Inventory Under Uncertainty

Introduction and Context

Inventory decisions are made before managers know exactly how much customers will buy or when replenishment will arrive. Demand may fluctuate because of seasonality, promotions, competition, or random customer behavior. Lead time may vary because of production delays, transportation congestion, supplier capacity, customs, or quality problems. A continuous-review inventory system must therefore protect the organization against uncertainty while avoiding excessive capital, space, obsolescence, and handling cost.

This chapter develops the logic of continuous-review policies, especially the reorder-point system. The central decision is not simply how much inventory to hold, but how much protection is appropriate for a specified service objective. The analysis distinguishes demand variability, lead-time variability, lost sales, and backorders, because these assumptions affect both policy design and performance measurement.

Learning Objectives

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

1. Explain the role of safety stock and reorder points under uncertain demand and lead time.
2. Model lead-time demand using probability distributions.
3. Calculate reorder points for fixed and variable lead times.
4. Distinguish cycle service level, fill rate, stockout probability, lost sales, and backorders.
5. Analyze the joint effect of demand and lead-time variability.
6. Design and critically evaluate continuous-review policies in practical settings.

1. The Continuous-Review Policy

In a continuous-review system, inventory position is monitored whenever a transaction occurs. An order is placed when inventory position reaches the reorder point. Inventory position is defined as:

where IP is inventory position, OH is on-hand inventory, OO is inventory on order, and BO is backorders. All quantities are measured in units.

The common policy is the (Q, R) policy:

• Q: fixed replenishment order quantity;
• R: reorder point.

An order of Q units is placed whenever IP ≤ R. The reorder point protects demand expected during replenishment lead time and provides additional safety stock.

For deterministic demand and lead time, the reorder point is simply lead-time demand. Under uncertainty, it is a selected percentile of lead-time demand:

where μ_L is mean demand during lead time and SS is safety stock. The units of all three terms must be identical, such as units per replenishment lead time.

A continuous-review policy differs from a periodic-review policy. In periodic review, inventory is examined at intervals and protection must cover both the review period and lead time. Continuous review can generally achieve the same service target with less safety stock, but it requires reliable transaction data and disciplined execution.

2. Demand During Lead Time

The relevant random quantity is not demand per day alone; it is demand accumulated while an order is outstanding. Let daily demand be D_t and lead time be L. Lead-time demand is:

If lead time is fixed at L periods, daily demands are independent with mean μ_D and standard deviation σ_D, then:

These relationships assume identical demand distributions, independence across periods, and a fixed lead time. The standard deviation grows with the square root of time, not linearly, because independent random fluctuations partially offset one another.

If lead-time demand is approximately normal, a target cycle service level can be translated into a quantile:

Here z is the standard-normal quantile corresponding to the desired probability of having sufficient stock throughout one replenishment cycle. Typical values are approximately 1.28 for 90%, 1.645 for 95%, and 2.326 for 99%.

Safety stock is therefore:

This formula is appropriate when the normal approximation is reasonable and the organization is targeting a cycle service level. It is not automatically a fill-rate formula, and it should not be used uncritically for highly intermittent, bounded, skewed, or promotional demand.

Example 1: Fixed Lead Time

A medical distributor experiences average daily demand of 80 units with a daily standard deviation of 20 units. Lead time is constant at 5 days. Management wants a 95% cycle service level.

Mean lead-time demand is:

The standard deviation of lead-time demand is:

Using z = 1.645:

Thus:

The manager should set the reorder point at approximately 474 units, subject to rounding, pack-size constraints, and validation of the distributional assumption.

3. Variable Lead Time

When demand varies but lead time is fixed, the preceding formula is sufficient. When lead time also varies, uncertainty has two sources. If daily demand and lead time are independent, with mean demand μ_D, demand standard deviation σ_D, mean lead time μ_L, and lead-time standard deviation σ_L, the variance of lead-time demand is:

Consequently:

The first term represents demand variability accumulated over the average lead time. The second represents additional demand caused by a longer-than-average lead time. This result assumes independent demand across periods, independence between demand and lead time, and a lead time sufficiently large for the random-sum approximation to be useful.

If demand and lead time are correlated, the independence formula is incomplete. A useful approximation adds a covariance component:

where Cov(D,L) is the covariance between demand conditions and lead time. Positive covariance occurs, for example, when peak demand coincides with congestion and longer supplier lead times. Ignoring it understates risk. In advanced applications, simulation or empirical lead-time-demand observations are preferable.

Example 2: Demand and Lead-Time Variability

Daily demand has mean 100 units and standard deviation 25 units. Lead time has mean 6 days and standard deviation 2 days. Assume independence and a normal approximation. Then:

Mean lead-time demand is 600 units. For a 95% cycle service level:

The large safety stock, approximately 344 units, is driven principally by lead-time variability. Reducing supplier lead-time variance may be more economical than increasing order quantities or accepting more inventory.

4. Probability Distributions and Empirical Modeling

A probability distribution describes possible demand or lead-time outcomes and their probabilities. The normal distribution is convenient because it is mathematically tractable, but convenience is not evidence of validity. Demand distributions may be:

• discrete, as with spare parts;
• skewed, as with new-product demand;
• intermittent, with many zero-demand periods;
• bounded, as with capacity-limited services;
• seasonal or regime-changing.

Managers should examine histograms, quantiles, autocorrelation, outliers, calendar effects, and forecast errors before choosing a model. For a nonnormal distribution, the reorder point can be set directly from the empirical quantile:

where F^-1_DL is the inverse cumulative distribution function of lead-time demand and CSL is the desired cycle service level. This approach avoids forcing a normal shape on the data.

If historical observations of lead-time demand are available, the empirical percentile may be calculated and adjusted for sample size, structural changes, and forecast bias. Forecast uncertainty should also be included. If demand is forecast as \hat{D} and forecast error has standard deviation σ_e, safety stock should be based on the distribution of forecast error accumulated over lead time rather than merely on historical demand variation.

5. Service-Level Measures

Service level is not a single universal measure. Different measures answer different managerial questions.

For a cycle service level:

Under a continuous distribution, the stockout probability during the cycle is approximately:

A 95% cycle service level does not mean that 95% of units are filled. A stockout may occur late in a cycle and affect only a small quantity, or it may occur early and affect substantial demand. Fill rate is therefore more sensitive to shortage magnitude.

For a continuous demand distribution with order quantity Q, expected shortage per cycle is:

where (x)^+ = max(x,0). If each cycle faces approximately Q units of demand, a common approximation is:

This relationship requires careful interpretation because cycle demand, order quantity, and replenishment timing may not align exactly. For a normal lead-time demand distribution with mean μ_L and standard deviation σ_L, define z = (R−μ_L)/σ_L. Expected shortage can be expressed as:

where φ(z) is the standard-normal density and Φ(z) is its cumulative distribution. This formula applies to a normally distributed lead-time demand and measures expected units short during the relevant cycle.

6. Lost Sales and Backorders

A lost-sales system assumes that unmet demand disappears. The customer does not wait, and the firm records a missed sale. In a backorder system, unmet demand remains in a queue and is delivered later. The distinction affects inventory position, customer experience, revenue, and performance measurement.

Under lost sales, shortage units are not added to future demand. Under backorders, they become commitments and reduce available inventory when replenishment arrives. Thus, the same physical stockout can produce different future states.

The choice should reflect the market. Perishable or easily substituted products often have lost sales. Industrial components, repair parts, and contractual supply arrangements more often create backorders. A model that assumes backorders when customers actually defect will overstate realized service. A lost-sales model may understate the operational burden when customers wait and demand accumulates.

Shortage cost should include more than purchase cost. Relevant components may include lost contribution margin, expediting, penalties, production interruption, customer dissatisfaction, regulatory consequences, and reputational damage. These costs should not be hidden by reporting only accounting inventory value.

7. Order Quantity and Reorder Point Are Different Decisions

The reorder point controls protection during replenishment lead time; the order quantity controls replenishment frequency and cycle-stock investment. The economic order quantity is often introduced as:

where D is annual demand in units per year, S is ordering cost per order, and H is annual holding cost per unit. The classical formula assumes constant demand, instantaneous replenishment, no quantity discounts, and no shortages. In uncertain systems, Q may be selected using this logic, supplier constraints, transportation economics, or optimization under service requirements. It should not be confused with safety stock.

Increasing Q generally increases cycle stock but does not necessarily solve lead-time uncertainty. Reducing supplier lead-time variation can lower safety stock without changing customer demand. Managers should therefore diagnose the source of inventory before changing the order quantity.

8. A Practical Design Procedure

A defensible continuous-review design follows these steps:

1. Define the item, customer promise, review transaction, and replenishment process.
2. Measure demand and lead-time data at a consistent time scale.
3. Correct data for returns, cancellations, substitutions, promotions, and stockout-censored demand.
4. Estimate demand during lead time, including forecast error where applicable.
5. Select a service measure and economic target.
6. Choose an appropriate distribution or empirical quantile.
7. Calculate the reorder point and round for packaging or minimum-order constraints.
8. Simulate or back-test the policy using historical or scenario data.
9. Monitor service, inventory, expedites, obsolescence, and bias.
10. Recalibrate when demand, supplier performance, or customer requirements change.

A particularly important issue is censored demand. If a product was out of stock, recorded sales may be less than true demand. Treating sales records as complete demand estimates can systematically underestimate the reorder point.

Illustrative Case: Regional Hospital Supply Network

Background

A regional hospital network uses a critical sterile connector in emergency procedures. The item is purchased from a certified supplier. Demand is variable, and deliveries are transported from a distant distribution center.

Problem

The purchasing team had set a 95% service target using average daily demand and average lead time only. Although average inventory appeared reasonable, emergency stockouts occurred during high-demand weeks and supplier delays. The team also reported a 95% service level without specifying whether this meant cycle service or fill rate.

Analysis

The analyst separated ordinary demand, emergency demand, and stockout periods. Mean daily demand was 100 units, with a standard deviation of 25 units. Mean lead time was 6 days, with a standard deviation of 2 days. Historical evidence showed that high-demand periods were associated with transport congestion, suggesting positive demand–lead-time dependence. The independent-variance approximation therefore understated risk.

The team first calculated the independent estimate of lead-time standard deviation as approximately 209 units and a reorder point of approximately 944 units for a 95% cycle service level. Back-testing showed that this policy produced unacceptable shortages during correlated high-demand and delayed-delivery events.

Approach and Solution

The organization adopted four measures:

• a higher protection target for this critical item;
• empirical lead-time-demand percentiles from historical observations;
• supplier performance agreements focused on lead-time variance, not only average lead time;
• a separate emergency reserve governed by clinical criticality.

The inventory system also distinguished backorders from lost sales. Since hospitals could not simply abandon emergency demand, unmet requirements were treated as backorders for planning, while clinical substitutions were recorded separately.

Outcome

The revised policy improved availability without relying solely on a large universal safety factor. Supplier lead-time variance declined after performance feedback and transport redesign. Inventory increased for the critical connector but declined for less critical items after differentiated service targets were introduced.

Learning Points

The case demonstrates that service targets must be defined precisely, lead-time variability may dominate safety stock, and correlation can invalidate an apparently sound formula. It also shows that inventory policy is a cross-functional governance decision involving procurement, logistics, finance, clinical operations, and data management.

Technology, Governance, and Contemporary Practice

Enterprise resource planning and warehouse-management systems can calculate inventory position continuously, trigger replenishment, and provide supplier-performance data. Machine-learning forecasts may improve the mean demand estimate, but they do not automatically produce a valid safety stock. The organization must quantify forecast-error distributions, detect drift, and preserve human accountability for critical items.

AI-supported inventory systems should be governed through version-controlled models, documented assumptions, audit trails, exception monitoring, and override policies. Managers should test whether algorithms systematically allocate lower service to particular customers or locations. In healthcare, food, aviation, and other regulated contexts, safety constraints may dominate cost minimization.

Best practice is not maximum automation. It is transparent automation combined with appropriate segmentation, data quality controls, scenario testing, and clear ownership of service consequences.

Challenges and Limitations

The principal limitations of simple reorder-point models are:

• demand may not be independent or stationary;
• lead time may depend on demand and capacity conditions;
• normal approximations may fail for intermittent or highly skewed demand;
• stockout observations may censor true demand;
• suppliers may impose minimum quantities or allocation rules;
• forecast error and parameter uncertainty are often omitted;
• service measures may be inconsistent across departments;
• inventory records may contain timing and transaction errors.

When these limitations are material, simulation, stochastic programming, robust optimization, or empirical distribution methods may be more appropriate. Sophisticated models are not automatically superior: a transparent model with reliable inputs often outperforms a complex model that users do not understand or maintain.

Chapter Summary

Uncertainty makes lead-time demand the central quantity in continuous-review inventory control. A reorder point combines expected lead-time demand with safety stock. With fixed lead time, demand variability is accumulated over time; with variable lead time, both demand variance and lead-time variance contribute to risk. When demand and lead time are correlated, independence formulas may materially understate required protection.

The (Q,R) policy separates order quantity from reorder point. Service must be specified carefully: cycle service level measures the probability of avoiding a stockout during a cycle, whereas fill rate measures the proportion of demand supplied immediately. Lost-sales and backorder assumptions change both system behavior and managerial consequences. Effective design requires sound data, appropriate distributions, empirical validation, segmentation by criticality, and ongoing governance.

Key Terms

Lead-time demand: Total demand occurring while replenishment is in transit.

Continuous review: Inventory monitoring conducted continuously or after each transaction.

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

Reorder point: Inventory-position level that triggers replenishment.

Safety stock: Inventory held to protect against uncertainty.

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

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

Stockout: Condition in which available inventory cannot satisfy current demand.

Lost sales: Demand that disappears because it cannot be supplied.

Backorder: Unfilled demand retained as a commitment for later delivery.

Lead-time variability: Dispersion in replenishment time.

Demand variability: Dispersion in demand over a specified period.

Service quantile: Distribution percentile used to set a protection level.

Forecast error: Difference between realized demand and forecast demand.

Censored demand: True demand obscured because inventory was unavailable.

Cycle stock: Inventory arising from replenishment order quantity.

Demand–lead-time correlation: Statistical association between demand conditions and replenishment time.

Empirical distribution: Distribution estimated directly from observed data.

Inventory segmentation: Differentiation of policies by value, criticality, or demand characteristics.

Review Questions

Short Answer Questions

1. Define inventory position and explain why backorders are subtracted.
2. What is the difference between safety stock and cycle stock?
3. Why does demand standard deviation grow with the square root of fixed lead time under independent demand?
4. Distinguish cycle service level from fill rate.
5. Explain one operational difference between lost sales and backorders.

Long Answer Questions

1. Explain how a continuous-review (Q,R) policy operates and discuss its assumptions.
2. Derive and interpret the variance formula for lead-time demand when demand and lead time are independent.
3. Critically evaluate the use of the normal distribution for inventory decisions.
4. Discuss how stockout-censored demand can bias safety-stock calculations.
5. Explain how technology and governance should be combined in AI-supported inventory control.

Analytical / Application Questions

1. Daily demand has mean 80 units and standard deviation 20 units; lead time is fixed at 5 days. Calculate the reorder point for a 95% cycle service level and state all assumptions.
2. Compare the inventory implications of a lost-sales assumption and a backorder assumption for a critical spare part. Recommend a policy and justify it.
3. A supplier has average lead time of 6 days but high lead-time variance. Design an analytical investigation to determine whether reducing lead-time variability is more economical than increasing safety stock.

Multiple Choice Questions with Answers

Question 1

In a continuous-review system, inventory position equals:

A. On-hand inventory plus backorders

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

C. On-order inventory minus on-hand inventory

D. Demand minus safety stock

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

Question 2

If daily demand is independent and lead time is fixed at L periods, the standard deviation of lead-time demand is:

A. σ_D L

B. σ_D/L

C. σ_D√L

D. σ_D + L

Correct Answer: C. σ_D√L

Question 3

For a normally distributed lead-time demand, the reorder point is commonly calculated as:

A. μ_L − zσ_L

B. μ_L + zσ_L

C. z/μ_L + σ_L

D. μ_Lσ_L/z

Correct Answer: B. μ_L + zσ_L

Question 4

A 95% cycle service level primarily means:

A. 95% of units are always filled

B. There is a 95% probability of no stockout during a replenishment cycle

C. Lead time is reduced by 95%

D. 95% of inventory is safety stock

Correct Answer: B. There is a 95% probability of no stockout during a replenishment cycle

Question 5

Which condition makes the independent demand–lead-time variance formula potentially underestimate risk?

A. Negative safety stock

B. Positive correlation between demand and lead time

C. Constant demand and lead time

D. Lower ordering cost

Correct Answer: B. Positive correlation between demand and lead time

Question 6

In a lost-sales system, unmet demand generally:

A. Becomes a backorder automatically

B. Disappears rather than waiting for later fulfillment

C. Increases inventory position

D. Reduces lead time

Correct Answer: B. Disappears rather than waiting for later fulfillment

Question 7

If mean daily demand is 80 units and fixed lead time is 5 days, mean lead-time demand is:

A. 16 units

B. 85 units

C. 400 units

D. 640 units

Correct Answer: C. 400 units

Question 8

Which measure focuses on the proportion of demand satisfied immediately from stock?

A. Fill rate

B. Lead-time variance

C. Cycle stock

D. Order frequency

Correct Answer: A. Fill rate

Question 9

The principal purpose of safety stock is to:

A. Eliminate all ordering costs

B. Protect against uncertainty in demand and/or lead time

C. Replace cycle stock in every system

D. Guarantee zero inventory

Correct Answer: B. Protect against uncertainty in demand and/or lead time

Question 10

Which data problem can cause a firm to underestimate true demand?

A. Stockout-censored demand

B. Recording supplier invoices

C. Measuring package size

D. Using inventory position

Correct Answer: A. Stockout-censored demand