Inventory Under Uncertainty

Introduction

Inventory decisions are rarely made under conditions of complete certainty. Customer demand fluctuates, suppliers deliver early or late, transportation time varies, and production processes may generate unexpected requirements. A replenishment policy that is adequate when demand and lead time are fixed can produce stockouts or excessive inventory when uncertainty is present.

Inventory under uncertainty is therefore concerned with balancing two competing costs: the cost of holding additional stock and the cost of failing to satisfy demand. The central managerial problem is not to eliminate all stockouts at any cost, but to select an economically and operationally appropriate level of protection.

This chapter develops methods for determining safety stock, service levels and reorder points when demand, lead time, or both are variable. It also explains continuous-review and periodic-review systems, probability-based stockout risk, assumptions behind common formulas, and the limitations of statistical inventory models.

Learning Objectives

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

1. Explain why uncertainty creates a need for safety stock.
2. Distinguish cycle service level, fill rate and stockout probability.
3. Calculate safety stock and reorder points under variable demand and lead time.
4. Apply continuous-review and periodic-review policies.
5. Interpret the effect of variability, lead time and target service level on inventory.
6. Evaluate the assumptions and limitations of probability-based inventory models.
7. Recommend practical controls for uncertain supply and demand environments.

1. Sources and Types of Inventory Uncertainty

Uncertainty may arise from several sources:

• Demand uncertainty: actual customer demand differs from the forecast.
• Lead-time uncertainty: the interval between placing an order and receiving it is variable.
• Supply uncertainty: delivered quantity, quality or timing differs from the order.
• Forecast uncertainty: the forecast itself contains estimation error.
• Process uncertainty: production yield, scrap, machine availability or usage rates fluctuate.
• Information uncertainty: inventory records, demand signals or supplier information are inaccurate.

Two distinctions are especially important. First, variability is not the same as uncertainty. Variability describes a distribution of possible outcomes; uncertainty may also include incomplete knowledge about the distribution itself. Second, independent variation differs from correlated variation. If demand increases precisely when suppliers become slower, the combined risk is greater than it would be under independence.

A useful inventory position is:

where IP is inventory position in units, OH is on-hand inventory, OO is inventory on order, and BO is backorders or unmet demand. Inventory position, rather than physical stock alone, is normally used to trigger replenishment.

2. Service Levels and Stockout Risk

A service level expresses the desired probability or extent of satisfying demand. Different service measures answer different questions and must not be used interchangeably.

2.1 Cycle service level

The cycle service level (CSL) is the probability that no stockout occurs during one replenishment cycle. If the probability of a stockout during a cycle is P(stockout), then:

A 95% CSL means that approximately 95% of replenishment cycles are expected to pass without a stockout. It does not mean that 95% of individual units of demand will be filled.

2.2 Fill rate

The fill rate is the proportion of total demand supplied immediately from available inventory:

A 95% fill rate means that 95% of demand units are supplied immediately, on average. A system can have a high CSL but a lower fill rate if infrequent stockouts are severe. Conversely, frequent but very small shortages may produce a relatively high fill rate.

2.3 Stockout probability and critical fractile

For a single replenishment or one-period decision, the critical fractile is:

where Cu is the cost of underestimating demand by one unit and Co is the cost of overestimating demand by one unit. This relationship is mainly associated with single-period or newsvendor decisions. It should not be confused with the reorder-point service-level approach used in continuous replenishment.

2.4 Selecting a service target

Service targets should reflect the consequences of failure. Critical medical items, production bottlenecks and contractual customer orders may justify high service targets. Low-value, easily substituted or obsolete-prone items may require lower targets. A uniform 99% target for every item is usually economically irrational because the cost of achieving the final percentage points rises sharply.

3. Safety Stock and the Reorder Point

The reorder point (ROP) is the inventory position at which a replenishment order is placed. Its purpose is to cover expected demand during replenishment lead time plus a protection allowance:

For constant demand rate d per period and constant lead time L periods:

where SS is safety stock in units. The first term covers expected lead-time demand. Safety stock covers random deviations above that expectation.

A common statistical form is:

and therefore:

Here, μ_LT is mean demand during lead time, σ_LT is the standard deviation of lead-time demand, and z is the standard-normal quantile corresponding to the selected cycle service level. For a normally distributed lead-time demand variable, approximately 90%, 95% and 99% CSLs correspond to z-values of 1.282, 1.645 and 2.326 respectively.

The formula assumes that lead-time demand is approximately normal, observations are reasonably stable, and the chosen z-value is appropriate. It is less reliable for intermittent demand, strongly skewed demand, short data histories or structural changes in the market.

4. Variability in Demand with Constant Lead Time

Suppose demand in each period has mean μ_d and standard deviation σ_d, and lead time is a constant L periods. If demands in different periods are independent, then:

Thus:

The square-root relationship is important. Average demand increases directly with lead time, but independent demand uncertainty increases with the square root of lead time. The independence assumption is essential. If demand is positively correlated across periods, the actual standard deviation may be larger.

Worked Example 1: Variable demand and constant lead time

A component has average daily demand of 80 units and a daily standard deviation of 20 units. Lead time is constant at 4 working days. The firm wants a 95% cycle service level, using z = 1.645. Calculate safety stock and reorder point.

Mean lead-time demand:

Standard deviation of lead-time demand:

Safety stock:

Reorder point:

The inventory position should therefore trigger replenishment at approximately 386 units. The result is not a guarantee of no stockout; it represents an estimated 95% probability of avoiding a stockout during a replenishment cycle under the stated assumptions.

5. Variable Demand and Variable Lead Time

When both demand and lead time vary, the variance of lead-time demand must include both sources of uncertainty. Let demand per period have mean μ_d and variance σ_d², and let lead time have mean μ_L and variance σ_L². If demand and lead time are independent, an important approximation is:

Therefore:

The two variance components have clear interpretations. The term μ_Lσ_d² represents demand variability over the average lead time. The term μ_d²σ_L² represents the effect of lead-time variability on demand exposure.

If demand and lead time are correlated, an additional covariance term is needed. A commonly used approximation is:

where Cov(D,L) is the covariance between demand and lead time in compatible units. Positive covariance increases risk; negative covariance reduces it. Because demand during a random lead time is a compound random variable, these formulas are approximations unless the underlying distributions and timing assumptions are carefully modelled.

Worked Example 2: Variable demand and lead time

Daily demand has mean 100 units and standard deviation 25 units. Lead time has mean 3 days and standard deviation 1 day. Assume independence and a 95% CSL. Then:

Safety stock is:

The reorder point is:

This example illustrates that lead-time variation can dominate the uncertainty. The lead-time component contributes 10,000 units² to variance, compared with 1,875 units² from demand variation.

6. Continuous-Review Systems

In a continuous-review, or (Q,R), system, inventory position is monitored continuously. When it reaches the reorder point R, an order of approximately Q units is placed.

The reorder point is normally based on demand during lead time. The order quantity may be determined using an economic order quantity, supplier minimum, container size, production lot size or an optimization model. Under basic assumptions, the economic order quantity is:

where D is annual demand in units per year, S is ordering or setup cost per order, and H is annual holding cost per unit. This formula assumes known demand, constant lead time, no quantity discounts, instantaneous replenishment and no shortages. Under uncertainty, Q* does not replace the safety-stock calculation; the two decisions address different issues.

Advantages of continuous review include rapid response, lower average cycle stock for some demand patterns and direct control of inventory position. Disadvantages include the need for reliable transaction data, frequent monitoring and system integration. It is especially suitable for high-value, critical or fast-moving items.

7. Periodic-Review Systems

In a periodic-review, or (P,S), system, inventory is reviewed at fixed intervals of P periods. At each review, the order raises inventory position to an order-up-to level S. If the current inventory position is IP, the order quantity is:

The protection period is the review period plus lead time, because a decision cannot be made again until the next review. For constant review interval P and lead time L, a basic order-up-to formula is:

where σ_P+L is the standard deviation of demand over the protection period. With independent, identically distributed period demand:

Periodic review generally requires more safety stock than continuous review for the same target service level because demand occurring between reviews is exposed before another order can be placed. It is, however, convenient where suppliers visit on fixed schedules, administrative ordering costs are significant or several items are ordered together.

Worked Example 3: Periodic review

Daily demand has mean 50 units and standard deviation 10 units. A warehouse reviews inventory every 7 days and supplier lead time is 5 days. The target CSL is 95%, with z = 1.645. The protection period is 12 days.

The order-up-to level is therefore approximately 657 units. If inventory position at review is 210 units, the order is approximately:

The order should be rounded in accordance with packaging, minimum-order and supplier constraints.

8. Choosing Between Policy Types

The decision should consider not only mathematics but also transaction cost, system capability, supplier arrangements, item criticality and the cost of monitoring.

9. Data, Forecasting and Statistical Governance

Safety-stock calculations are only as credible as their data. A professional implementation should:

1. Define the demand unit and time bucket consistently.
2. Separate genuine demand from cancellations, returns and abnormal one-off events.
3. Measure actual supplier lead time from order release to usable receipt.
4. Distinguish requested, promised and actual delivery dates.
5. Recalculate parameters when product, supplier or process conditions change.
6. Use rolling performance measures such as forecast error, bias and lead-time variance.
7. Investigate outliers rather than automatically deleting them.

Forecast bias is particularly important. A forecast that is consistently low can create stockouts even when its standard deviation appears acceptable. A simple forecast error for period t is:

Mean absolute deviation can be calculated as:

where n is the number of observations. Standard deviation is more sensitive to large errors, whereas MAD is easier to interpret operationally. Safety stock should be based on residual forecast error where a forecast is already used, rather than blindly using the standard deviation of raw demand.

10. Detailed Applied Case: Northstar Pumps

Background

Northstar Pumps manufactures industrial water pumps. A seal assembly is a low-cost but production-critical component. A shortage stops final assembly, while excess stock has moderate holding cost and a risk of deterioration.

Problem

The existing policy uses a reorder point based only on average demand: 1,200 units. Daily demand averages 200 units with a standard deviation of 50 units. Supplier lead time averages 4 days with a standard deviation of 1.5 days. Demand and lead time are treated as independent. Management wants a 95% cycle service level.

Analysis

Mean lead-time demand is:

Variance of lead-time demand is approximated by:

Thus:

Safety stock is:

The statistically estimated reorder point is:

Approach and solution

The company raises the reorder point from 1,200 to 1,320 units, but does not immediately treat the result as an unquestionable optimum. It first validates supplier lead-time records, checks whether demand and lead time are correlated, and confirms that the normal approximation is reasonable. It also negotiates a lead-time reduction and introduces supplier performance monitoring.

Outcome

The initial policy improves protection by recognizing lead-time variability. Supplier improvement subsequently reduces both average lead time and its standard deviation, allowing Northstar to reduce safety stock without reducing the target service level.

Learning points

• Average lead-time demand alone does not protect against variation.
• Lead-time variability can be more influential than demand variability.
• Statistical outputs require operational validation.
• Service-level decisions should be linked to the cost and consequence of shortage.
• Improving the process may be more economical than holding more inventory.

This is an illustrative case, not a report of an actual company.

11. Limitations and Advanced Considerations

Non-normal and intermittent demand

The normal model can assign positive probability to negative demand and may poorly represent intermittent or lumpy demand. For slow-moving items, empirical quantiles, Poisson-type models, Croston-style forecasting approaches or simulation may be more appropriate, depending on the demand process.

Multiple echelons

In a multi-echelon supply chain, safety stock should not be independently added at every location without considering upstream and downstream relationships. Pooling risk at a central location may reduce total safety stock, although it can increase response time.

Capacity and constrained supply

If supplier capacity is limited, raising the reorder point may not solve the problem. Allocation rules, priority classes, alternate sources and production flexibility may be necessary.

Correlation and common-cause risk

Demand surges, transport disruption and supplier failure may occur together. Historical variance based on normal periods can understate risk during such events. Scenario analysis and stress testing should complement statistical calculations.

Simulation

When distributions, policies or interactions are complex, discrete-event or Monte Carlo simulation can estimate stockout frequency, fill rate and working capital under alternative policies. Simulation does not remove uncertainty; it makes assumptions explicit and allows policies to be compared consistently.

12. Technology, AI and Governance

Enterprise resource planning systems can calculate inventory position, lead-time statistics and exception alerts in near real time. Machine-learning models may improve demand forecasts by using promotions, weather, customer behaviour or external indicators. However, a more sophisticated forecast does not automatically produce a better inventory policy.

AI-supported inventory decisions require data-quality controls, explainability, monitoring for forecast drift and human accountability for critical items. Automated recommendations should be subject to approval rules, especially where stockouts affect safety, regulated products or contractual obligations. Personal data used to infer demand must be governed according to applicable privacy requirements. Audit trails should record the data, model version, parameters and decision associated with each policy change.

13. Best-Practice Implementation Framework

A practical implementation can follow these steps:

1. Classify items by value, criticality, demand pattern and substitutability.
2. Select a service measure appropriate to the business problem.
3. Establish clean demand and lead-time histories.
4. Estimate mean and variability using a suitable forecasting method.
5. Select continuous or periodic review based on operational requirements.
6. Calculate the protection level and document assumptions.
7. Validate results against capacity, supplier constraints and physical storage.
8. Pilot the policy on a representative item group.
9. Monitor CSL, fill rate, inventory turns, stockout duration and obsolete stock.
10. Review parameters after structural changes rather than relying only on calendar updates.

Chapter Summary

Inventory under uncertainty requires protection against variation in demand, lead time and supply. Safety stock is not expected demand; it is a buffer against deviations from expected demand. For normally distributed lead-time demand, safety stock is estimated as zσ_LT, and the reorder point is expected lead-time demand plus safety stock. When demand and lead time are both variable, their separate variance contributions must be considered, and correlation may materially change the result.

Cycle service level measures the probability of avoiding a stockout during a cycle, whereas fill rate measures the proportion of demand supplied immediately. Continuous-review systems trigger a fixed or calculated order at a reorder point. Periodic-review systems order up to a target level and must protect demand over the review interval plus lead time. Statistical formulas are useful decision aids, but their validity depends on data quality, distributional assumptions, independence, stable processes and correct service definitions. Effective inventory management combines quantitative analysis with supplier improvement, forecasting governance, segmentation, monitoring and managerial judgment.

Key Terms

Safety stock: Inventory held to protect against uncertainty in demand or replenishment.

Reorder point: Inventory position at which a replenishment order is triggered.

Lead-time demand: Demand expected during the replenishment lead time.

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

Cycle service level: Probability of completing a replenishment cycle without a stockout.

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

Stockout: Condition in which available inventory cannot satisfy demand.

Demand variability: Variation in demand across periods or customers.

Lead-time variability: Variation in the time required to receive replenishment.

Protection period: Time span for which inventory must provide coverage.

Continuous review: Inventory policy monitored continuously or transaction by transaction.

Periodic review: Inventory policy assessed at fixed time intervals.

Order-up-to level: Target inventory position in a periodic-review system.

Service-level quantile: Standard-normal value associated with a selected probability.

Forecast error: Difference between actual demand and forecast demand.

Safety factor: The z-value used to translate standard deviation into safety stock.

Covariance: Measure of how two variables vary together.

Intermittent demand: Demand occurring irregularly with periods of zero demand.

Inventory policy: Rules specifying when and how much to replenish.

Review Questions

Short Answer Questions

1. Why is safety stock required when average demand is known?
2. Distinguish cycle service level from fill rate.
3. What is meant by inventory position?
4. Why does independent demand variability grow with the square root of lead time?
5. State one advantage and one disadvantage of periodic review.

Long Answer Questions

1. Explain the logic of the reorder point model and discuss its principal assumptions.
2. Compare continuous-review and periodic-review systems in terms of protection period, information requirements and application.
3. Explain how variable lead time affects safety stock and why supplier improvement can be preferable to simply increasing inventory.
4. Discuss the limitations of using a normal distribution for inventory demand.
5. Explain how cycle service level, fill rate and shortage cost should influence service-target selection.

Analytical / Application Questions

1. Daily demand has mean 120 units and standard deviation 30 units. Lead time is constant at 5 days. Calculate safety stock and reorder point for a 95% cycle service level using z = 1.645.
2. Daily demand has mean 75 units and standard deviation 15 units. Mean lead time is 4 days with a standard deviation of 1 day. Assuming independence, calculate the approximate 90% reorder point using z = 1.282.
3. A firm reviews stock every 6 days, has a lead time of 4 days, and daily demand has mean 40 units and standard deviation 8 units. Determine the 95% order-up-to level and explain why the protection period is not merely the supplier lead time.

Multiple Choice Questions with Answers

Question 1

What does the cycle service level measure?

A. The percentage of inventory carrying cost recovered

B. The probability of avoiding a stockout during a replenishment cycle

C. The percentage of orders placed electronically

D. The average number of units ordered per cycle

Correct Answer: B. The probability of avoiding a stockout during a replenishment cycle

Question 2

A product has mean lead-time demand of 500 units and lead-time-demand standard deviation of 40 units. At a 95% cycle service level, using z = 1.645, what is the safety stock?

A. 24 units

B. 40 units

C. 66 units

D. 540 units

Correct Answer: C. 66 units

Question 3

If average daily demand is 60 units, constant lead time is 3 days, and safety stock is 50 units, what is the reorder point?

A. 110 units

B. 180 units

C. 230 units

D. 300 units

Correct Answer: C. 230 units

Question 4

In a continuous-review (Q,R) system, what normally triggers an order?

A. The calendar reaching the end of the month

B. Inventory position falling to or below the reorder point

C. The supplier increasing its price

D. The annual demand reaching zero

Correct Answer: B. Inventory position falling to or below the reorder point

Question 5

In a periodic-review system, why is the protection period equal to review interval plus lead time?

A. Because demand is assumed to be zero during lead time

B. Because inventory is reviewed continuously during the interval

C. Because no new replenishment decision can normally be made until the next review, and delivery then requires lead time

D. Because order quantity is always equal to annual demand

Correct Answer: C. Because no new replenishment decision can normally be made until the next review, and delivery then requires lead time

Question 6

Which variance component in the variable-demand-and-lead-time approximation represents lead-time uncertainty?

A. μ_Lσ_d²

B. μ_d²σ_L²

C. μ_dμ_L

D. zσ_LT

Correct Answer: B. μ_d²σ_L²

Question 7

A 95% fill rate means that, on average:

A. 95% of replenishment cycles have no stockout

B. 95% of demand units are supplied immediately from stock

C. 95% of suppliers deliver early

D. 95% of inventory is safety stock

Correct Answer: B. 95% of demand units are supplied immediately from stock

Question 8

Daily demand has mean 50 units and standard deviation 10 units. For a 12-day protection period with independent demand, what is the standard deviation of protection-period demand?

A. 10 units

B. 12 units

C. 34.6 units approximately

D. 120 units

Correct Answer: C. 34.6 units approximately

Question 9

Which condition most directly makes the basic normal safety-stock model less reliable?

A. Stable, high-volume demand

B. Accurate lead-time records

C. Intermittent and strongly skewed demand

D. A clearly defined service target

Correct Answer: C. Intermittent and strongly skewed demand

Question 10

If demand and lead time are positively correlated, what is the likely effect on lead-time-demand variance compared with an independence assumption?

A. It decreases to zero

B. It generally increases because a positive covariance contribution is added

C. It becomes equal to average demand

D. It is unaffected in every case

Correct Answer: B. It generally increases because a positive covariance contribution is added