Periodic Review, Multi-Echelon and Special Inventory Systems

Introduction

Inventory policy is not determined solely by whether demand is predictable. It is also shaped by review frequency, replenishment lead time, the number of locations in the supply network, product life, demand intermittency, and the economic consequences of shortage and surplus. Continuous-review models are powerful when inventory records and ordering decisions can be updated immediately. In many organisations, however, replenishment decisions are made weekly, monthly, or at scheduled supplier visits. Other settings involve several stocking points, rapidly perishing products, seasonal demand, or expensive service parts with highly intermittent demand.

This chapter develops a unified framework for these situations. It explains periodic-review, base-stock and min–max policies; derives the principal cost and service equations; and evaluates applications to multi-echelon, perishable, seasonal, spare-parts, and slow-moving inventories. The emphasis is managerial as well as quantitative: a mathematically attractive policy can fail if data, governance, capacity, or execution are inadequate.

Learning objectives

By the end of this chapter, the learner should be able to:

1. Distinguish continuous-review and periodic-review control.
2. Design and interpret order-up-to, base-stock, and min–max policies.
3. Apply the newsvendor logic to single-period and seasonal decisions.
4. Explain echelon inventory and the logic of multi-echelon optimisation.
5. Select appropriate approaches for perishable, spare-parts, and slow-moving items.
6. Evaluate assumptions, service implications, technology requirements, and governance risks.

1. The decision architecture of inventory control

An inventory policy specifies at least four elements:

• Review mechanism: continuous or periodic.
• Replenishment rule: how much to order when a review occurs.
• Service objective: cycle service level, fill rate, response time, or a financial target.
• Scope of optimisation: one item at one location, or a network of facilities and inventory stages.

The inventory position is the quantity that is economically relevant for replenishment:

where OH_t is on-hand inventory, OO_t is on-order inventory, and BO_t is backorders at time t. All terms must use the same unit and location scope. In a periodic system, an order is normally based on inventory position at the review epoch, not merely on physical stock.

A policy must also distinguish inventory level from inventory position. Inventory level excludes material still in transit; inventory position includes the replenishment pipeline and therefore avoids double ordering. Errors in this distinction are a frequent cause of excess stock.

2. Periodic-review control

Under a periodic-review policy, stock is inspected at intervals of length P. An order is then placed to raise inventory position to a target. During the interval until the next review, demand can occur without a replenishment decision. Consequently, protection is required for the supplier lead time plus the review interval.

For a single item with deterministic lead time L, the order-up-to policy is:

Here Q_t is the order quantity and S_t is the order-up-to level. A common stochastic target is:

where μ_L+P and σ_L+P are the mean and standard deviation of demand during the protection period L+P, and z is the standard-normal quantile associated with the chosen cycle-service probability. The formula assumes that demand during the protection period is approximately normal, lead time is fixed or already incorporated into the demand distribution, replenishment is unconstrained, and shortages are recorded consistently.

If independent, identically distributed period demand has mean μ and standard deviation σ, then:

The second relationship assumes independent demand across periods. If demand is autocorrelated, the variance is instead:

where γ_k is the covariance between demands separated by k periods and n = L+P. Positive autocorrelation increases uncertainty relative to the independent-demand approximation.

Example: weekly periodic review

A distributor reviews an item every two weeks. Supplier lead time is three weeks. Weekly demand has mean 100 units and standard deviation 30 units. The desired cycle service level is 95%, for which z ≈ 1.645. Assuming independent weekly demand:

Protection period = L+P = 3+2 = 5 weeks.

The operational target is approximately 611 units. If inventory position at review is 438 units, the order is approximately 173 units. The calculation does not imply that average inventory is 611 units: the target includes expected demand during the protection period and is reset as inventory position changes.

Service measures

Cycle service level is the probability of avoiding a stockout during a replenishment cycle. It is not identical to fill rate, which is the proportion of demand supplied immediately. A policy may have a high cycle service level but a lower fill rate when a rare stockout is large. For high-volume products, fill rate and expected shortage are often more decision-relevant than cycle service alone.

If protection-period demand is normal with mean μ_D and standard deviation σ_D, and safety stock is zσ_D, expected shortage per cycle is:

where φ and Φ are the standard-normal density and cumulative distribution functions. This expression assumes shortages are measured as units short at the replenishment point and demand is continuous enough for the normal approximation to be useful.

3. Base-stock and min–max policies

A base-stock policy replenishes every unit consumed, so that the desired inventory position is continually restored to a base-stock level. In its ideal form, it is a continuous-review order-up-to policy with target S. It is effective where demand is frequent, replenishment is responsive, and inventory records are accurate. Its weakness is transaction intensity: it may generate many small orders and can amplify administrative or transportation costs.

A periodic base-stock policy uses the same order-up-to logic but acts only at review times. The review interval is therefore part of the protection period.

A min–max policy is generally represented by (s,S): when inventory position is at or below s, order enough to reach S. In a periodic implementation, the equivalent is often described as (s,S,P), with the threshold checked every P periods. A useful design is:

and

The first target protects against lead-time demand; the second covers the review interval as well. The two service factors need not be equal because the organisation may use different targets for reorder protection and replenishment-up-to coverage.

The min–max policy is operationally attractive because it is easy to explain and compatible with fixed ordering routines. It is less optimal than a fully cost-based dynamic policy when demand, lead time, capacity, and substitution are strongly coupled. It can also generate undesirable order sizes if the gap S−s is not aligned with supplier minimums or transport economics.

4. Choosing review intervals and targets

Shortening P reduces the time during which demand can accumulate before action, but it may increase ordering, transport, counting, and administrative cost. Lengthening P reduces transaction frequency but increases safety stock because uncertainty grows with the protection period.

A practical design sequence is:

1. Segment items by value, criticality, demand pattern, and lead-time risk.
2. Estimate demand and lead-time distributions rather than relying only on averages.
3. Select a review interval consistent with supplier schedules and operational capacity.
4. Set a service objective appropriate to the consequence of shortage.
5. Calculate an initial target and simulate it against historical or synthetic demand.
6. Stress-test demand surges, late deliveries, minimum order quantities, and capacity limits.
7. Monitor realised service, inventory, obsolescence, and forecast bias.

A periodic policy should not be judged by average inventory alone. Relevant measures include fill rate, stockout frequency, order frequency, inventory turns, working capital, obsolete inventory, emergency shipments, and customer response time.

5. The newsvendor model and single-period inventory

The newsvendor model applies when a replenishment decision is made before uncertain demand and leftover stock has limited future value. Examples include seasonal fashion, newspapers, event merchandise, fresh food, and capacity reserved for a short selling window.

Let:

• c_u = underage cost per unit, the contribution lost or penalty incurred when one unit is unavailable;
• c_o = overage cost per unit, the cost of one unit left after the selling period;
• D = random demand;
• Q = order quantity.

The expected cost is:

where x^+ = max(x,0). Under a continuous demand distribution, the optimal quantity satisfies the critical-fractile rule:

or equivalently:

The closing parenthesis in the inverse expression should be read as Q* = F^-1(c_u/(c_u+c_o)). The rule assumes a single decision period, known cost parameters, no replenishment during the period, and a specified salvage or disposal treatment embedded in c_o. If unit purchase cost is c, selling price is p, and salvage value is v, then, ignoring other costs:

Example: seasonal product

A retailer buys a seasonal item for 20 currency units, sells it for 50, and clears leftovers for 8. Thus c_u = 50−20 = 30 and c_o = 20−8 = 12. The critical fractile is:

If demand is normal with mean 1,000 and standard deviation 250, the 71.43rd percentile has z ≈ 0.566. Therefore:

The result is not a 50th-percentile forecast. The retailer orders above mean demand because the margin lost from a stockout is greater than the net loss on a leftover unit.

For seasonal products with multiple opportunities to replenish, the basic newsvendor model is incomplete. Dynamic programming, rolling forecasts, capacity reservation, and postponement may dominate a single initial buy. Similarly, if leftover stock can be sold in later seasons, c_o is not simply the purchase cost less salvage value.

6. Multi-echelon inventory systems

A multi-echelon system contains several inventory stages, such as a central factory, regional distribution centres, and local warehouses. Optimising each location independently can create excessive total stock because every stage protects itself against uncertainty that may already be buffered elsewhere.

Echelon inventory at stage i is inventory at that stage plus inventory held downstream, according to the defined network direction. In a serial system, echelon inventory can be expressed as:

where I_j includes the relevant on-hand and pipeline quantities at downstream stage j. The exact definition must be fixed before measurement; otherwise echelon stock can be double counted.

The key principle is risk pooling. A central stock point serving several markets aggregates demand variability, while local stock points provide faster response. Centralisation usually lowers safety stock but may increase transport time and response risk. Decentralisation can improve customer responsiveness but duplicates protection stock.

For independent demands with standard deviations σ₁, σ₂, …, σ_n, pooled standard deviation is:

If all n locations have equal standard deviation σ, the pooled standard deviation is σ√n, whereas independently buffered locations require total safety stock proportional to nσ. Thus pooling reduces variability when demand is not perfectly correlated. With correlation, the general variance is:

where ρ_ij is the correlation between locations i and j. High positive correlation substantially reduces the benefit of pooling.

Echelon order-up-to logic

In a serial system, a stage may set an echelon order-up-to level based on demand over the cumulative lead time from that stage to the customer. A simplified target is:

where T_i is the relevant downstream protection time. Local physical inventory targets are then derived from differences between adjacent echelon targets. In practice, optimisation must also include capacity, transportation, minimum batch sizes, substitution, allocation rules, and service differentiation.

Multi-echelon methods include guaranteed-service models, installation-stock models, stochastic programming, simulation optimisation, and approximate dynamic programming. No method eliminates the need for accurate lead times and explicit service definitions.

7. Perishable inventory

Perishable inventory loses value through expiration, deterioration, quality degradation, or regulatory restrictions. The fundamental decision is not only how much to stock, but also which units to issue first and when to dispose, discount, rework, or substitute.

Important controls include:

• FEFO (first-expiry, first-out): issue the unit with the earliest expiry date.
• Expiry-date visibility at receipt, storage, picking, and shipment.
• Shelf-life acceptance rules for suppliers and customers.
• Age-based safety stock rather than quantity-only safety stock.
• Markdown, donation, reprocessing, or disposal rules.
• Temperature and handling monitoring where quality is condition-dependent.

A simple perishability model may divide stock into age cohorts. If I_a is inventory of age a, then the issue decision should normally prioritise low remaining life, subject to customer shelf-life requirements. A quantity policy that ignores age can report high service while generating unacceptable waste.

For food and pharmaceuticals, the overage cost includes disposal, reverse logistics, environmental impact, reputational damage, and compliance exposure. The underage cost may include lost sales, patient risk, or service disruption. Therefore, the apparent newsvendor costs must be expanded to reflect the full consequence of decisions.

8. Seasonal and intermittent demand

Seasonality creates systematic variation over time, while intermittency creates many zero-demand periods separated by sporadic transactions. Their combination is common in spare parts and seasonal maintenance items.

Seasonal planning should separate:

1. baseline demand;
2. calendar or event effects;
3. trend and lifecycle effects;
4. promotions and planned interventions;
5. uncertainty around the forecast.

A multiplicative seasonal model may be written as:

where L_t is level or trend, S_t is the seasonal index, and ε_t is a random factor with expected value approximately 1. This representation is appropriate when seasonal variability grows with the level. An additive model is preferable when seasonal amplitude is approximately constant:

The chosen model must be tested rather than assumed.

For intermittent demand, ordinary moving averages may produce biased timing and excessive volatility. Croston-type methods estimate demand size and the interval between non-zero demands separately. Their variants, including SBA and TSB, address some bias and obsolescence issues, but no forecasting method solves structural uncertainty such as an unannounced product redesign.

9. Spare-parts inventory

Spare parts differ from ordinary finished goods in several ways: demand may be intermittent, failure consequences may be severe, lead times may be long, and substitution or repair may be possible. The correct service metric may be probability of having a part when needed, mean time to repair, fleet availability, or mission completion—not conventional sales fill rate.

A useful classification considers:

• criticality of equipment failure;
• demand frequency and variability;
• replenishment lead time;
• purchase price and repairability;
• commonality across assets;
• obsolescence and engineering-change risk.

For a fleet with independent component failures, if each asset has failure probability p during a planning period and there are N assets, expected failures are:

When failures are independent, demand variance is:

This binomial approximation is useful for a common component over a defined period. It becomes unreliable when failures are correlated, operating conditions differ, or maintenance policy changes exposure.

Pooling parts across a fleet can reduce total stock, but location matters. A central emergency depot may be economical for expensive, low-demand parts, while local stock is justified for critical parts with severe downtime cost. Repair loops, cannibalisation, loaners, and alternate part numbers should be modelled explicitly.

10. Slow-moving and obsolete inventory

Slow-moving inventory has low transaction frequency and often high uncertainty in the timing of demand. A high service target based only on unit demand can produce excessive capital tied up in stock. Conversely, reducing stock indiscriminately can create long customer delays for critical items.

Governance should distinguish:

• low demand but high criticality;
• low demand and low criticality;
• obsolete or superseded items;
• items with recoverable value through repair or transfer;
• items required by contract or regulation.

Useful measures include demand-weighted availability, age profile, months of supply, forecast bias, last-demand date, excess-and-obsolete value, and probability of future use. For a slow-moving item, an empirical demand distribution or a Bayesian model may be more defensible than a normal approximation. Decisions should incorporate the probability of future demand and the value of alternatives such as emergency purchase, substitution, or redesign.

11. Technology, analytics and governance

Modern inventory systems combine enterprise resource planning, warehouse-management systems, transport data, supplier portals, IoT sensing, and advanced analytics. AI can assist with demand classification, anomaly detection, lead-time prediction, and policy simulation. It does not remove the need for causal reasoning, master-data control, or human accountability.

Good governance requires:

• traceable forecasts and parameter changes;
• clear ownership of service and inventory targets;
• version-controlled policies;
• monitoring for data drift and forecast bias;
• human override with documented rationale;
• cybersecurity for connected warehouses and supplier interfaces;
• protection of commercially sensitive and customer data.

Automated recommendations may systematically favour high-volume customers or penalise locations with poorer data. Managers should test allocation rules for fairness, contractual compliance, and operational safety. In regulated sectors, an explainable, validated model is generally preferable to an opaque model with marginally better historical accuracy.

12. Illustrative case: medical-device regional inventory

Background

A medical-device manufacturer supplies a critical replacement component through one central distribution centre and four regional hospitals. Demand is intermittent, the component has a 12-week supplier lead time, and the manufacturer reviews hospital replenishment requests weekly. A stockout can delay surgery, while unused units expire after 18 months.

Problem

Each hospital independently set a high min–max target using its own demand history. The result was high total inventory, uneven availability, and frequent emergency transfers. Some hospitals held old stock while another experienced a shortage.

Analysis

The manufacturer first reconciled inventory position, including in-transit stock, and established FEFO issuing. It then classified the component as critical, intermittent, and perishable. Hospital demand was pooled for central planning, but local response requirements were retained. Correlation was estimated from procedure schedules rather than assumed to be zero. The policy objective was changed from a generic 98% cycle service level to a combination of emergency availability and expiry-adjusted total cost.

Approach

The central facility adopted a periodic order-up-to policy with a protection period equal to supplier lead time plus the weekly review interval. Regional hospitals retained smaller local buffers, with emergency replenishment from the central facility. A monthly reallocation rule moved stock approaching expiry to hospitals with near-term scheduled procedures. The target parameters were simulated under normal demand, demand surges, late supplier deliveries, and expiry constraints.

Outcome

In this illustrative scenario, total stock fell by 19%, emergency transfers fell by 31%, and critical-case availability improved because stock was positioned according to procedure timing rather than historical averages alone. Expiry losses initially increased during the transition because old stock was discovered; after FEFO and reallocation were implemented, expiry losses fell below the original level.

Learning points

The case demonstrates that pooling alone is not sufficient. Multi-echelon design must balance centralisation against response time; periodic-review targets must include the review interval; inventory age must be visible; and service metrics must reflect clinical consequence rather than merely average fill rate.

13. Limitations and implementation cautions

The formulas in this chapter are approximations when demand is non-normal, lead times are stochastic, orders are constrained, or multiple items compete for capacity. A normal safety-stock calculation can be particularly misleading for intermittent demand, very short lifecycles, or highly skewed lead times.

Implementation commonly fails for non-mathematical reasons: inaccurate inventory records, unrecorded returns, ignored supplier minimums, inappropriate forecast aggregation, poor expiry data, and conflicting performance incentives. Purchasing may minimise unit price while increasing obsolete stock; sales may promise availability without funding the required buffer; warehouses may optimise local utilisation while damaging network service.

Best practice is therefore iterative: define the service promise, build a transparent model, validate the data, simulate realistic constraints, pilot the policy, and review outcomes using both financial and operational measures.

Chapter Summary

Periodic-review systems make replenishment decisions at scheduled intervals and must protect against demand during both lead time and the review interval. Order-up-to policies use inventory position, not merely on-hand stock. Base-stock policies continuously restore a target, while min–max policies trigger replenishment when inventory position reaches a threshold.

The newsvendor model selects a single-period quantity through the critical fractile, balancing underage and overage costs. Multi-echelon systems require attention to echelon inventory, risk pooling, correlation, centralisation, and response time. Perishable inventory adds age, expiry, waste, and compliance dimensions. Seasonal demand requires explicit seasonal structure, whereas intermittent and slow-moving demand often requires specialised forecasting and empirical distributions. Spare-parts policies must reflect failure processes, criticality, fleet pooling, repair, substitution, and downtime consequences.

Across all systems, the quality of the policy depends on data integrity, realistic assumptions, appropriate service metrics, simulation under constraints, and accountable governance. Technology and AI can improve prediction and visibility, but they must be controlled, explainable, and aligned with operational and ethical requirements.

Key Terms

Inventory position: On-hand plus on-order inventory minus backorders.

Periodic review: Inventory control in which stock and replenishment decisions are assessed at fixed intervals.

Protection period: Time covered by a replenishment target, commonly lead time plus review interval.

Order-up-to level: Target inventory position to which a replenishment order raises stock.

Base-stock policy: A policy that restores inventory position to a specified target after demand.

Min–max policy: A policy that orders up to a maximum when inventory position reaches a minimum threshold.

Cycle service level: Probability of no stockout during a replenishment cycle.

Fill rate: Proportion of demand supplied immediately from available stock.

Safety stock: Inventory held to absorb uncertainty in demand, supply, or lead time.

Newsvendor model: A single-period model balancing underage and overage costs.

Critical fractile: The optimal cumulative demand probability in the newsvendor model.

Echelon inventory: Inventory at a stage plus relevant downstream inventory in a supply network.

Risk pooling: Reduction in aggregate uncertainty achieved by combining demand streams.

Perishable inventory: Stock whose value or usability declines with time or age.

FEFO: First-expiry, first-out issuing discipline.

Intermittent demand: Demand containing many zero-demand periods and occasional transactions.

Spare part: Component stocked to restore or maintain equipment operation.

Slow-moving inventory: Inventory with low demand frequency or turnover.

Obsolescence: Loss of economic or functional value due to age, redesign, or market change.

Review Questions

Short Answer Questions

1. Why is the protection period in a periodic-review system generally longer than supplier lead time?
2. Distinguish inventory level from inventory position.
3. What is the difference between cycle service level and fill rate?
4. State the critical-fractile rule and explain its two cost components.
5. Why can demand pooling reduce safety stock in a multi-echelon network?

Long Answer Questions

1. Explain the design and managerial trade-offs of periodic-review, base-stock, and min–max policies.
2. Derive and interpret the newsvendor critical-fractile rule, including the effects of salvage value.
3. Discuss how echelon inventory and risk pooling influence the design of a multi-echelon system.
4. Compare inventory-control challenges for perishable goods, seasonal products, and slow-moving items.
5. Develop a policy framework for critical spare parts, including service metrics, repair, substitution, and obsolescence.

Analytical / Application Questions

1. Weekly demand has mean 120 units and standard deviation 25 units. A firm reviews stock every two weeks, lead time is four weeks, and the target cycle service level is 95%. Assuming independent normal demand, calculate the order-up-to level.
2. A seasonal product costs 30 currency units, sells for 75, and has a salvage value of 10. Demand is normal with mean 2,000 and standard deviation 400. Calculate the critical fractile and estimate the optimal order quantity using the corresponding standard-normal quantile.
3. A company has four regional warehouses with correlated demand and a central emergency depot. Design an analysis to decide which parts should be pooled centrally and which should remain locally stocked. Specify data, service measures, constraints, and tests you would use.

Multiple Choice Questions with Answers

Question 1

A periodic-review system has a two-week review interval and a five-week supplier lead time. Which period should normally be covered by the order-up-to target?

A. Two weeks

B. Five weeks

C. Seven weeks

D. The average order cycle only

Correct Answer: C. Seven weeks

Question 2

Which quantity correctly represents inventory position?

A. On-hand inventory minus on-order inventory

B. On-hand inventory plus on-order inventory minus backorders

C. On-hand inventory plus backorders

D. On-order inventory minus on-hand inventory

Correct Answer: B. On-hand inventory plus on-order inventory minus backorders

Question 3

A product has underage cost 40 and overage cost 10. What is its newsvendor critical fractile?

A. 0.20

B. 0.40

C. 0.75

D. 0.80

Correct Answer: D. 0.80

Question 4

Which statement best distinguishes a fill rate from a cycle service level?

A. Fill rate measures the proportion of demand supplied immediately

B. Fill rate is always equal to the probability of no stockout

C. Cycle service level measures inventory value only

D. Cycle service level ignores replenishment cycles

Correct Answer: A. Fill rate measures the proportion of demand supplied immediately

Question 5

If two independent locations each have demand standard deviation σ, the standard deviation of their pooled demand is:

A. 2σ

B. σ/2

C. σ√2

D. σ²

Correct Answer: C. σ√2

Question 6

Which issuing rule is most appropriate for inventory with different expiry dates?

A. Last-expiry, first-out

B. First-expiry, first-out

C. Highest-cost, first-out

D. Random issue to equalise bins

Correct Answer: B. First-expiry, first-out

Question 7

What is a central feature of intermittent demand?

A. Demand is positive and constant every period

B. Demand contains many zero-demand periods

C. Demand is always normally distributed

D. Demand has no forecasting uncertainty

Correct Answer: B. Demand contains many zero-demand periods

Question 8

For a fleet of N assets with independent failure probability p during a period, which expression gives expected failures?

A. N+p

B. N/p

C. Np

D. p−N

Correct Answer: C. Np

Question 9

What is the principal purpose of echelon inventory analysis?

A. To count only warehouse floor space

B. To optimise each location without considering downstream stock

C. To represent inventory across a stage and its downstream network

D. To eliminate all pipeline inventory

Correct Answer: C. To represent inventory across a stage and its downstream network

Question 10

Which is a major governance risk when AI recommends inventory policies?

A. It makes inventory position unnecessary

B. It guarantees unbiased allocations

C. It can reproduce data bias or drift without accountable oversight

D. It removes the need for service definitions

Correct Answer: C. It can reproduce data bias or drift without accountable oversight