Demand Forecasting and Inventory Planning

Introduction

Inventory decisions begin with an estimate of future demand. A forecast is not a promise about what will happen; it is a structured estimate based on information available at the time of planning. Inventory managers use forecasts to determine purchasing quantities, production schedules, workforce requirements, warehouse capacity, transport needs, safety stock, and customer-service commitments.

Forecasting is particularly important because inventory creates both value and cost. Too little inventory causes stockouts, lost sales, production interruptions, and dissatisfied customers. Too much inventory ties up working capital, increases storage and insurance costs, creates obsolescence risk, and may conceal quality or process problems. Effective inventory planning therefore combines demand estimation with lead-time analysis, service-level objectives, replenishment policies, and continuing review.

Forecasting cannot eliminate uncertainty. Its purpose is to make uncertainty visible, measurable, and manageable.

Learning Objectives

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

1. Explain the role, scope, and limitations of demand forecasting.
2. Distinguish qualitative and quantitative forecasting methods.
3. Calculate and interpret moving averages, weighted moving averages, and exponential smoothing forecasts.
4. Recognize trend, seasonality, cyclical movement, random variation, intermittent demand, and dependent demand.
5. Measure forecast error using appropriate statistical indicators.
6. Translate forecasts into reorder points, safety stock, and time-phased inventory plans.
7. Evaluate forecasting systems ethically and operationally, including the use of software and artificial intelligence.

1. The Nature of Demand and Forecasting

Demand is the quantity of a product or service that customers are willing and able to purchase during a specified period at a specified price and under specified conditions. Demand is not always equal to sales. Sales may be lower than true demand when stockouts, delivery delays, credit restrictions, or poor availability prevent customers from purchasing.

A forecast should specify:

• the item or product family;
• the forecast horizon and time bucket, such as daily or monthly;
• the unit of measure;
• the location or market;
• the information cut-off date; and
• the purpose of the forecast.

Forecasts for strategic capacity planning may cover several years, while replenishment forecasts may cover only the supplier lead time plus a planning buffer. Aggregated forecasts, such as total demand for a product family, are usually more accurate than forecasts for individual stock-keeping units (SKUs), because random variations partly offset one another.

Demand patterns

Four components are commonly used to describe a time series:

• Level: the underlying average magnitude of demand.
• Trend: a persistent upward or downward movement.
• Seasonality: a repeating pattern associated with a calendar, climatic, institutional, or operational cycle.
• Random variation: irregular movement that cannot be predicted reliably from the available data.

Cyclical movement may also occur, particularly in industrial and capital-goods markets. Unlike seasonality, cycles do not necessarily repeat at a fixed frequency or amplitude.

Independent and dependent demand

Independent demand arises from external customers or markets. Finished goods, replacement parts, and retail products commonly have independent demand. Dependent demand is derived from the requirements of another item or process. For example, the demand for tyres, seats, and engines depends on the planned production of vehicles.

Independent demand is often forecast statistically. Dependent demand is usually calculated through a master production schedule and a bill of materials using materials requirements planning (MRP). Forecasting a component independently when its demand is dependent can produce inaccurate and contradictory plans.

2. The Forecasting Process

A professional forecasting process normally follows these steps:

1. Define the decision and forecast purpose.
2. Identify the item, location, unit, horizon, and time bucket.
3. Collect and clean historical demand and causal information.
4. Examine the demand pattern using plots, summaries, and operational knowledge.
5. Select one or more suitable methods.
6. Generate the forecast and a prediction interval or uncertainty estimate where possible.
7. Compare the forecast with actual demand and calculate errors.
8. Review exceptional events and obtain informed managerial judgment.
9. Translate the forecast into purchasing, production, and inventory decisions.
10. Monitor performance and revise the method when conditions change.

Data cleansing requires care. A very large observation may represent genuine demand, a promotion, a one-time project, an order backlog, or a recording error. Automatically deleting it may be as harmful as accepting it without investigation. The reason for an adjustment should be documented, approved, and retained for auditability.

3. Qualitative Forecasting Methods

Qualitative methods are valuable when historical data are limited, the market is changing, or the forecast concerns a new product. They rely on structured judgment rather than solely on numerical history.

Executive or expert opinion

Senior managers or technical specialists provide estimates based on market knowledge. This method is quick, but group hierarchy and optimism bias can dominate the result. Anonymous estimates followed by structured discussion can reduce conformity pressure.

Sales-force composite

Sales representatives estimate demand in their territories and the estimates are aggregated. Sales staff possess customer knowledge, but may understate demand to create easier targets or overstate it to secure stock.

Market research and customer surveys

Interviews, surveys, test marketing, and conjoint studies can reveal customer preferences. Results depend on sample quality, question design, response bias, and the difference between stated intention and actual purchasing behaviour.

Delphi method

A panel of experts submits forecasts independently through several confidential rounds. A coordinator summarizes the responses and circulates the summary until a reasoned convergence or stable disagreement is reached. Delphi is useful for long-range and technology forecasts, but it is time-consuming and does not guarantee correctness.

Scenario analysis

Managers construct alternative futures, such as high-growth, base-case, and disruption scenarios. Scenario analysis is not a single-point forecast; it is a way to test the robustness of inventory and capacity decisions under different assumptions.

Qualitative judgment should complement, not automatically replace, quantitative evidence. A useful practice is to record the statistical baseline separately from the judgmental adjustment and monitor both against actual results.

4. Quantitative Forecasting Methods

Quantitative methods assume that useful information exists in historical observations or explanatory variables. They are especially useful for repetitive demand with reasonably stable relationships.

4.1 Naïve forecast

For a stable series, the next-period forecast may equal the most recent actual demand:

Here, F_t+1 is the forecast for period t+1, and A_t is actual demand in period t. The method is simple and provides an important benchmark. A complex method should not be adopted unless it performs better than a naïve benchmark after fair testing.

4.2 Simple moving average

A moving average uses the most recent n observations:

The units of the forecast are the same as the units of demand. A larger n smooths random variation but responds more slowly to changes. A smaller n reacts more quickly but is more sensitive to noise.

Worked example. Monthly demand for the last three months was 420, 500, and 460 units. The three-month moving-average forecast for the next month is:

F_t+1 = (420 + 500 + 460) ÷ 3 = 1,380 ÷ 3 = 460 units.

The method assumes that recent observations are equally relevant and that no strong trend or seasonal pattern has been ignored.

4.3 Weighted moving average

A weighted moving average assigns different weights to the observations:

The weights must satisfy:

The weights are dimensionless, so the resulting forecast retains the units of demand. Greater weight on recent data makes the forecast more responsive.

If the last three observations are 420, 500, and 460 units and the weights for the most recent, second-most recent, and oldest observations are 0.50, 0.30, and 0.20 respectively:

F_t+1 = (0.50 × 460) + (0.30 × 500) + (0.20 × 420) = 230 + 150 + 84 = 464 units.

Weights should be selected using validation, operational knowledge, or both. They should not be changed after observing the test results merely to obtain a preferred answer.

4.4 Exponential smoothing

Simple exponential smoothing updates the previous forecast using the most recent error:

Equivalently:

α is the smoothing constant, with 0 < α ≤ 1. A_t is actual demand and F_t is the forecast for period t. A high α responds quickly to changes; a low value produces greater smoothing. Simple exponential smoothing is appropriate for level demand without a systematic trend or seasonality. Trend and seasonal versions are required when those components are material.

Worked example. Suppose F_t = 460 units, A_t = 500 units, and α = 0.30. Then:

F_t+1 = 0.30(500) + 0.70(460) = 150 + 322 = 472 units.

The forecast increases by 12 units, which is 30% of the observed error of 40 units.

The initial forecast may be set equal to the first observation, an average of early observations, or a managerial estimate. The initialization rule should be documented because it affects early forecasts.

4.5 Trend and seasonality

A trend-adjusted method is appropriate when demand changes persistently over time. Holt’s method uses a level and a trend. One common form is:

L_t is the estimated level, T_t is the estimated trend, α and β are smoothing constants between 0 and 1, and m is the number of periods ahead. The method assumes that the trend is reasonably stable over the forecast horizon.

Seasonality can be represented using seasonal indices. Under a multiplicative model:

Seasonal indices are commonly normalized so that their average over a complete season equals 1.00. An index of 1.20 indicates demand approximately 20% above the deseasonalized level; an index of 0.80 indicates demand approximately 20% below it. Additive seasonal models use seasonal effects in the same units as demand and are preferable when seasonal variation is approximately constant in absolute terms. Multiplicative models are preferable when seasonal variation grows with the level of demand.

Seasonal estimates must be based on sufficient historical cycles and should be reviewed when pricing, product mix, climate, regulation, or customer behaviour changes.

5. Intermittent Demand

Intermittent demand contains many zero-demand periods and occasional non-zero transactions. It occurs for slow-moving spare parts, maintenance items, emergency equipment, and specialized components. Ordinary moving averages may respond poorly because zeros reduce the average while a large, irregular order can create a sudden increase.

Two characteristics should be distinguished:

• demand occurrence: whether demand occurs in a period; and
• demand size: the quantity demanded when it occurs.

Croston-type approaches estimate these components separately. A basic Croston method updates the average non-zero demand size and the average interval between non-zero demands, then divides the former by the latter. Variants such as SBA and TSB address known biases or changing demand occurrence. These methods require careful implementation and should be tested against suitable benchmarks.

For intermittent items, inventory policy should also consider criticality, supplier minimum order quantities, repairability, substitution, and the cost of a stockout. A purely statistical forecast is insufficient when a single part can stop an entire production line.

6. Forecast Error and Performance Measurement

Forecast error for period t is commonly defined as:

A positive error means actual demand exceeded the forecast; a negative error means the forecast exceeded actual demand. Organizations must state their sign convention because alternative definitions exist.

Important measures include:

MAPE is undefined when A_t = 0 and can be distorted when actual demand is very small. Weighted absolute percentage error or scaled measures may be better for intermittent demand. Forecast performance should be evaluated over a representative holdout period rather than only the data used to fit the method.

Bias is operationally important. Persistent under-forecasting increases stockouts and emergency purchases; persistent over-forecasting increases inventory and obsolescence. A forecast can have low average absolute error but still be biased if positive and negative errors do not balance.

Forecast tracking should examine error by product family, location, customer segment, and demand condition. A single aggregate accuracy percentage can conceal serious failures in high-value or critical items.

7. Translating Forecasts into Inventory Plans

A forecast becomes useful only when it is converted into a time-phased inventory plan. The plan should connect expected demand with on-hand stock, scheduled receipts, open orders, lead times, safety stock, and service objectives.

Reorder point

For reasonably stable demand and constant lead time:

ROP is the reorder point in units, d is average demand per time unit, L is lead time in the same time unit, and SS is safety stock in units. Unit consistency is essential: daily demand must be paired with lead time in days.

If demand and lead time are variable and approximately independent, a common approximation for the standard deviation of demand during lead time is:

where σ_d is the standard deviation of demand per period and σ_L is the standard deviation of lead time, both expressed in compatible units. A common safety-stock rule is:

z is the standard-normal value associated with the desired cycle-service probability. This approximation assumes that demand and lead time are sufficiently stable, distributions are reasonably represented by the model, and demand and lead time are independent. If they are correlated, an additional covariance term is required; the simple formula should not be used uncritically.

Worked example. Daily demand has mean 100 units and standard deviation 20 units. Lead time has mean 5 days and standard deviation 1 day. Assume independence and a service factor z = 1.65. Then:

σ_DLT = √(5 × 20² + 100² × 1²) = √(2,000 + 10,000) = √12,000 ≈ 109.54 units.

Thus:

SS = 1.65 × 109.54 ≈ 180.74 units, rounded operationally to 181 units.

The reorder point is:

ROP = (100 × 5) + 181 = 681 units.

This result is a planning approximation, not a guarantee of a particular fill rate. Service metrics must be defined precisely; cycle service level and fill rate are not interchangeable.

Periodic review and order-up-to level

Under periodic review, inventory is checked every R time units and replenishment must cover the review period plus lead time. A simple order-up-to level is:

The order quantity is generally the order-up-to level minus the inventory position:

Inventory position is on-hand inventory plus on-order inventory minus backorders. In practice, order multiples, minimum order quantities, shelf life, capacity, and supplier constraints may require rounding or adjustment.

Time-phased planning

A time-phased plan lays out forecast demand by period and subtracts available inventory and scheduled receipts. MRP performs this logic for dependent demand, using the master production schedule, bill of materials, inventory records, lead times, lot-sizing rules, and planned-order releases. Forecasts may influence the master schedule for finished products, but component requirements should normally be exploded from planned production rather than independently forecast.

8. Applied Case: Illustrative Spare-Parts Planning

Background

A regional water utility stocks a pump seal used in maintenance. Monthly issues over six months were 0, 0, 12, 0, 0, and 20 units. The part is inexpensive but operationally critical because pump failure can interrupt water service. Supplier lead time is variable, and emergency procurement is costly.

Problem

A buyer proposes a three-month moving average. This gives:

(0 + 0 + 20) ÷ 3 = 6.67 units for the next month.

The estimate appears precise but hides the intermittent pattern. It may also produce an inadequate replenishment quantity if a maintenance event occurs.

Analysis

The planner separates occurrence from size. There were two non-zero demands, of 12 and 20 units, over six months. The average non-zero demand size is:

(12 + 20) ÷ 2 = 16 units.

The average interval between non-zero demands is approximately three months in this small sample. A Croston-style baseline is therefore approximately:

16 ÷ 3 ≈ 5.33 units per month.

This baseline is not treated as a complete answer. The planner reviews the maintenance schedule, identifies two pumps due for overhaul, checks whether seals can be refurbished, and confirms the supplier’s minimum order quantity. Because the item is critical, the utility chooses a higher service objective than it would for a non-critical consumable.

Approach and solution

The final plan combines:

• a documented intermittent-demand baseline;
• known maintenance work orders as dependent or scheduled requirements;
• a criticality-based safety-stock decision;
• supplier lead-time and minimum-order constraints; and
• monthly review of demand occurrence, demand size, and stockout exposure.

The utility avoids blindly applying a high statistical safety stock to every slow-moving part. Instead, it uses operational consequences to classify the item and records the judgmental adjustment separately from the statistical baseline.

Outcome and learning points

The case illustrates that a mathematically simple average may be unsuitable for intermittent demand. Forecasting and inventory planning must be integrated with maintenance schedules, criticality, lead-time risk, and the economics of failure. The appropriate method is not necessarily the most sophisticated method; it is the method that produces a transparent and controllable decision.

9. Technology, Governance, and Contemporary Practice

Enterprise resource planning, warehouse-management, and advanced planning systems can automate data collection, forecasting, exception alerts, and replenishment proposals. Machine-learning methods may improve forecasts when large, clean datasets contain nonlinear relationships involving price, promotions, weather, web traffic, or economic indicators. However, algorithmic complexity does not compensate for poor data, structural market change, or an incorrect business objective.

Good governance requires:

• documented data definitions and ownership;
• version control for forecasts and overrides;
• separation of forecast generation from performance evaluation;
• monitoring for bias across customers, locations, or product categories;
• human review of exceptional and critical items;
• protection of commercially sensitive and personal data; and
• clear accountability when an automated recommendation is accepted.

Forecast overrides should be evidence-based. Repeated overrides may indicate that the model is missing a causal variable, that the forecast horizon is unsuitable, or that planners are responding to incentives unrelated to accuracy. Promotions, shortages, supply disruptions, and one-time projects should be represented explicitly where possible rather than concealed in unexplained adjustments.

Sustainability is also relevant. Excess inventory consumes materials, energy, space, and working capital, while stockouts may generate expedited transport and emergency production. Better segmentation, postponement, repair, reuse, and end-of-life planning can reduce both environmental and financial waste.

10. Limitations and Best Practices

No forecasting method is universally superior. Historical methods fail when the market changes structurally. Judgment fails when incentives and cognitive biases dominate. Forecast accuracy can be high while inventory performance remains poor if lead times, lot sizes, service policies, or master data are wrong.

Best practices include:

1. Use a clear forecast hierarchy from aggregate demand to SKU-location detail.
2. Benchmark complex methods against naïve alternatives.
3. Match the method to the demand pattern, including intermittency and seasonality.
4. Evaluate forecasts out of sample and over a meaningful horizon.
5. Report both accuracy and bias.
6. Use units, calendars, and lead times consistently.
7. Separate independent demand forecasting from dependent-demand planning.
8. Link safety stock to variability, lead time, criticality, and service objectives.
9. Document overrides, assumptions, and data adjustments.
10. Review forecasts continuously rather than assuming that historical parameters remain valid.

Chapter Summary

Demand forecasting estimates future requirements so that organizations can coordinate purchasing, production, capacity, and inventory. Demand may contain level, trend, seasonality, cyclical movement, and random variation. Independent demand is influenced by external customers, whereas dependent demand is derived from planned production and bills of materials.

Qualitative methods, including expert opinion, sales-force estimates, market research, Delphi studies, and scenarios, are valuable when data are scarce or structural change is significant. Quantitative methods include naïve forecasts, moving averages, weighted moving averages, exponential smoothing, trend-adjusted methods, and seasonal models. Intermittent demand requires special treatment because the timing and size of demand are separately uncertain.

Forecast error should be measured using appropriate metrics such as ME, MAD, MSE, RMSE, and carefully applied percentage measures. Accuracy must be considered alongside bias and operational consequences. Forecasts become inventory plans through reorder points, safety stock, periodic-review order-up-to levels, and time-phased planning. These calculations depend on consistent units and explicit assumptions about demand, lead time, variability, and service objectives.

Effective practice combines statistical evidence, operational judgment, technology, governance, and continuous learning. The central managerial task is not to produce a perfect forecast, but to make robust inventory decisions under measurable uncertainty.

Key Terms

Demand forecast: A structured estimate of future demand for a specified item, period, and location.

Independent demand: Demand originating from external customers or markets.

Dependent demand: Demand calculated from the requirements of another item or production plan.

Level: The underlying average magnitude of a demand series.

Trend: Persistent upward or downward movement in demand.

Seasonality: A repeating pattern associated with a regular calendar or operating cycle.

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

Moving average: An average of the most recent observations used as a forecast.

Weighted moving average: A moving average in which observations receive unequal weights.

Exponential smoothing: A recursive forecasting method that updates a forecast using the latest error.

Smoothing constant: The parameter controlling responsiveness in exponential smoothing.

Forecast error: The difference between actual demand and forecast demand.

Bias: Persistent tendency to over-forecast or under-forecast.

MAD: Mean absolute deviation, the average absolute forecast error in demand units.

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

Reorder point: The inventory position at which replenishment should be initiated.

Lead time: The elapsed time between placing and receiving a replenishment order.

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

Service level: A defined probability or performance target for meeting demand from available stock.

Materials requirements planning: A system that calculates dependent material requirements from production plans, bills of materials, inventory records, and lead times.

Review Questions

Short Answer Questions

1. Why can sales be lower than actual customer demand?
2. Distinguish independent demand from dependent demand.
3. What effect does increasing the number of periods in a moving average have on responsiveness?
4. Why is MAPE unsuitable when actual demand is zero?
5. Define safety stock and state its purpose.

Long Answer Questions

1. Explain the forecasting process from defining the decision through monitoring forecast performance.
2. Compare qualitative forecasting methods and identify situations in which each may be appropriate.
3. Explain simple exponential smoothing, including the role of the smoothing constant and its assumptions.
4. Discuss the difference between trend and seasonality and explain how each can be incorporated into forecasting.
5. Explain how a forecast is translated into a reorder point and a periodic-review order-up-to level.

Analytical / Application Questions

1. Demand for the last four weeks was 180, 220, 200, and 240 units. Calculate the four-week moving-average forecast for the next week and explain one limitation of the result.
2. A product has average daily demand of 80 units, a five-day lead time, and safety stock of 120 units. Calculate the reorder point and state the unit-consistency assumption.
3. A spare part has demand of 0, 0, 15, 0, 25, and 0 units over six months. Explain why a conventional moving average may be inadequate and propose an inventory-planning approach.

Multiple Choice Questions with Answers

Question 1

Which demand type is normally calculated from a bill of materials and a production schedule?

A. Independent demand

B. Dependent demand

C. Seasonal demand

D. Random demand

Correct Answer: B. Dependent demand

Question 2

A three-period moving average uses demands of 120, 150, and 180 units. What is the next-period forecast?

A. 135 units

B. 145 units

C. 150 units

D. 180 units

Correct Answer: C. 150 units

Question 3

In simple exponential smoothing, a higher value of α generally makes the forecast:

A. More responsive to recent observations

B. Less responsive to recent observations

C. Independent of actual demand

D. Equal to the longest historical average

Correct Answer: A. More responsive to recent observations

Question 4

If actual demand is 500 units and the forecast is 460 units, using e_t = A_t − F_t, what is the forecast error?

A. −40 units

B. 0 units

C. 40 units

D. 960 units

Correct Answer: C. 40 units

Question 5

Which measure reports average absolute forecast error in the original demand units?

A. MAPE

B. MAD

C. MSE

D. Seasonal index

Correct Answer: B. MAD

Question 6

Average daily demand is 100 units, lead time is 4 days, and safety stock is 150 units. What is the reorder point?

A. 250 units

B. 400 units

C. 450 units

D. 550 units

Correct Answer: C. 450 units

Question 7

Which characteristic most strongly indicates intermittent demand?

A. A smooth upward trend

B. Regular monthly peaks

C. Many zero-demand periods with occasional orders

D. Constant demand every period

Correct Answer: C. Many zero-demand periods with occasional orders

Question 8

Under the common reorder-point model, safety stock primarily protects against:

A. Demand and supply uncertainty

B. Product packaging design

C. Accounting depreciation

D. Supplier marketing activity

Correct Answer: A. Demand and supply uncertainty

Question 9

Why must demand and lead-time units be consistent in ROP = dL + SS?

A. To ensure the reorder point is expressed in demand units

B. To force safety stock to equal zero

C. To make all forecasts seasonal

D. To eliminate forecast error

Correct Answer: A. To ensure the reorder point is expressed in demand units

Question 10

Which method is especially useful when historical data are limited and a new product is being introduced?

A. Qualitative expert judgment

B. Three-year moving average only

C. Dependent-demand explosion only

D. Random replenishment

Correct Answer: A. Qualitative expert judgment