Demand Management and Forecasting

Introduction

Production planning and control begins with a practical question: how much will customers require, and when will they require it? Demand forecasts provide an informed estimate of future requirements, while demand management coordinates the actions used to influence, shape, prioritize, and respond to those requirements. Together, they connect market information with decisions about capacity, materials, labour, inventory, scheduling, purchasing, and distribution.

A forecast is not a prediction of an inevitable future. It is an estimate based on available information, assumptions, and a chosen method. Forecasts are therefore uncertain and should be expressed, reviewed, and used as part of a planning process rather than treated as exact commitments. Effective production planning combines statistical evidence with managerial judgement, market knowledge, and collaboration across the supply chain.

Learning Objectives

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

1. Explain the purpose and scope of demand management in production planning and control.
2. Distinguish qualitative and quantitative forecasting methods.
3. Calculate and interpret moving averages, weighted moving averages, and exponential smoothing forecasts.
4. Identify trend and seasonality and explain their implications for production decisions.
5. Measure forecast error, accuracy, and bias using appropriate indicators.
6. Explain how collaborative forecasting reduces demand uncertainty.
7. Evaluate forecasting limitations and select suitable methods for operational decisions.

1. Demand Management in Operations

Demand may be classified in several ways. Independent demand originates outside the production system, such as customer purchases of finished goods. Dependent demand is derived from another item’s requirements, such as components needed to manufacture a product. Independent demand is commonly forecast, whereas dependent demand is usually calculated from the master production schedule and bill of materials.

Demand management includes four related activities:

• collecting and organizing demand information;
• forecasting future demand;
• influencing or negotiating demand where possible; and
• aligning demand with available capacity and supply.

Demand may be order-driven, where production begins after receipt of a customer order, or forecast-driven, where products are made in anticipation of demand. Most organizations use a combination. Standard products may be produced for stock, while customized products are made to order.

Demand management also distinguishes between demand volume and demand timing. Two products may have the same annual demand but very different weekly patterns. Timing affects overtime, inventory, transport, supplier capacity, and customer service. A useful demand-management process therefore examines the level, trend, seasonality, variability, and business causes of demand.

Demand shaping

Organizations cannot always control demand, but they may influence it through:

• pricing and discounts;
• promotions and advertising;
• product substitution;
• delivery-date negotiation;
• product allocation during shortages;
• reservations and contracts; and
• incentives for off-peak ordering.

Demand shaping must be coordinated with operations. A promotion that increases sales without checking materials and capacity may produce stockouts, expediting costs, and damaged customer relationships.

2. Demand Patterns and Forecasting Horizons

A time series of demand commonly contains four components:

1. Level: the underlying average magnitude of demand.
2. Trend: a persistent upward or downward movement.
3. Seasonality: a repeating pattern at a known frequency, such as monthly or weekly.
4. Random variation: irregular movement that cannot be reliably predicted.

Some analyses also consider cycles, which are longer-term fluctuations associated with economic or industry conditions. Unlike seasonality, cycles may not have a fixed length or stable amplitude.

Forecasting horizons should match the decision being supported:

Short-term forecasts often rely heavily on recent observations. Long-term forecasts require broader assumptions and are generally more uncertain. Forecast accuracy usually declines as the horizon increases.

3. Qualitative Forecasting Methods

Qualitative methods are especially valuable when historical data are limited, the market is changing, or a new product is being introduced. They rely on structured judgement rather than solely on numerical history.

Executive opinion

Senior managers combine knowledge of strategy, markets, finance, and operations to create a forecast. This is fast and inexpensive but may be distorted by hierarchy, optimism, or political pressure. A documented assumption record and independent estimates can reduce these risks.

Sales-force composite

Sales personnel estimate demand in their territories or customer groups. This method benefits from direct customer contact but may be affected by incentives, negotiation tactics, or excessive pessimism.

Market research and customer surveys

Surveys, interviews, focus groups, and customer panels provide information about intentions and preferences. Stated intentions do not always become actual purchases, so survey results should be combined with behavioural data where possible.

Delphi method

The Delphi method obtains forecasts from a panel of experts through several anonymous rounds. A coordinator summarizes responses between rounds, allowing experts to reconsider their estimates without direct pressure from dominant participants. It is useful for long-term or uncertain issues but can be time-consuming and depends on panel quality.

Analogy and lifecycle reasoning

Demand for a new product may be estimated by comparing it with a similar product, market, or technology. The analyst must adjust for differences in price, distribution, customer group, regulation, and maturity.

Qualitative methods should not be regarded as unscientific. Their quality depends on transparent assumptions, appropriate expertise, independence, and systematic comparison with later outcomes.

4. Quantitative Forecasting Methods

Quantitative forecasting uses numerical data and an explicit procedure. Two broad categories are:

• Time-series methods, which project patterns in past demand into the future.
• Causal methods, which relate demand to explanatory variables such as price, income, advertising, weather, or economic activity.

Time-series methods are practical for operational planning when demand history is available and conditions are reasonably stable. Causal models may be more useful when management needs to understand drivers or evaluate scenarios.

4.1 Naïve forecasting

The simplest forecast for the next period equals the most recent actual demand:

where F_t+1 is the forecast for period t+1, and A_t is actual demand in period t. The method requires no estimated parameters and provides a useful benchmark. It performs poorly when demand has strong trend or seasonality unless the comparison is made with the corresponding prior season.

4.2 Simple moving average

A moving average uses the most recent n observations:

Here, n is the number of periods in the window. All observations receive equal weight. A larger n produces a smoother forecast but responds more slowly to genuine changes. A smaller window responds more rapidly but may retain more random variation.

Assumptions and applicability: demand observations are measured consistently; the recent past is relevant to the next period; and no strong unadjusted trend or seasonal pattern invalidates the averaging process.

Worked Example 1: Moving average

Weekly demand for a component was 420, 450, 390, and 440 units in the last four weeks. The three-week moving-average forecast for week 5 is:

The result may be rounded to 427 units if whole units must be scheduled. Rounding should be performed only after the calculation, and operational constraints may require a further decision about batch size or safety stock.

4.3 Weighted moving average

A weighted moving average assigns different weights to recent observations:

The weights should normally satisfy:

The most recent observation commonly receives the greatest weight, although the choice must be justified by demand behaviour. If weights do not sum to 1, the result is not on the same unit scale unless a separate normalization is applied.

Using the previous example, suppose weights of 0.5, 0.3, and 0.2 are assigned to the latest three weeks:

Weighted averages can respond more quickly than simple averages, but poorly chosen weights may amplify noise or create systematic lag.

4.4 Exponential smoothing

Simple exponential smoothing updates the previous forecast using the latest forecast error:

Equivalently:

where α is the smoothing constant, with 0 < α ≤ 1. A_t is actual demand and F_t is the forecast for period t. The units of F_t+1 and A_t are the same, such as units per week.

A high α responds strongly to recent changes; a low α creates a smoother, slower-moving forecast. Simple exponential smoothing is most suitable when demand has a relatively stable level and no pronounced trend or seasonality. The initial forecast F₁ must be specified, often as the first actual observation or an average of early observations.

Worked Example 2: Exponential smoothing

Suppose the previous forecast was 500 units, actual demand was 560 units, and α = 0.3. Then:

The forecast rises by only 18 units, not by the full 60-unit error, because exponential smoothing deliberately filters random variation.

4.5 Trend-adjusted forecasting

When demand has a sustained trend, simple smoothing tends to lag. Holt’s method estimates both level and trend:

Here, L_t is the smoothed 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 a reasonably stable local trend can be estimated. It should be monitored because an apparent trend may be temporary.

4.6 Seasonality

Seasonality is a repeating pattern with a known period, such as 12 months, 7 days, or 4 quarters. Two common representations are:

• Additive seasonality: seasonal effects are approximately constant in absolute units.
• Multiplicative seasonality: seasonal effects vary in proportion to the level of demand.

For additive decomposition:

For multiplicative decomposition:

where L_t is level, T_t is trend, S_t is seasonal effect or index, and E_t is irregular error. These models are conceptual decompositions; the selected form must fit the data.

For multiplicative seasonal indices, the average index across one complete seasonal cycle is commonly normalized to 1.00. An index of 1.20 indicates demand approximately 20% above the non-seasonal baseline, while 0.80 indicates approximately 20% below it.

A seasonal forecast may be written as:

where B_t is the estimated baseline demand and S_t is the seasonal index. Additive forecasts instead use F_t = B_t + S_t. The choice depends on whether seasonal variation is proportional to or independent of the demand level.

4.7 Regression and causal forecasting

A simple linear trend model is:

where y_t is demand, t is time, a is the intercept, b is the estimated change per period, and ε_t is the unexplained error. With explanatory variables, a multiple regression may be expressed as:

Regression analysis requires careful attention to data quality, causality, omitted variables, changing relationships, and correlated predictors. A strong statistical association does not automatically prove that one variable causes another.

5. Measuring Forecast Error, Accuracy, and Bias

For period t, define forecast error as:

A positive error means actual demand exceeded the forecast; a negative error means the forecast was higher than actual demand. Organizations must state their sign convention because alternative definitions are possible.

Mean error and bias

Mean error, also called mean forecast error under this convention, is:

A persistent positive ME indicates under-forecasting, while a persistent negative ME indicates over-forecasting. Positive and negative errors can cancel, so ME alone is not an adequate measure of accuracy.

Tracking signal is one common bias indicator:

where RSFE = ∑e_t is running sum of forecast errors and MAD is mean absolute deviation:

Tracking signal is dimensionless. It should be interpreted using an organizational control limit; no universal limit is appropriate for every product or process.

Accuracy measures

Mean squared error is:

Root mean squared error is:

MSE gives greater influence to large errors. RMSE returns to the original demand unit and is useful when large misses are especially costly.

Mean absolute percentage error is:

MAPE is interpretable as an average percentage error, but it is undefined when actual demand is zero and unstable when actual demand is close to zero. For intermittent demand, MAE, MAD, RMSE, or scaled measures may be more appropriate.

The choice of metric should reflect the decision. If stockouts are extremely costly, large errors may deserve greater weight. If demand is intermittent, percentage measures may mislead. Forecasts should be evaluated on a holdout period or through rolling-origin testing rather than only on the data used to fit the method.

Worked Example 3: Error measures

Actual demand for three periods was 100, 120, and 80 units. Forecasts were 90, 130, and 70 units. Errors are 10, −10, and 10 units.

The forecasts have a small positive average error, suggesting some under-forecasting, but their average absolute error is 10 units. The result should be interpreted with more observations before changing the method.

6. Selecting and Implementing a Forecasting Method

Method selection should be based on purpose, data, demand behaviour, cost, and required explanation. A practical process is:

1. Define the item, customer segment, forecast horizon, time bucket, and required unit of measure.
2. Clean the data by checking missing values, returns, stockouts, one-time orders, calendar effects, and changes in product definition.
3. Plot the demand history and examine level, trend, seasonality, outliers, and intermittency.
4. Establish a naïve benchmark.
5. Compare suitable methods using rolling historical tests.
6. Examine both accuracy and bias.
7. Incorporate known events such as promotions, closures, contracts, or product launches.
8. Review the forecast with relevant experts.
9. Record assumptions, overrides, and reasons for changes.
10. Monitor forecast performance and revise the method when conditions change.

A statistically superior method is not automatically operationally superior. A method that is slightly less accurate but transparent, stable, and easy to maintain may be preferable to a complex model that users do not understand or trust.

7. Collaborative Forecasting and Demand Uncertainty

Demand uncertainty is often increased by information delays, order batching, promotions, long lead times, and limited visibility. In a supply chain, small changes in downstream demand can create larger fluctuations upstream, a phenomenon commonly called the bullwhip effect.

Collaborative forecasting seeks to reduce this effect by sharing information and coordinating assumptions among parties. Participants may include customers, sales teams, marketing, production, purchasing, logistics, and suppliers. A collaborative process should distinguish:

• statistical baseline: the forecast generated from historical data;
• known-event adjustments: changes supported by promotions, contracts, launches, or closures;
• constraints: capacity, material, transport, or supplier limitations; and
• consensus forecast: the agreed planning value after review.

A useful governance cycle is often called collaborative planning, forecasting, and replenishment. It generally includes joint business planning, forecast creation, exception review, order or replenishment planning, and performance evaluation.

Collaboration does not mean that every participant simply averages their opinion. It requires common definitions, agreed data ownership, documented assumptions, conflict-resolution rules, and a clear distinction between demand and supply constraints. If a supplier reduces the forecast because it lacks capacity, the number becomes a constrained supply plan rather than an unbiased estimate of customer demand.

Managing uncertainty operationally

Forecast uncertainty should be translated into decisions through:

• safety stock and reorder-point policies;
• capacity flexibility and cross-training;
• postponement and modular product design;
• alternative suppliers;
• scenario planning;
• order-priority rules; and
• rapid sensing of actual demand.

A simple reorder-point relationship is:

where μ_L is expected demand during lead time and SS is safety stock. The equation does not determine safety stock by itself; safety stock depends on service objectives, demand variability, lead-time variability, and the selected probability assumptions. Therefore, a forecast should never be confused with the total inventory requirement.

8. Technology, Analytics, and AI

Enterprise resource planning systems, point-of-sale data, customer relationship systems, and cloud platforms allow organizations to collect demand signals more frequently. Dashboards can display forecast accuracy, bias, stockouts, overrides, and exceptions by product or location.

Machine-learning methods may identify nonlinear relationships and interactions among price, promotions, weather, online searches, and customer behaviour. However, they require sufficient data, stable definitions, careful validation, and governance. They may fail when a market experiences a structural break for which no relevant historical example exists.

AI-generated forecasts should therefore be subject to:

• data-quality checks;
• out-of-sample validation;
• human review of unusual results;
• explainability appropriate to the decision;
• access controls and privacy protection; and
• monitoring for systematic bias across customers or regions.

Technology improves calculation and visibility, but it does not eliminate uncertainty or replace operational judgement.

9. Applied Case: Northstar Appliances

The following is an illustrative case.

Background

Northstar Appliances assembles compact air-cooling units for commercial buildings. Demand is highest from March through June. The company sells through distributors, who often place large orders shortly before the warm season. Production lead time is six weeks, while key imported motors require ten weeks.

Problem

Northstar’s sales department increased the annual forecast after a promotional campaign, but operations received no information about the campaign timing. The result was excess finished-goods inventory after the season and shortages during the first promotional month. The company also discovered that forecasts were being measured with different error definitions by sales and production.

Analysis

The planning team separated historical demand into monthly observations, identified the seasonal pattern, and removed two months affected by a one-time distributor closure from the baseline analysis. It compared a 3-month moving average, exponential smoothing, and a seasonal baseline. The moving average lagged during the rapid seasonal increase. Simple exponential smoothing also underestimated the initial rise. A seasonal model gave better rolling-test accuracy and less bias, provided that promotional information was added separately.

The team defined error consistently as e_t = A_t − F_t. It reported MAD for operational interpretation, RMSE for the cost of large misses, and tracking signal for bias. It also classified distributor orders as either normal replenishment or promotion-related demand.

Approach and solution

Northstar introduced a monthly consensus meeting involving sales, marketing, production, purchasing, and two major distributors. Marketing supplied promotion calendars eight weeks before launch. The statistical seasonal forecast became the baseline; documented event adjustments were then reviewed. Purchasing received a range consisting of a central forecast and an uncertainty indicator rather than one apparently precise number.

Production adopted a flexible capacity arrangement using trained temporary labour during the seasonal peak. The company postponed final packaging until distributor requirements were clearer. A dashboard displayed forecast bias, stockout incidents, excess inventory, and forecast overrides by product family.

Outcome

After two planning cycles, the company reduced early-season stockouts and avoided repeating the previous post-season surplus. The greatest improvement came not from selecting a more complicated algorithm, but from separating baseline demand from promotion effects and sharing information earlier.

Learning points

1. Seasonality must be represented explicitly when it is stable and material.
2. Sales events should be recorded as causal information, not hidden inside unexplained adjustments.
3. A consensus forecast requires governance and shared definitions.
4. Forecast accuracy and operational performance are related but not identical.
5. Capacity flexibility and postponement can reduce the cost of residual uncertainty.

10. Challenges, Limitations, and Ethical Considerations

Forecasting is vulnerable to structural breaks, new products, competitor actions, supply shortages, inflation, weather events, and changes in customer behaviour. Historical data may also contain censoring: observed sales may be lower than true demand because inventory was unavailable. Treating all observed sales as unconstrained demand can systematically understate future requirements.

Human judgement introduces risks such as anchoring, overconfidence, groupthink, and strategic bias. Forecasts may be deliberately understated to make targets easier or overstated to secure capacity. Clear accountability, independent review, and an audit trail for overrides help reduce these problems.

Customer and transaction data may contain personal or commercially sensitive information. Organizations should follow applicable privacy, data-protection, retention, and access-control requirements. Forecasting systems should not discriminate against customer groups through biased data or inappropriate allocation rules. During shortages, allocation criteria should be transparent, consistent, and aligned with contractual and ethical obligations.

Best-Practice Checklist

• Forecast at the level at which decisions are made.
• Use a simple benchmark before adopting a complex method.
• Separate demand from constrained shipments where possible.
• Represent trend and seasonality explicitly when evidence supports them.
• Measure both accuracy and bias.
• Use rolling, out-of-sample evaluation.
• Record overrides and their business reasons.
• Share promotions, launches, closures, and contract information early.
• Translate uncertainty into inventory, capacity, and sourcing decisions.
• Review models after structural changes rather than assuming historical relationships continue.

Chapter Summary

Demand management links customer requirements with production, inventory, capacity, purchasing, and distribution decisions. Forecasting provides estimates of future demand, but forecasts are uncertain and must be interpreted with their assumptions and limitations. Qualitative methods such as executive judgement, sales-force estimates, market research, Delphi panels, and analogy are particularly useful when data are limited or markets are changing. Quantitative methods include naïve forecasts, moving averages, weighted moving averages, exponential smoothing, trend models, seasonal models, and causal regression.

Moving averages smooth historical observations; weighted averages emphasize selected periods; exponential smoothing updates forecasts in response to recent error; and trend or seasonal methods address systematic patterns that simpler methods may miss. Forecast performance should be assessed with measures such as ME, MAD, MSE, RMSE, MAPE, and tracking signal, recognizing that each has limitations. Collaborative forecasting combines statistical baselines with market knowledge, event information, constraints, and agreed governance. Technology and AI can improve speed and visibility, but they require sound data, validation, monitoring, and human accountability.

Key Terms

Forecast: An estimate of future demand based on information and assumptions.

Demand management: The process of sensing, forecasting, influencing, prioritizing, and coordinating demand.

Independent demand: Demand originating from external customers or markets.

Dependent demand: Demand derived from the requirement for another item.

Time series: A sequence of observations recorded over ordered time periods.

Level: The underlying average magnitude of demand.

Trend: A persistent upward or downward movement in demand.

Seasonality: A repeating demand pattern with a known frequency.

Moving average: A forecast calculated from the average of recent observations.

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

Exponential smoothing: A recursive method that updates a forecast using recent forecast error.

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

Bias: A persistent tendency to over-forecast or under-forecast.

MAD: Mean absolute deviation, the average absolute forecast error.

RMSE: Root mean squared error, which emphasizes large errors and is expressed in demand units.

MAPE: Mean absolute percentage error.

Tracking signal: A ratio used to monitor cumulative forecast error relative to MAD.

Collaborative forecasting: Forecast development involving multiple functions or supply-chain partners.

Bullwhip effect: Amplification of demand variation as information moves upstream in a supply chain.

Demand shaping: Actions intended to influence the volume or timing of demand.

Review Questions

Short Answer Questions

1. Distinguish independent demand from dependent demand.
2. What is the main effect of increasing the window size in a simple moving average?
3. Interpret a positive forecast error when error is defined as actual demand minus forecast demand.
4. State one limitation of MAPE.
5. Why should a consensus forecast distinguish demand from supply constraints?

Long Answer Questions

1. Explain the role of demand management in production planning and control, including at least four ways an organization may shape demand.
2. Compare qualitative forecasting with quantitative forecasting, and identify situations in which each is appropriate.
3. Explain simple exponential smoothing, define its variables, and discuss the effects of a high and low smoothing constant.
4. Discuss the difference between additive and multiplicative seasonality and explain how the choice affects forecasting.
5. Explain how collaborative forecasting can reduce the bullwhip effect and identify the governance conditions required for success.

Analytical / Application Questions

1. Demand in the last four periods was 280, 310, 295, and 325 units. Calculate the three-period simple moving-average forecast for the next period. Explain one limitation of this forecast.
2. A forecast is 800 units and actual demand is 920 units. Calculate the forecast error using e_t = A_t − F_t. If the smoothing constant is 0.25, calculate the next exponential-smoothing forecast.
3. A seasonal product has a baseline forecast of 1,500 units for a month with a seasonal index of 1.30. Calculate the multiplicative seasonal forecast and explain two operational actions that could manage the associated uncertainty.

Multiple Choice Questions with Answers

Question 1

Which statement best describes independent demand?

A. Demand calculated from a bill of materials

B. Demand generated by external customers

C. Demand created only by internal maintenance

D. Demand that is always constant

Correct Answer: B. Demand generated by external customers

Question 2

A three-period moving average uses demands of 100, 120, and 140 units. What is the forecast for the next period?

A. 110 units

B. 115 units

C. 120 units

D. 130 units

Correct Answer: C. 120 units

Question 3

In simple exponential smoothing, what happens when α is increased?

A. The forecast responds more strongly to recent observations

B. The forecast ignores the latest observation

C. The forecast becomes independent of actual demand

D. The forecast must become seasonal

Correct Answer: A. The forecast responds more strongly to recent observations

Question 4

Using e_t = A_t − F_t, what does a negative error indicate?

A. Actual demand exceeded the forecast

B. Actual demand equalled the forecast

C. The forecast exceeded actual demand

D. The forecast has no measurable bias

Correct Answer: C. The forecast exceeded actual demand

Question 5

Which measure gives greater influence to large forecast errors?

A. Mean error

B. Mean squared error

C. Mean absolute percentage error in every circumstance

D. Seasonal index

Correct Answer: B. Mean squared error

Question 6

A seasonal index of 1.20 in a multiplicative model indicates that demand is approximately:

A. 20% below the baseline

B. Equal to the baseline

C. 12% above the baseline

D. 20% above the baseline

Correct Answer: D. 20% above the baseline

Question 7

What is a principal purpose of a naïve forecast?

A. To provide a simple benchmark

B. To remove all random variation

C. To estimate causal effects automatically

D. To guarantee unbiased forecasts

Correct Answer: A. To provide a simple benchmark

Question 8

Why can MAPE be unsuitable when actual demand is zero?

A. It becomes dimensionless

B. It cannot be calculated because division by zero occurs

C. It always equals zero

D. It automatically becomes RMSE

Correct Answer: B. It cannot be calculated because division by zero occurs

Question 9

Which practice supports collaborative forecasting?

A. Allowing each department to use a different error definition

B. Hiding promotion information until orders arrive

C. Separating the statistical baseline from documented event adjustments

D. Treating supply capacity as customer demand

Correct Answer: C. Separating the statistical baseline from documented event adjustments

Question 10

A baseline forecast is 1,000 units and the multiplicative seasonal index is 0.80. What is the seasonal forecast?

A. 800 units

B. 1,000 units

C. 1,080 units

D. 1,250 units

Correct Answer: A. 800 units