Demand Forecasting and Operations Planning

Introduction

Operations cannot be planned intelligently unless an organisation has a reasoned view of future demand. Demand forecasting provides that view. It does not eliminate uncertainty; rather, it makes uncertainty visible and supports decisions about capacity, labour, inventory, procurement, scheduling and customer-service policies.

Forecasting is particularly important in total quality management (TQM). Poor forecasts can create overproduction, shortages, expediting, excessive inventory, unstable schedules and dissatisfied customers. These outcomes increase process variation and encourage corrective action after problems have occurred. A disciplined forecasting and planning process is preventive: it anticipates likely requirements and aligns resources before failure occurs.

Forecasts should therefore be treated as managerial inputs, not as unquestionable predictions. A forecast is a statement about expected demand under specified assumptions, at a specified time horizon, using specified information. Its quality depends not only on statistical accuracy but also on data integrity, process discipline, cross-functional participation and appropriate managerial judgement.

Learning Objectives

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

1. Explain the purpose, horizons and components of demand forecasting.
2. Select suitable qualitative and quantitative forecasting methods.
3. Calculate moving-average, weighted-moving-average, exponential-smoothing and simple-regression forecasts.
4. Evaluate forecasts using error measures and interpret their limitations.
5. Explain aggregate planning and sales and operations planning (S&OP).
6. Analyse the relationship between forecasting, capacity, inventory, workforce and demand management.
7. Apply forecasting and planning principles to an operations case.

1. Demand, Forecasts and Planning Horizons

Demand is the quantity of a product or service that customers are willing and able to purchase during a defined period. It is not always identical to sales. Sales may be constrained by stockouts, production limits, distribution problems or credit restrictions. For forecasting purposes, organisations should distinguish:

• Observed sales: transactions actually completed.
• True demand: customer requirements, including unmet requirements where they can be estimated.
• Dependent demand: demand calculated from the requirement for another item, such as components in a bill of materials.
• Independent demand: demand determined primarily by external customers or markets.
• Derived demand: demand generated by another activity, such as transport demand generated by trade.

Forecasting horizons normally include:

The appropriate method depends on the horizon. Short-term forecasts often rely heavily on recent observations and operational information. Long-term forecasts require market research, economic assumptions, technology analysis and scenario planning because historical patterns may not continue.

A demand series may contain four useful components:

1. Level: the underlying average quantity.
2. Trend: a persistent upward or downward movement.
3. Seasonality: a repeating pattern linked to calendar or operating cycles.
4. Random variation: irregular movements that cannot be reliably predicted.

A series may also contain cycles, which are longer and less regular movements associated with economic or industry conditions. A useful conceptual model is additive:

Here, Y_t is observed demand in period t, L_t is level, T_t is trend, S_t is seasonal effect and ε_t is random error. The additive form is appropriate when seasonal fluctuations are approximately constant in absolute units. A multiplicative model, Y_t = L_t × T_t × S_t × ε_t, is more appropriate when seasonal variation grows in proportion to the level. The multiplicative model requires positive, ratio-scale data and is often analysed using logarithms.

2. The Forecasting Process

A professional forecasting process usually follows these steps:

1. Define the purpose, item, unit of measure and forecast horizon.
2. Identify the decision maker and the consequences of error.
3. Collect, clean and reconcile historical and external data.
4. Examine the series for trend, seasonality, outliers, promotions and structural changes.
5. Select one or more methods.
6. Produce the statistical forecast.
7. Incorporate documented managerial information.
8. Measure error and monitor bias.
9. Communicate assumptions, confidence and risks.
10. Review the forecast after actual demand becomes available.

The unit and time bucket must be consistent. For example, a monthly forecast of units cannot be directly compared with weekly labour hours without a conversion. Data should also distinguish regular demand from one-time events, cancelled orders, returns and stockout periods.

3. Qualitative Forecasting Methods

Qualitative methods are especially valuable when historical data are limited, the market is changing, or the forecast concerns a new product. They are also useful for challenging a statistical forecast with information that has not yet appeared in historical data.

3.1 Executive opinion

Senior managers combine knowledge of strategy, markets, finance and operations. The method is rapid but can suffer from hierarchy effects, overconfidence and anchoring. A documented rationale and independent estimates reduce these risks.

3.2 Sales-force composite

Sales representatives estimate demand in their territories, and the estimates are aggregated. This uses local customer knowledge but may be distorted by incentives, optimism or deliberate understatement of targets.

3.3 Customer and market research

Surveys, interviews, pilot sales and conjoint studies provide information about customer intentions and preferences. Stated intention is not always equal to actual purchase behaviour, so results should be validated where possible.

3.4 Delphi method

A panel of experts provides estimates anonymously through several rounds. A facilitator summarises the responses without identifying individuals, and participants revise their estimates after reviewing the group summary. Delphi is useful when evidence is sparse and consensus is needed, but the choice of experts and wording of questions strongly influence the result.

3.5 Analogy and scenario analysis

A new product may be compared with a similar product, market or technology. Scenario analysis develops coherent alternatives, such as high-growth, base and low-growth cases. Scenarios are not probability-free forecasts; they are structured assumptions used to test the resilience of plans.

Qualitative judgement should be traceable. Organisations should record who supplied it, what evidence was used, which assumptions changed and whether the judgement improved subsequent accuracy.

4. Quantitative Time-Series Methods

Quantitative methods use numerical observations. They are most effective when the historical process is sufficiently stable and the data are measured consistently. They should not be applied mechanically when a product has been discontinued, a market has changed fundamentally or historical sales were constrained by stockouts.

4.1 Naïve forecast

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

F_t+1 is the forecast for the next period and A_t is actual demand in the current period. This method is a useful benchmark. A more seasonal naïve forecast uses the observation from the same season in the previous cycle:

where m is the number of periods in a seasonal cycle.

4.2 Simple moving average

A moving average uses the most recent n observations:

The method is appropriate when demand has no strong trend or seasonality and random variation is expected to average out. A larger n produces greater smoothing but responds more slowly to changes. A smaller n responds more quickly but is more sensitive to random fluctuations.

Worked example. Demand during four consecutive months was 420, 460, 440 and 480 units. A three-month moving-average forecast for month 5 is:

The oldest observation, 420, is excluded because only the latest three observations are used.

4.3 Weighted moving average

A weighted moving average assigns different weights to recent observations:

The weights should satisfy w_i ≥ 0 and ∑w_i = 1. Larger weights on recent observations make the forecast more responsive. The method remains subjective because the weights must be selected and validated.

Using the same demand data, suppose the weights for the latest three months are 0.5, 0.3 and 0.2. Then:

4.4 Exponential smoothing

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

Equivalently:

Here, α is the smoothing constant, where 0 < α < 1. A high value gives more emphasis to recent demand; a low value produces greater smoothing. The basic method assumes no systematic trend or seasonality. If these exist, trend-adjusted or seasonal methods are more suitable.

Worked example. If the current forecast is 450 units, actual demand is 480 units and α = 0.3, then:

The initial forecast can affect early results. For a long series its influence usually declines, but analysts should document the starting value.

4.5 Trend-adjusted exponential smoothing

When a series has trend, Holt’s method estimates both level and trend. One common formulation is:

L_t is the estimated level, T_t is the estimated trend, h is the number of periods ahead, and α and β are smoothing constants between 0 and 1. This method assumes that a locally approximately linear trend is useful for the forecast horizon. It can overreact when a temporary increase is incorrectly interpreted as a permanent trend.

4.6 Seasonal forecasting

Seasonal indices express the typical relative effect of a season. If a deseasonalised forecast is D_t and the seasonal index is S_t, a multiplicative seasonal forecast is:

Seasonal indices should normally average 1.00 across a complete cycle. For example, an index of 1.20 indicates demand approximately 20% above the deseasonalised level, while 0.80 indicates demand approximately 20% below it. The method requires enough repeated cycles and stable seasonal behaviour.

5. Causal Forecasting and Regression

Time-series methods explain demand through its own history. Causal methods relate demand to explanatory variables such as price, advertising expenditure, temperature, income, population or delivery lead time.

For one explanatory variable, the simple linear regression model is:

The fitted forecast is:

Y_t is observed demand, X_t is the explanatory variable, a is the intercept, b is the slope and ε_t is random error. In ordinary least squares regression:

The slope has units of demand per unit of X. The intercept may have no practical interpretation if zero is outside the observed range. Regression is useful only when the relationship is plausible, data quality is adequate and the future value of X can itself be estimated.

Worked example. Suppose a fitted model is:

where X is advertising expenditure measured in thousands of currency units and Ŷ is weekly demand in units. If planned expenditure is 10, the forecast is:

This is not proof that advertising causes every unit of demand. The model may omit price, competitor activity, seasonality and market conditions. Analysts should examine residuals, influential observations, correlation, coefficient uncertainty and out-of-sample performance. A high R² alone does not establish causation or guarantee useful forecasts.

6. Measuring Forecast Error

Let the forecast error be actual demand minus forecast:

A positive error means actual demand exceeded the forecast. Consistency in the sign convention is essential.

Here, n is the number of evaluated periods. MAPE is undefined when actual demand is zero and unstable when actual demand is close to zero. In such cases, MAD, RMSE or a scaled measure is preferable. Percentage errors can also be misleading when over- and under-forecasting have different operational costs.

Worked comparison. Actual demand over three periods was 100, 120 and 80 units; forecasts were 90, 130 and 100 units. Errors are 10, −10 and −20 units.

The negative ME indicates overall overforecasting. Accuracy measures should be calculated on a common test period; comparing methods on different periods can produce an unfair conclusion.

Forecast monitoring should consider both accuracy and bias. A forecast with small average absolute error but persistent bias can systematically create shortages or excess stock. A tracking signal is sometimes defined as:

where RSFE = ∑e_t is the running sum of forecast errors. Control limits must be established according to the organisation’s policy and the statistical context; a tracking signal is a diagnostic, not a universal law.

7. Aggregate Planning and Sales and Operations Planning

Forecasts become operationally valuable when translated into a feasible plan. Aggregate planning determines medium-term levels of production, workforce, inventory, backorders and capacity for product families rather than individual stock-keeping units.

Two broad strategies are common:

• Level strategy: maintain relatively stable production and workforce, absorbing demand variation through inventory, backorders or promotions.
• Chase strategy: vary production or workforce to follow demand more closely.

Hybrid strategies combine regular labour, overtime, subcontracting, inventory and flexible staffing. The appropriate choice depends on labour regulations, hiring and dismissal costs, inventory carrying costs, service requirements, product perishability and capacity flexibility.

A simplified aggregate-planning balance is:

where I_t is ending inventory, P_t is production and D_t is demand in period t. If backorders are permitted, I_t may be negative, although the organisation should define how backlog is recorded and costed. With workforce and capacity restrictions, production must also satisfy feasible bounds such as P_min ≤ P_t ≤ P_max.

A typical cost objective is:

where c_p is unit production cost, I_t^+ is positive inventory, I_t^- is backlog, c_h is holding cost, c_b is backorder cost, O_t is overtime, S_t is subcontracting and the c terms are their respective costs. This is a simplified model; real plans may include setup costs, minimum batch sizes, multiple resources and service constraints.

Sales and operations planning (S&OP) is a recurring cross-functional process that reconciles demand and supply. Sales and marketing provide market intelligence; operations assesses capacity and materials; finance tests profitability and cash implications; senior management approves trade-offs. A typical cycle includes data review, demand review, supply review, reconciliation of scenarios and an executive decision meeting.

S&OP is not merely a forecast meeting. It is a governance process that establishes one agreed set of assumptions and a feasible plan. It should include exceptions, risks, responsible owners and triggers for replanning.

8. Demand Management and Quality Integration

Demand management influences the timing, quantity and mix of demand rather than simply predicting it. Techniques include pricing, promotions, reservations, appointment systems, product substitution, postponement, delivery-date negotiation and customer segmentation. These actions must be evaluated carefully because a promotion can shift demand between periods without increasing total demand.

In TQM, forecasting and planning should be integrated with the plan–do–check–act (PDCA) cycle:

• Plan: select methods, define assumptions and establish performance targets.
• Do: issue the forecast and execute the aggregate plan.
• Check: compare actual demand, forecast error, service and cost.
• Act: revise methods, data processes, parameters and management responses.

Forecasting is also a process subject to variation. A process map can identify defects such as late data, inconsistent product codes, manual overrides without justification and failure to close the feedback loop. Root-cause analysis should distinguish model inadequacy from process failure. For example, a forecast may be inaccurate because the method is unsuitable, because sales were lost during a stockout, or because a promotion was not recorded in the data.

9. Technology, Data and Artificial Intelligence

Enterprise resource planning, point-of-sale systems, cloud analytics and integrated planning platforms can automate data collection and forecast updates. Machine-learning models may identify nonlinear relationships, interactions and high-dimensional patterns. They can be valuable when large, clean and representative datasets exist.

However, technological sophistication does not remove fundamental requirements. Models require reliable definitions, stable data pipelines, appropriate validation and explainable outputs. A complex model can overfit historical noise and perform poorly after a market change. Forecasts should be tested using rolling-origin or time-ordered validation rather than random shuffling that allows future information to leak into the training set.

Governance should address access rights, data privacy, cybersecurity, model version control, audit trails and human accountability. Demand forecasts may reveal commercially sensitive customer behaviour. Automated recommendations should not create discriminatory service priorities or conceal assumptions from decision makers. Human overrides should be permitted when justified, but excessive overrides can destroy the value of a statistical process and should be analysed.

10. Illustrative Case: Meridian Appliances

Background

Meridian Appliances assembles a small domestic water pump. Monthly demand is seasonal, with higher sales before the agricultural planting season. The company has experienced both stockouts and excess finished goods. Sales representatives provide optimistic estimates, while production managers prefer stable schedules.

Problem

The company must establish a six-month plan. Historical sales contain one month affected by a stockout and one month containing a large one-time infrastructure order. Treating both observations as ordinary demand would distort the forecast.

Analysis

The planning team first separates regular retail demand from the exceptional order and estimates lost sales during the stockout using distributor information. It then creates a baseline seasonal forecast using historical seasonal indices. A three-month moving average is retained as a benchmark, while a causal model incorporating planting-area indicators is tested for medium-term planning. Forecast accuracy is assessed using a rolling holdout period and MAD and RMSE, rather than MAPE alone, because some regional demand values are close to zero.

The demand review identifies a likely 8% increase in the next season because of distributor expansion. The supply review finds that normal production is 4,000 units per month, overtime can add 600 units, and subcontracting can add 800 units at higher cost. Finished-goods capacity is 1,500 units. The team compares a level plan, a chase plan and a hybrid plan.

Approach and solution

The approved hybrid plan keeps a core workforce and stable regular production, builds inventory before the seasonal peak, uses limited overtime during the peak and reserves subcontracting for demand above the agreed risk threshold. The exceptional infrastructure order is excluded from the baseline but retained as a scenario. The plan includes a weekly exception review when actual orders exceed the forecast by more than a defined tolerance.

Sales and operations agree that a promotion will be scheduled in a low-demand month only if production and distribution capacity are available. Sales targets are separated from unconstrained market demand so that management can distinguish an ambitious target from a statistically supported forecast.

Outcome

In the illustrative scenario, the hybrid plan reduces stockout risk without requiring the permanent workforce changes associated with a full chase strategy. The most important improvement is not a particular algorithm; it is the disciplined separation of data, assumptions, scenarios and decisions. After three cycles, the team finds that unexplained overrides are a major source of bias and introduces an approval log.

Learning points

1. Sales history may not equal unconstrained demand.
2. Exceptional events require investigation, not automatic deletion.
3. Forecast accuracy and plan feasibility are separate questions.
4. Cross-functional agreement improves implementation.
5. A forecast should trigger decisions, review rules and accountability.

11. Limitations and Best Practices

No forecasting method is universally superior. Moving averages lag turning points; exponential smoothing depends on parameter choice; regression can confuse correlation with causation; qualitative methods can be biased; machine learning can be opaque and unstable under structural change.

Best practice is therefore comparative and disciplined:

• establish a transparent naïve benchmark;
• use a method appropriate to level, trend and seasonality;
• segment items by volume, variability, value and strategic importance;
• evaluate forecasts out of sample and over multiple periods;
• measure bias as well as absolute accuracy;
• record promotions, stockouts and structural breaks;
• combine statistical evidence with documented expert knowledge;
• translate forecasts into capacity, inventory and service decisions;
• maintain one agreed demand plan while preserving alternative scenarios;
• review forecast performance and improve the process continuously.

Forecasting should not be judged solely by numerical accuracy. A slightly less accurate forecast may be preferable if it is timely, interpretable and supports better decisions. Conversely, a highly accurate forecast is of limited value if it cannot be converted into a feasible operating plan.

Chapter Summary

Demand forecasting provides an informed estimate of future customer requirements and supports decisions about resources, capacity, inventory and service. Forecast horizons and data characteristics determine method choice. Qualitative methods use expert, customer and market knowledge; quantitative methods include naïve forecasts, moving averages, exponential smoothing, seasonal methods and regression.

Moving averages smooth random variation, weighted averages emphasise selected observations and exponential smoothing updates forecasts in response to recent error. Trend-adjusted and seasonal methods are required when level alone is inadequate. Regression relates demand to explanatory variables but requires careful interpretation and validation. Forecast error should be evaluated using measures such as ME, MAD, MSE, RMSE and MAPE, recognising their assumptions and limitations.

Aggregate planning converts demand expectations into medium-term decisions concerning production, workforce, inventory, overtime, subcontracting and backorders. S&OP provides a cross-functional governance process for reconciling demand, supply and financial implications. Demand management can shape demand through pricing, promotions and scheduling. TQM principles require forecasting to be treated as a controlled, measured and continuously improved process, supported by reliable data, ethical governance and appropriate technology.

Key Terms

Demand forecast: An estimate of future customer requirements for a defined item, period and set of assumptions.

Independent demand: Demand determined primarily by external customers.

Dependent demand: Demand calculated from the requirement for another item or product.

Forecast horizon: The future time span covered by a forecast.

Level: The underlying average value of a demand series.

Trend: A persistent upward or downward movement in demand.

Seasonality: A repeating demand pattern associated with a regular cycle.

Naïve forecast: A forecast based on the most recent, or corresponding seasonal, observation.

Moving average: An average of a specified number of recent observations.

Weighted moving average: A moving average assigning different weights to observations.

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

Regression: A statistical method relating a dependent variable to one or more explanatory variables.

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

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

RMSE: Root mean squared error, an error measure that penalises large errors.

Forecast bias: Persistent tendency to overforecast or underforecast.

Aggregate planning: Medium-term planning of resources for product families or service groups.

Sales and operations planning: A cross-functional process reconciling demand, supply and business plans.

Demand management: Actions that influence the timing, quantity or mix of demand.

Tracking signal: A diagnostic comparing cumulative forecast error with an error scale such as MAD.

Review Questions

Short Answer Questions

1. Distinguish between sales and true demand.
2. What is the effect of increasing the number of periods in a simple moving average?
3. State the basic exponential-smoothing equation and define its smoothing constant.
4. Why can MAPE be unsuitable when actual demand is zero or close to zero?
5. What is the purpose of an S&OP process?

Long Answer Questions

1. Explain the main qualitative forecasting methods and discuss the bias risks associated with each.
2. Compare simple moving averages, weighted moving averages and exponential smoothing in terms of responsiveness and data requirements.
3. Explain the assumptions and managerial interpretation of a simple linear regression forecast.
4. Discuss how aggregate planning balances level and chase strategies.
5. Explain how forecasting can be integrated into TQM and the PDCA cycle.

Analytical / Application Questions

1. Monthly demand for the last four months was 600, 640, 620 and 680 units. Calculate the three-month moving-average forecast and a weighted moving-average forecast using weights 0.5, 0.3 and 0.2 for the latest three observations. Which forecast is more responsive, and why?
2. A forecast of 950 units was issued for a period in which actual demand was 1,000 units. Calculate the forecast error and absolute percentage error. Explain one operational consequence of underforecasting.
3. A company faces stable base demand, strong seasonal peaks and limited finished-goods capacity. Recommend a forecasting and aggregate-planning approach, including data treatment, forecast evaluation, workforce policy and demand-management actions.

Multiple Choice Questions with Answers

Question 1

Which condition most directly supports the use of a simple moving average?

A. Demand has no strong trend or seasonality and random variation should be smoothed

B. Demand is determined entirely by advertising expenditure

C. The product has never been sold before

D. Historical observations are all from different units of measure

Correct Answer: A. Demand has no strong trend or seasonality and random variation should be smoothed

Question 2

Demand in the latest three months was 460, 440 and 480 units. What is the three-month moving-average forecast for the next month?

A. 440 units

B. 450 units

C. 460 units

D. 480 units

Correct Answer: C. 460 units

Question 3

Using a smoothing constant of 0.3, a previous forecast of 450 units and actual demand of 480 units, what is the next exponential-smoothing forecast?

A. 453 units

B. 459 units

C. 465 units

D. 480 units

Correct Answer: B. 459 units

Question 4

In the convention e_t = A_t − F_t, what does a positive forecast error indicate?

A. Actual demand exceeded the forecast

B. The forecast exceeded actual demand

C. The forecast was exactly accurate

D. The seasonal index was below 1

Correct Answer: A. Actual demand exceeded the forecast

Question 5

Which measure is expressed in the original units of demand and averages absolute errors?

A. MAPE

B. MSE

C. MAD

D. R-squared

Correct Answer: C. MAD

Question 6

A fitted regression model is Ŷ = 120 + 8X. What is the forecast when X = 10?

A. 128 units

B. 180 units

C. 200 units

D. 960 units

Correct Answer: C. 200 units

Question 7

Which statement best describes a chase aggregate-planning strategy?

A. Production and workforce are kept constant regardless of demand

B. Production or workforce is varied to follow demand more closely

C. All demand variation is handled by permanent inventory

D. Demand is ignored until the end of the planning horizon

Correct Answer: B. Production or workforce is varied to follow demand more closely

Question 8

Why can MAPE be problematic when actual demand is zero?

A. It becomes a currency measure

B. It cannot be calculated because division by zero occurs

C. It always equals 100%

D. It removes all forecast bias

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

Question 9

Which activity is characteristic of S&OP?

A. Reconciling demand, supply and financial implications across functions

B. Replacing all managerial judgement with one formula

C. Measuring only machine maintenance time

D. Forecasting only individual component requirements

Correct Answer: A. Reconciling demand, supply and financial implications across functions

Question 10

If actual demand is 100 units and the forecast is 90 units, what is the error under e_t = A_t − F_t?

A. −10 units

B. 0 units

C. 10 units

D. 190 units

Correct Answer: C. 10 units