demand variance becomes \(\mu^2 \mathrm{Var}(L) + \mathbb{E}[L]\sigma^2\); safety stock grows with both demand and lead-time uncertainty.

Multi-product joint replenishment

Several SKUs share the same delivery truck. Joint replenishment problem: coordinate cycles to share fixed cost \(K\). Roundy's 98%-power-of-two heuristic.

Perishable variant

Add a fixed shelf life; expired units cost more than they salvage. See grocery ordering for the perishable inventory variant.

Capacitated supplier

DC cannot ship more than \(Q_{\max}\) per order. Modified (s, S) caps the order quantity; lost-sales recourse when \(S - s > Q_{\max}\).

Periodic review (T, R, S)

Inventory checked every \(T\) periods rather than continuously; if below reorder point \(R\), order up to \(S\). Review interval adds \(T/2\) to the safety-stock-protection horizon.

Wagner-Whitin (deterministic)

For known time-varying demand, exact DP yields zero-inventory-ordering optimal lot sizes. See the inventory family for Wagner-Whitin and Silver-Meal.

Multi-echelon coordination

DC inventory and store inventory are jointly optimised; Clark-Scarf (1960) base-stock decomposition. Multi-echelon →

Omnichannel pooling

Pool store inventory with online fulfillment and ship-from-store. Pooled \(\sigma\) shrinks by \(\sqrt{n}\) under independence; safety stock falls accordingly.

Key references

Foundational (s, S) inventory theory and texts

Scarf, H. (1960).
The optimality of (s, S) policies in the dynamic inventory problem.
In K. J. Arrow, S. Karlin & P. Suppes (eds.), Mathematical Methods in the Social Sciences, Stanford University Press.
Veinott, A. F. & Wagner, H. M. (1965).
Computing optimal (s, S) inventory policies.
Management Science 11(5): 525–552. doi:10.1287/mnsc.11.5.525
Federgruen, A. & Zipkin, P. (1984).
An efficient algorithm for computing optimal (s, S) policies.
Operations Research 32(6): 1268–1285. doi:10.1287/opre.32.6.1268
Hadley, G. & Whitin, T. M. (1963).
Analysis of Inventory Systems.
Prentice-Hall, Englewood Cliffs, NJ. (Cost-balance derivation of \(z_\alpha = \Phi^{-1}(p/(p+h))\).)
Silver, E. A., Pyke, D. F. & Peterson, R. (1998).
Inventory Management and Production Planning and Scheduling.
3rd ed., Wiley. (Standard practitioner text on (s, S) and (R, s, S).)
Axäter, S. (2015).
Inventory Control.
3rd ed., Springer International Series in OR & Management Science. doi:10.1007/978-3-319-15729-0
Zipkin, P. (2000).
Foundations of Inventory Management.
McGraw-Hill / Irwin. (Comprehensive theoretical treatment.)
Porteus, E. L. (2002).
Foundations of Stochastic Inventory Theory.
Stanford Business Books, Stanford University Press.

Back to the retail domain

Store replenishment sits in the Place × Operational cell of the 4P decision matrix — the policy that turns demand forecasts into back-room receipts at the right moment.

Open Retail Landing
Educational solver · Normal-demand approximation, constant lead time, single-SKU · validate against your own demand distribution and lead-time variability before deploying in production.