Inventory and Replenishment Analytics

Introduction

Inventory is a buffer between uncertain supply and uncertain demand. It enables customer service, production continuity and transport consolidation, but it also ties up capital, consumes space, incurs insurance and handling costs, and creates obsolescence or deterioration risk. Inventory analytics therefore concerns more than deciding how much to order. It links demand classification, replenishment policy, uncertainty, service objectives, network structure and financial consequences.

A sound policy balances three outcomes:

1. Availability: the probability that demand can be met when required.
2. Working-capital efficiency: the amount of cash committed to stock.
3. Risk control: exposure to shortages, excess, obsolescence, expiry and supply disruption.

The central managerial challenge is that higher availability usually requires more inventory, while lower inventory increases the probability and cost of shortage. Analytics makes this trade-off explicit.

Learning objectives

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

• segment items using ABC and XYZ analysis;
• calculate and interpret EOQ, reorder points and safety stock;
• distinguish cycle service level from fill rate;
• select continuous-review and periodic-review replenishment policies;
• evaluate multi-echelon inventory decisions;
• adapt policies to perishability, lead-time uncertainty and intermittent demand;
• assess analytical limitations, data requirements and governance risks.

1. Inventory decisions and relevant cost

Inventory decisions operate at several time scales. Strategic decisions determine network locations and postponement policies. Tactical decisions establish target service levels, review frequencies and supplier contracts. Operational decisions determine order quantities, release dates, allocation and exception handling.

The relevant cost of an inventory policy commonly includes:

• purchase or production cost;
• ordering or setup cost;
• holding cost, including capital, storage, insurance and handling;
• shortage cost, including lost margin, expediting, penalties and reputational damage;
• obsolescence, spoilage and disposal cost;
• transport and administrative cost.

A useful distinction is between cycle stock, created by ordering in batches, and safety stock, held to protect against uncertainty. Pipeline inventory is stock currently in transit or otherwise committed but not yet available for use. Available inventory should not be confused with inventory position:

The inventory position is the relevant quantity for many replenishment decisions because an order already placed will affect future availability.

2. ABC–XYZ segmentation

ABC analysis

ABC analysis classifies items by annual consumption value (ACV):

where D_i is annual demand for item i in units per year and c_i is unit cost in currency per unit. Items are ranked from highest to lowest ACV and cumulative percentages are calculated. A typical, but not universal, classification is:

The cut-offs are policy choices, not natural laws. A low-value item may still be strategically critical if its absence stops production or compromises safety.

XYZ analysis

XYZ analysis classifies demand predictability. A common measure is the coefficient of variation:

where σ_i is the standard deviation of periodic demand and μ_i is its mean. Low-CV items are often classified X, medium-CV items Y and high-CV items Z. Thresholds must reflect the business context and the demand interval used.

XYZ is not identical to demand volume. An item can have high demand but be unpredictable, or low demand but highly stable. For intermittent demand, the ordinary coefficient of variation can be misleading because many zero observations inflate variability. Croston-type methods, demand-interval analysis and intermittent-demand measures may be more suitable.

Combined segmentation

Combining the classifications creates a portfolio. AX items have high value and predictable demand; AZ items have high value and uncertain demand; CX items have low value and predictable demand; CZ items have low value and uncertain demand. The combination supports differentiated control:

A mature segmentation may add criticality, lead-time risk, substitutability, shelf life, margin and carbon impact. Segmentation should be dynamic: classifications are recalculated when demand, price or supply conditions materially change.

3. The economic order quantity model

The basic EOQ model determines the order quantity that minimizes ordering and holding cost under restrictive assumptions.

Here Q* is the economic order quantity in units per order, D is annual demand in units per year, S is ordering or setup cost in currency per order, and H is annual holding cost in currency per unit per year.

If holding cost is expressed as a percentage i of unit price C, then H = iC. The annual relevant cost, excluding purchase cost when unit price is constant, is:

The first term is annual ordering cost and the second is annual cycle-stock holding cost. Differentiating with respect to Q gives:

Setting the derivative to zero yields Q* = √(2DS/H). The second derivative, 2DS/Q³, is positive for positive demand and order quantity, confirming a minimum.

The model assumes constant and known demand, constant lead time, instantaneous replenishment, no shortages, no quantity discounts, and independent orders. These assumptions are rarely exact. EOQ is therefore a benchmark rather than an automatic answer.

Worked example: EOQ

An item has annual demand of 24,000 units, ordering cost of £60 per order and annual holding cost of £2.40 per unit.

Orders per year are D/Q* ≈ 21.9, and the average cycle stock is approximately Q*/2 = 547.5 units, assuming no safety stock. The result should be rounded to a feasible pack size and checked against supplier minimums, transport economics and storage capacity.

Extensions

Quantity discounts require comparing total cost across price-break quantities because a lower purchase price can outweigh additional holding cost. If replenishment is gradual rather than instantaneous, the economic production quantity replaces EOQ and accounts for simultaneous demand during production. Joint ordering, supplier truckload constraints and shared setup costs may require a portfolio or mixed-integer optimization model rather than item-by-item EOQ.

4. Uncertainty, safety stock and service levels

Demand during lead time

A reorder decision must protect demand occurring between order release and replenishment availability. If demand per period has mean μ_D and standard deviation σ_D, and lead time has mean μ_L and standard deviation σ_L, then, under independence and a suitable normal approximation, the mean demand during lead time is:

The variance is:

Thus:

The first term represents demand variation during an average lead time; the second represents variation caused by lead-time uncertainty. If demand and lead time are correlated, the covariance term must be included:

The approximation is most defensible when demand is reasonably regular and the distribution is not severely skewed. Simulation or empirical quantiles are preferable for intermittent, promotional or highly skewed demand.

Safety stock

For a continuous-review policy with a target cycle service level and normally distributed lead-time demand:

where SS is safety stock in units, z is the standard-normal quantile corresponding to the target probability of no stockout during a replenishment cycle, and σ_LT is the standard deviation of lead-time demand. For example, a 95% cycle service level corresponds approximately to z = 1.645, while 99% corresponds approximately to z = 2.326.

The formula assumes that the forecast mean is unbiased, uncertainty is represented by σ_LT, replenishment is available when expected, and the selected distribution is appropriate. It does not automatically produce a 95% fill rate.

Reorder point

The reorder point is the inventory position at which replenishment is triggered. It is not necessarily the physical on-hand quantity because on-order stock and backorders affect inventory position.

Worked example: variable demand and lead time

Weekly demand has mean 500 units and standard deviation 80 units. Lead time has mean 2 weeks and standard deviation 0.5 week. Assume independence and a 95% cycle service level.

Therefore:

The large contribution from lead-time variation demonstrates why improving supplier reliability can reduce safety stock more effectively than refining the demand forecast alone.

Service-level measures

Cycle service level (CSL) is the probability of completing a replenishment cycle without a stockout. It is a probability measure.

Fill rate is the proportion of demand quantity supplied immediately from available stock:

A policy may have a high CSL but a lower fill rate if rare stockouts are large. Conversely, a low CSL can coexist with a high fill rate if shortages are small. Management must specify which service measure matters, over what time horizon and for which customer or product class.

For a one-period newsvendor decision with underage cost C_u and overage cost C_o, the critical fractile is:

This is appropriate for a single selling opportunity or strongly perishable product, not automatically for replenishable inventory.

5. Replenishment policies

Continuous review: (Q, R)

Under a continuous-review policy, inventory position is monitored continuously. When it reaches reorder point R, an order of fixed quantity Q is placed. The policy is written (Q,R). It responds quickly to demand but requires reliable transaction data and monitoring.

Periodic review: (P, S)

Under periodic review, inventory is checked every P time units and raised to an order-up-to level S. Protection must cover the review period plus lead time:

where σ_P+L is the standard deviation of demand over the protection period. If demand and lead time are independent and measured consistently:

The order quantity at review is:

Periodic review is convenient for route-based purchasing and supplier consolidation, but it generally requires more safety stock because a stockout can occur before the next review.

Choosing a policy

Policy selection should consider demand variability, value, criticality, order cost, system capability, supplier schedule, shelf life and review economics. A high-value, volatile item may need continuous review and exception alerts. Low-value stable items may be managed by periodic review, two-bin control or supplier-managed inventory. The policy should be evaluated using simulation when interactions, constraints and non-normal demand are important.

6. Multi-echelon inventory analytics

In a multi-echelon network, inventory exists at suppliers, central warehouses, regional distribution centres and customer-facing locations. Optimizing each node independently can produce excessive total stock because every echelon protects itself against the same uncertainty.

An echelon inventory position includes stock at a node and all downstream stock or commitments, depending on the modelling convention. Echelon-based policies coordinate buffers across the network. Upstream locations often hold aggregate safety stock because demand variability is pooled, while downstream locations hold stock where response time and customer service require it.

Important design questions include:

• Where should safety stock be positioned?
• Which demand signals should be forecast at each level?
• Can delayed differentiation or postponement reduce finished-goods variety?
• What capacity and transportation constraints limit pooling benefits?
• How are shortages allocated among locations and customers?

Risk pooling generally reduces relative variability when independent demand streams are aggregated. However, correlation, common promotions, regional disruption and lead-time dependence weaken the benefit. Multi-echelon models may use stochastic programming, simulation, optimization or approximate dynamic programming. Their value depends heavily on accurate lead-time, demand, capacity and allocation data.

7. Perishability and intermittent demand

Perishable inventory loses value with time. The effective holding cost includes expiry, markdown, disposal and quality loss. Policies must incorporate age, not only quantity. First-expiry-first-out (FEFO), lot tracking, shelf-life constraints and minimum remaining life at delivery are operational requirements rather than optional refinements.

For a short-life item, the decision may be a newsvendor problem: choose a quantity before uncertain demand is realized and accept a cost for both shortage and excess. For repeated replenishment, a rolling-horizon model is more appropriate because today’s order affects future age profiles.

Intermittent demand has frequent zero periods and occasional positive demand. Standard moving averages and normal safety-stock formulas can perform poorly. Analysts should distinguish:

• demand occurrence probability;
• size of non-zero demand;
• demand interval;
• obsolescence and substitution effects.

Croston-type methods estimate occurrence and size separately, while newer approaches use state-space models, count distributions and machine learning. Any method must be benchmarked against simple baselines and evaluated with a decision-relevant cost function, not forecast accuracy alone.

8. Analytics, technology and governance

Modern inventory systems combine enterprise-resource-planning transactions, warehouse scans, point-of-sale data, supplier confirmations, transport events and external signals. A control tower can monitor inventory position, projected stockout dates, supplier reliability and excess exposure.

AI and machine learning can support probabilistic forecasting, anomaly detection, lead-time prediction and dynamic safety-stock recommendations. They do not remove the need for causal reasoning. A model trained on historical stockouts may learn censored demand: observed sales are lower than true demand when the item was unavailable. Promotions, substitutions and deliberate allocations also distort the data.

Governance should include:

• clear ownership of master data and parameter changes;
• versioned forecasts and auditable overrides;
• monitoring for drift and bias;
• explainable exception rules for high-impact decisions;
• access controls and segregation of duties;
• validation of units, calendars, pack sizes and lead-time definitions;
• human review for critical, hazardous or regulated products.

Inventory algorithms can create inequitable allocation if scarce products are assigned solely by historical volume, disadvantaging smaller customers or vulnerable communities. Ethical allocation rules should be explicit, defensible and consistent with contractual and regulatory duties.

9. Applied case: illustrative medical-device network

Background

A fictional medical-device distributor supplies hospitals from one national distribution centre and four regional depots. A sterile implant is expensive, has uncertain hospital demand and a finite shelf life. Stockouts can delay surgery, while expiry creates substantial disposal cost.

Problem

The distributor used the same monthly order-up-to rule at every depot. Regional managers increased buffers after shortages, causing total inventory to rise by 28% while expiry also increased. The organization measured only on-hand stock and monthly fill rate, not inventory position, age profile or cycle service level.

Analysis

An ABC–XYZ review classified the implant as AZ: high annual consumption value and highly variable demand. A criticality override confirmed that ordinary value-based cost minimization was inappropriate. Analysis separated demand uncertainty from supplier lead-time uncertainty and found that lead-time variation accounted for most safety-stock variance. The central depot had correlated demand during national tenders, so simple pooling assumptions were tested rather than accepted.

The team simulated three alternatives:

1. independent regional buffers;
2. a larger central buffer with smaller regional presentation stocks;
3. postponement, in which a common component was held centrally and final configuration occurred after hospital demand was known.

The simulation incorporated age, expiry, emergency transshipments, minimum order quantities and hospital service targets.

Approach and solution

The distributor adopted a differentiated policy. Regional depots used continuous review for the highest-criticality configurations, with safety stock based on empirical lead-time-demand quantiles rather than a normal approximation. The central depot held pooled common components and used FEFO. Supplier contracts added lead-time performance clauses and advance shipment notices. Allocation rules prioritized scheduled surgery, then clinical urgency, then transparent contractual priorities.

Dashboard measures included CSL, fill rate, emergency shipments, expiry percentage, inventory turns, working capital and forecast bias. Parameter changes required documented approval.

Outcome

In this illustrative case, six months after implementation, total inventory value fell by 14%, expiry fell by 31% and the target hospital service level was maintained. Emergency shipments declined because supplier lead-time variance was reduced. The outcome did not result from a single EOQ calculation; it resulted from segmentation, network pooling, age-aware control, better data and explicit service governance.

Learning points

• Criticality can override ABC classification.
• Safety stock should address both demand and lead-time uncertainty.
• Network-level optimization can outperform local optimization.
• Perishability requires age-aware policies.
• Service metrics must be specified precisely and audited.

10. Limitations and best practices

No inventory formula is universally valid. Forecast error may be non-stationary; demand may be censored by stockouts; supplier lead times may be dependent on order size; and customer service may be nonlinear. Normal approximations can understate tail risk. EOQ can recommend infeasible quantities. A mathematically precise model with poor data can be less useful than a simple, well-governed rule.

Best practice is to:

1. define the service objective and cost of failure;
2. cleanse demand, lead-time and inventory-position data;
3. segment by value, variability and criticality;
4. select a policy appropriate to the demand process and perishability;
5. test assumptions using historical back-testing and simulation;
6. include capacity, pack-size, supplier and transport constraints;
7. monitor outcomes, not merely parameter compliance;
8. review parameters after structural changes, promotions or disruptions;
9. document overrides and maintain human accountability.

Chapter Summary

Inventory and replenishment analytics balances availability, capital and risk. ABC analysis identifies value concentration, while XYZ analysis identifies demand predictability; their combination supports differentiated control, but criticality and shelf life may require overrides. EOQ provides a benchmark under stable-demand assumptions by balancing ordering and holding cost. Safety stock protects against uncertainty, and its calculation must distinguish demand variation, lead-time variation and their possible dependence. Reorder points are based on demand during lead time plus safety stock, using inventory position rather than on-hand stock alone.

Cycle service level measures the probability of avoiding a stockout during a cycle, whereas fill rate measures the quantity supplied immediately. Continuous-review and periodic-review policies have different data, responsiveness and safety-stock implications. Multi-echelon decisions should position inventory across the network rather than optimize each location independently. Perishable and intermittent-demand items require age-aware and distribution-appropriate methods. AI can improve forecasting and exception management, but governance, data quality, explainability and ethical allocation remain essential. Effective practice combines quantitative models with simulation, operational constraints, managerial judgment and continuous performance measurement.

Key Terms

ABC analysis: Classification of items by annual consumption value.

XYZ analysis: Classification of items by demand variability or predictability.

Annual consumption value: Annual demand multiplied by unit cost.

Cycle stock: Inventory created by ordering in batches.

Safety stock: Buffer inventory held against uncertainty.

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

EOQ: Order quantity minimizing basic ordering and holding cost.

Ordering cost: Cost incurred each time an order is placed or setup performed.

Holding cost: Cost of carrying one unit for a stated time period.

Lead-time demand: Demand occurring between replenishment release and availability.

Reorder point: Inventory-position threshold that triggers replenishment.

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

Fill rate: Proportion of demand quantity supplied immediately.

Continuous review: Monitoring inventory position continuously.

Periodic review: Checking inventory at predetermined intervals.

Multi-echelon inventory: Inventory managed across multiple supply-chain levels.

FEFO: First-expiry-first-out issuing or allocation rule.

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

Critical fractile: Target cumulative probability based on underage and overage costs.

Inventory pooling: Combining demand or stock across locations to reduce aggregate uncertainty.

Review Questions

Short Answer Questions

1. What is the difference between cycle stock and safety stock?
2. Why can a CZ item require tighter control than an AX item?
3. Define inventory position.
4. Distinguish cycle service level from fill rate.
5. Why is FEFO important for perishable inventory?

Long Answer Questions

1. Explain ABC–XYZ segmentation and evaluate its strengths and limitations for a complex product portfolio.
2. Derive the basic EOQ expression and discuss the assumptions under which it is valid.
3. Explain how demand and lead-time variability contribute to safety stock.
4. Compare continuous-review and periodic-review replenishment policies.
5. Discuss the strategic benefits and risks of multi-echelon inventory optimization.

Analytical / Application Questions

1. Weekly demand has mean 600 units and standard deviation 90 units. Lead time has mean 3 weeks and standard deviation 0.4 week. Assuming independence and a 98% cycle service level, calculate lead-time demand mean, lead-time demand standard deviation, safety stock and reorder point.
2. A distributor has high-value, highly variable demand for a short-life product. Design an inventory policy covering segmentation, service measurement, replenishment, expiry control and governance. Justify each element.
3. Two regional warehouses independently hold safety stock for the same item, although demand is partly uncorrelated and the supplier lead time is variable. Explain how you would test whether central pooling would reduce total inventory without reducing service.

Multiple Choice Questions with Answers

Question 1

Which measure is used most directly in ABC analysis?

A. Annual consumption value

B. Coefficient of variation only

C. Supplier defect rate

D. Shelf-life percentage

Correct Answer: A. Annual consumption value

Question 2

Annual demand is 12,000 units, ordering cost is £40, and annual holding cost is £2 per unit. What is the basic EOQ?

A. 245 units

B. 490 units

C. 600 units

D. 1,200 units

Correct Answer: B. 490 units

Question 3

Which quantity is normally used to trigger a continuous-review replenishment order?

A. On-hand inventory only

B. Inventory position

C. Annual purchase value

D. Average unit cost

Correct Answer: B. Inventory position

Question 4

Under the normal approximation, what does safety stock equal for a target cycle service level?

A. μLT ÷ z

B. z × σLT

C. D × S

D. H × Q

Correct Answer: B. z × σLT

Question 5

A 95% cycle service level corresponds approximately to which standard-normal z value?

A. 0.95

B. 1.28

C. 1.645

D. 2.576

Correct Answer: C. 1.645

Question 6

Which statement correctly distinguishes fill rate from cycle service level?

A. Fill rate measures the probability of no stockout in a cycle

B. Cycle service level measures the percentage of units supplied

C. Fill rate measures the proportion of demand quantity supplied immediately

D. They are always numerically identical

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

Question 7

In a periodic-review system, the protection period generally includes:

A. Only the supplier lead time

B. Only the review interval

C. The review interval plus lead time

D. The supplier payment period

Correct Answer: C. The review interval plus lead time

Question 8

Which feature most directly distinguishes perishable-inventory control from ordinary cycle-stock control?

A. Ignoring item age

B. Tracking expiry and remaining shelf life

C. Using purchase cost alone

D. Eliminating service targets

Correct Answer: B. Tracking expiry and remaining shelf life

Question 9

Why can multi-echelon pooling reduce safety stock?

A. It guarantees zero demand variability

B. It can aggregate partially independent demand risks

C. It removes all supplier lead time

D. It makes every warehouse unnecessary

Correct Answer: B. It can aggregate partially independent demand risks

Question 10

A high-value item with highly variable demand is commonly classified as:

A. AX

B. AY

C. AZ

D. CZ

Correct Answer: C. AZ