Cold-Chain Planning and Inventory Control

Introduction

Cold-chain inventory decisions differ fundamentally from conventional inventory decisions. A unit of chilled, frozen, or fresh product is not simply “in stock” or “out of stock.” Its value depends on remaining shelf life, temperature history, quality grade, customer requirements, and the probability that it can be sold before deterioration. Consequently, managers must balance availability against freshness, waste, working capital, energy use, markdowns, and regulatory compliance.

The central planning problem is to determine how much to procure, where to position it, when to replenish it, and how to allocate it among customers while demand and usable life are uncertain. This chapter develops an integrated approach using demand forecasting, service-level planning, FEFO issuing, safety stock, spoilage and markdown models, substitution, and perishable-inventory methods such as the newsvendor model.

Learning Objectives

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

1. Forecast demand for temperature-sensitive products and recognize the effects of bias, seasonality, promotions, and weather.
2. Apply FEFO and age-based inventory visibility to operational decisions.
3. Calculate safety stock and reorder points under specified service-level assumptions.
4. Estimate expected spoilage, markdown exposure, and working-capital consequences.
5. Use substitution, critical-fractile, and newsvendor logic for short-life products.
6. Evaluate cold-chain inventory policies using availability, freshness, waste, cost, and compliance measures.

1. The Nature of Perishable Inventory

Perishability may be physical, when the product becomes unsafe or unusable, or economic, when it remains technically saleable but customers reject it because of reduced freshness or quality. A product’s effective inventory position is therefore multidimensional:

• Quantity: units physically available.
• Age: elapsed time since production or receipt.
• Remaining shelf life: usable time before expiry or quality failure.
• Temperature condition: whether storage and transport remained within specification.
• Disposition: saleable, restricted, quarantined, discounted, donated, returned, or discarded.

A useful planning distinction is between chronological inventory and effective inventory. Chronological inventory counts all units, whereas effective inventory counts only units that satisfy a customer’s remaining-life requirement. If a retailer requires at least four days of remaining life and a batch has only three days left, that batch may be physically present but unavailable for that customer.

Cold-chain planning also involves several interacting lead times: supplier production lead time, transport lead time, receiving and quality-inspection time, put-away time, and customer order-to-delivery time. Each consumes part of the product’s usable life. A replenishment policy that ignores this “lead-time shelf-life loss” can create apparently adequate stock that is commercially obsolete.

2. Demand Forecasting for Perishables

Forecasting should begin with a clean demand history. Sales data must be separated from lost sales, cancellations, substitutions, returns, stockouts, and abnormal promotional events. Observed sales during a stockout are not true demand; they are censored demand.

2.1 Basic forecasting methods

For a moving average of the previous n periods:

where D_t is demand in period t and F_{t+1} is the forecast for the next period. This method is appropriate when demand is relatively stable and recent observations are representative. It reacts slowly to rapid structural changes.

Exponential smoothing gives greater weight to recent demand:

where 0 ≤ α ≤ 1. A high α responds quickly but may overreact to noise; a low α produces a smoother forecast. The units of F and D must be identical, such as cases per day.

When seasonality is material, managers should use seasonal indices, decomposition, regression, or hierarchical forecasting. Perishable demand commonly varies with weekday, holiday, temperature, rainfall, school calendars, tourist flows, and promotional price. Causal variables should be included only when they are available at the time of forecasting and have demonstrated predictive value.

2.2 Forecast evaluation

Mean absolute error is:

Mean absolute percentage error is:

MAPE is problematic when actual demand is zero or very small. Forecast error standard deviation is more useful for safety-stock design when errors are approximately stable and independent. Bias should also be monitored:

A persistent positive bias means actual demand systematically exceeds the forecast, while a persistent negative bias indicates overforecasting. Forecast accuracy should be measured at the decision level: SKU–location–day, SKU–customer segment, or product family, rather than only at an aggregate level.

3. FEFO and Age-Based Inventory Control

First-expired, first-out (FEFO) issues the unit with the earliest expiry date first. It differs from FIFO, which issues the unit received first. FIFO is acceptable only when receipt order reliably corresponds to expiry order. In a multi-supplier cold chain, this assumption often fails.

FEFO requires lot-level records containing receipt date, production date, expiry date, remaining shelf life, quantity, temperature status, and customer minimum-life requirements. A practical allocation sequence is:

1. Remove quarantined or temperature-abused lots from available stock.
2. Rank eligible lots by earliest expiry.
3. Apply customer minimum-life rules.
4. Allocate the earliest-expiring eligible lot first.
5. Record the resulting age and remaining-life distribution.

FEFO does not mean shipping the oldest product regardless of customer need. If a customer requires ten days of remaining life, a lot with eight days remaining is not eligible even if it expires soonest. In such cases, FEFO operates within the feasible eligibility set.

Age-based control can be represented by an age vector, such as (20, 35, 15) units with one, two, and three days of life remaining. This is more informative than a single stock balance because two inventories of 70 units can have very different service and waste risks.

4. Service Levels and Safety Stock

A cycle service level is the probability of avoiding a stockout during a replenishment cycle. It differs from fill rate, which is the proportion of demand units supplied immediately. A policy can have a high cycle service level but a lower fill rate when occasional stockouts are large.

For normally distributed demand during lead time, the reorder point is:

where μ_LT is mean demand during lead time, σ_LT is its standard deviation, and z is the standard-normal quantile associated with the target cycle service level. Safety stock is:

If daily demand has mean μ_d and standard deviation σ_d, and lead time is a constant L days, then:

These formulas assume independent daily demand, a stable distribution, known lead time, and no significant forecast bias. If lead time is also variable and demand and lead time are independent:

where E[L] is mean lead time and σ_L is lead-time standard deviation. Correlation between demand and lead time requires a covariance term and should not be ignored during weather events or capacity shortages.

Worked example: reorder point

A distribution center sells 120 cases per day on average. Daily standard deviation is 30 cases, lead time is four days, and the desired cycle service level is 95%, for which z ≈ 1.645.

Mean lead-time demand is 4 × 120 = 480 cases. Lead-time standard deviation is √4 × 30 = 60 cases. Therefore:

The result is a statistical trigger, not an instruction to order 579 cases. Order quantity must also consider case packs, minimum order quantities, capacity, remaining shelf life, and expected waste.

For perishables, safety stock should be constrained by usable life. If replenishment arrives with only five days of acceptable life and average demand is 120 cases per day, a calculated stock position of 579 cases may be operationally impossible to sell before expiry. Service-level targets must therefore be integrated with shelf-life feasibility.

5. Spoilage, Markdown, and Economic Inventory

Let Q be the quantity ordered, D random demand during the selling period, and S the maximum quantity that remains saleable during that period. Actual sales are:

If S is deterministic and demand is uncertain, expected waste is:

provided Q does not exceed the saleable-life capacity and unsold units are discarded. If shelf-life capacity is also random, waste is better represented as:

In practice, waste should be measured by lot and cause: expiry, temperature excursion, damage, quality rejection, forecast error, handling loss, or customer rejection. A single waste percentage conceals actionable differences.

Markdowns can recover value from aging inventory but may alter demand and perceived quality. A simple expected-cost expression is:

where c is purchase cost per unit, h holding cost, I average inventory, p disposal or spoilage cost, W waste, m markdown cost or margin loss, M markdown units, r regular selling revenue, and v recovery value from donation, processing, or salvage. These terms must be defined consistently: either all as costs or all as contribution-margin consequences.

Markdown timing is a control decision. Late markdowns protect regular price but increase expiry risk; early markdowns improve sell-through but may train customers to wait for discounts. Dynamic markdown systems should use remaining life, demand elasticity, inventory age, local demand, and downstream capacity.

6. Substitution and Availability Management

Substitution allows a customer’s need to be met by an alternative SKU, pack size, grade, or temperature state. It can reduce lost sales but may create margin loss, allergen risk, customer dissatisfaction, or additional handling.

A substitution policy should specify:

• acceptable product equivalence;
• customer consent requirements;
• allergen, dietary, and regulatory restrictions;
• price and margin treatment;
• traceability and lot recording;
• rules for refrigerated versus frozen alternatives.

For planning, demand may be represented by a substitution matrix. If product j can satisfy a fraction a_ij of demand for product i, effective available supply for demand i includes eligible substitute stock. However, substitution is not additive without limits: one unit cannot satisfy two customers, and using a substitute for one demand class may create shortage in its primary class.

A useful managerial distinction is between physical availability and functional availability. Physical availability asks whether units exist; functional availability asks whether the right customer can accept them at the required quality, temperature, pack, and remaining shelf life.

7. The Newsvendor Model for Short-Life Products

The single-period newsvendor model applies when an order is selected before uncertain demand is realized and leftover product has a different value from a lost sale. Let:

• c = unit acquisition cost;
• p = selling price;
• v = recovery value of an unsold unit;
• g = shortage or lost-margin cost per unmet unit;
• D = random demand;
• Q = order quantity.

With no separate shortage penalty, the underage cost is Cu = p − c, and the overage cost is Co = c − v. The critical fractile is:

or, equivalently,

when p > c > v. If a shortage penalty g is explicitly included, use Cu = p − c + g only when the penalty represents an incremental consequence not already included in lost contribution margin.

Worked example: fresh product order

A processor sells meal kits for $12 per unit. Acquisition and preparation cost is $7, and an unsold kit can be recovered for $2. Thus:

The critical fractile is 5 ÷ (5 + 5) = 0.50. If demand is normally distributed with mean 1,000 units and standard deviation 200 units, the optimal quantity is the 50th percentile, approximately 1,000 units.

This result is conditional on the model assumptions: one selling period, known distribution, one ordering opportunity, constant unit economics, and no capacity or shelf-life constraint beyond the period. Real cold chains often require an extended model with multiple replenishment opportunities, aging states, markdowns, and correlated demand.

8. Integrated Planning Architecture

A robust planning process links tactical and operational decisions:

Figure: An iterative cold-chain planning cycle linking demand, inventory, execution, and learning.

The cycle begins with demand sensing and proceeds through probabilistic forecasting, procurement and capacity planning, FEFO allocation, and performance review. The review stage should update parameters rather than merely report outcomes. For example, repeated expiry may require a lower order-up-to level, shorter procurement cycles, improved supplier shelf life, or a revised markdown trigger.

9. Illustrative Case: Regional Fresh Meal Distributor

Background

This illustrative case concerns FreshRoute, a distributor serving hospitals, universities, and convenience stores. Its chilled meal kits have a seven-day supplier shelf life. Customers require at least three days of remaining life at delivery. Demand is higher on weekdays, promotions create spikes, and late deliveries occasionally reduce usable life.

Problem

FreshRoute carried excessive average inventory but still experienced stockouts on high-demand days. Waste averaged 8% of receipts, while the reported fill rate was 96%. Investigation showed that the fill-rate calculation counted short-life units as available even when customers rejected them. FIFO issuing also left earlier-expiring lots in storage.

Analysis

The firm segmented demand by customer and weekday, removed stockout-censored observations, and estimated forecast error by SKU–location–day. It then introduced a functional-availability measure: a unit counted as available only if its temperature status was valid and its remaining life met the customer rule.

The planning team calculated reorder points using lead-time demand and forecast-error variation, but applied a shelf-life feasibility check. Orders exceeding expected sell-through before the customer minimum-life threshold were capped or redirected to a lower-life-tolerance channel. FEFO was implemented at lot level, and a two-stage markdown policy was introduced: a modest discount at four days remaining and a deeper discount at three days, subject to customer acceptance.

Approach and outcome

FreshRoute also permitted substitution between two nutritionally equivalent meal kits, but prohibited substitution across allergen categories. A dashboard tracked service level, functional fill rate, waste by cause, average remaining life at shipment, markdown recovery, temperature excursions, and inventory value.

After three planning cycles, the company reported lower expiry waste and fewer customer rejections. The apparent service level declined slightly because the new metric excluded noncompliant short-life stock, but customer-accepted fill rate improved. Working capital fell because aging inventory was identified earlier and replenishment was aligned with usable rather than nominal capacity.

Learning points

1. A high physical fill rate can conceal poor functional availability.
2. FEFO requires eligibility rules, not merely sorting by expiry date.
3. Safety stock must be bounded by shelf-life feasibility.
4. Substitution can improve service only when quality, allergen, and customer rules are explicit.
5. Metrics should expose trade-offs among service, waste, freshness, margin, and compliance.

10. Technology, Governance, and Contemporary Practice

Warehouse-management systems increasingly combine lot tracking, barcode or RFID identification, temperature sensors, and expiry alerts. IoT data can support exception-based control, but sensors do not automatically establish product safety; sensor calibration, placement, sampling frequency, data integrity, and validated time–temperature limits remain necessary.

Artificial-intelligence forecasting can detect nonlinear effects from weather, promotions, holidays, and local events. Its value depends on clean data, stable deployment, explainable overrides, and monitoring for drift. An algorithm that optimizes sales without incorporating expiry, food safety, or customer shelf-life constraints can increase waste or regulatory exposure.

Governance should define authority for quarantine release, markdown approval, substitution, donation, disposal, and temperature-excursion disposition. Traceability must connect supplier lot, transport record, storage location, customer shipment, and recall action. Food-safety decisions must follow applicable national regulations, validated hazard-control plans, and product-specific storage requirements; an inventory model cannot replace those controls.

Sustainability analysis should include product waste, refrigeration energy, packaging, reverse logistics, and disposal emissions. A lower-waste policy is not automatically superior if it causes excessive emergency transport or energy-intensive storage. Multi-objective evaluation is therefore preferable to a single cost metric.

Challenges and Limitations

The models in this chapter simplify reality. Demand may be intermittent, non-normal, or censored by stockouts. Forecast errors can be correlated across products during heat waves or disruptions. Shelf life is not always deterministic, and temperature abuse may reduce quality before the printed expiry date. Costs such as reputation damage, food-safety risk, and patient harm are difficult to monetize. The newsvendor model is elegant but limited for multi-period, multi-echelon systems with substitution and dynamic pricing.

Managers should therefore use models as disciplined approximations. They should test sensitivity to demand distribution, lead time, recovery value, service target, shelf-life loss, and forecast bias. Scenario analysis and simulation are often appropriate when analytical assumptions fail.

Best-Practice Checklist

• Forecast unconstrained demand, not only recorded sales.
• Segment by SKU, location, customer, and demand pattern.
• Track lot age, remaining life, and temperature status.
• Use FEFO within customer eligibility constraints.
• Distinguish cycle service level, fill rate, and functional availability.
• Calculate safety stock from lead-time uncertainty and validate shelf-life feasibility.
• Record waste and markdowns by cause and lot.
• Establish controlled substitution rules.
• Use newsvendor logic only when its single-period assumptions are credible.
• Review model performance continuously and govern overrides.

Chapter Summary

Cold-chain inventory control is a joint problem of quantity, time, quality, and uncertainty. Demand forecasting provides the expected consumption pattern, while forecast-error analysis supports safety stock and reorder points. FEFO protects freshness, but it must operate with customer minimum-life and temperature-status rules. Service-level formulas are useful only when their distributional and lead-time assumptions are understood, and calculated safety stock must be tested against shelf-life capacity.

Spoilage, markdowns, recovery value, and substitution determine the economic consequences of inventory decisions. The newsvendor model clarifies the balance between underage and overage costs for short-life products, while integrated planning connects forecasting, procurement, allocation, execution, and continuous review. Effective management evaluates functional availability rather than nominal stock and balances service, freshness, waste, working capital, sustainability, and regulatory compliance.

Key Terms

Perishable inventory: Stock whose physical or economic value declines with time.

Remaining shelf life: Usable time before expiry or unacceptable quality.

Functional availability: Inventory that meets quantity, quality, temperature, and customer eligibility requirements.

FEFO: First-expired, first-out issuing logic.

FIFO: First-in, first-out issuing based on receipt sequence.

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

Fill rate: Proportion of demand supplied immediately.

Safety stock: Inventory held to protect against uncertainty.

Reorder point: Inventory position that triggers replenishment.

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

Forecast bias: Systematic tendency to overforecast or underforecast.

Spoilage: Loss caused by expiry, deterioration, or unacceptable quality.

Markdown: Price reduction used to accelerate sale of aging inventory.

Substitution: Fulfilling demand with an approved alternative product.

Newsvendor model: Single-period model balancing overage and underage costs.

Critical fractile: Target demand percentile derived from underage and overage costs.

Underage cost: Economic consequence of ordering too little.

Overage cost: Economic consequence of ordering too much.

Temperature excursion: Departure from specified temperature conditions.

Review Questions

Short Answer Questions

1. Why is physical inventory different from functional availability in a cold chain?
2. What is the operational difference between FIFO and FEFO?
3. Define cycle service level and fill rate.
4. Why can MAPE be misleading for intermittent demand?
5. State two assumptions behind the basic reorder-point formula.

Long Answer Questions

1. Explain how demand forecasting, safety stock, and shelf-life feasibility should be integrated for chilled products.
2. Compare FEFO with FIFO and discuss the data requirements for reliable FEFO execution.
3. Explain the newsvendor model, including underage cost, overage cost, and the critical fractile.
4. Discuss how markdowns and substitution can reduce waste while creating new operational and ethical risks.
5. Design a performance-measurement system for a regional cold-chain distributor.

Analytical / Application Questions

1. Daily demand has a mean of 120 cases and standard deviation of 30 cases. Lead time is four days and the target cycle service level is 95% with z = 1.645. Calculate safety stock and reorder point, then explain why the result may need adjustment for a five-day shelf-life constraint.
2. A fresh product sells for $12, costs $7, and has a recovery value of $2. Demand is normally distributed with mean 1,000 and standard deviation 200. Calculate the critical fractile and recommended order quantity under the single-period newsvendor assumptions.
3. A warehouse has 100 units with two days of life remaining and 100 units with six days of life remaining. Customer A requires at least five days of life, while Customer B accepts two days. Develop a FEFO allocation plan and explain why total inventory alone is an inadequate availability measure.

Multiple Choice Questions with Answers

Question 1

Which measure describes the probability of avoiding a stockout during a replenishment cycle?

A. Cycle service level

B. Fill rate

C. Markdown recovery rate

D. Inventory turnover

Correct Answer: A. Cycle service level

Question 2

A product is issued according to earliest expiry date rather than receipt date. Which policy is being used?

A. LIFO

B. FIFO

C. FEFO

D. Cross-docking

Correct Answer: C. FEFO

Question 3

If daily demand is independent with standard deviation σ_d and lead time is a constant L, which expression gives lead-time demand standard deviation?

A. Lσ_d

B. √L × σ_d

C. σ_d ÷ L

D. L ÷ σ_d

Correct Answer: B. √L × σ_d

Question 4

In the basic newsvendor model, the overage cost represents the cost of:

A. An unmet customer order

B. An unsold unit relative to its recovery value

C. A supplier delivery delay only

D. A temperature sensor calibration

Correct Answer: B. An unsold unit relative to its recovery value

Question 5

A customer requires ten days of remaining life. A lot with eight days remaining is physically present but should normally be classified as:

A. Eligible inventory for that customer

B. Functional stock for every customer

C. Ineligible inventory for that customer

D. Safety stock automatically

Correct Answer: C. Ineligible inventory for that customer

Question 6

Which forecasting problem occurs when recorded sales are lower than true demand because the product stocked out?

A. Demand censoring

B. Seasonal decomposition

C. Overage costing

D. Lot consolidation

Correct Answer: A. Demand censoring

Question 7

For a product with selling price $10, cost $6, and recovery value $2, the basic critical fractile is:

A. 0.25

B. 0.40

C. 0.50

D. 0.67

Correct Answer: C. 0.50

Question 8

Which action is most consistent with responsible substitution governance?

A. Substitute any product without recording the change

B. Ignore allergen restrictions if the customer is short

C. Define equivalence, consent, and traceability rules

D. Use frozen products for all chilled orders automatically

Correct Answer: C. Define equivalence, consent, and traceability rules

Question 9

Why can a high reported fill rate coexist with poor customer service in a cold chain?

A. Fill rate always excludes stockouts

B. Short-life or temperature-compromised units may be counted as available

C. FEFO increases all stockouts

D. Markdowns eliminate functional availability

Correct Answer: B. Short-life or temperature-compromised units may be counted as available

Question 10

Which statement about safety stock is most accurate for perishables?

A. It should be maximized regardless of shelf life

B. It is unrelated to service-level targets

C. It should reflect uncertainty but also be checked against usable-life feasibility

D. It replaces the need for demand forecasting

Correct Answer: C. It should reflect uncertainty but also be checked against usable-life feasibility