Demand Forecasting and Sales and Operations Planning

Introduction

Production systems rarely fail because managers cannot calculate capacity. They fail because decisions are based on an incomplete or poorly interpreted view of future demand. Forecasting provides an informed estimate of what customers may require, while Sales and Operations Planning (S&OP) converts that estimate into a feasible, financially coherent operating plan. Together, they connect markets with purchasing, production, inventory, logistics, finance and workforce decisions.

A forecast is not a prediction of the future in a deterministic sense. It is a structured estimate based on information available at a particular time. Its value depends not only on statistical accuracy but also on whether it improves decisions. A highly accurate forecast at product-family level may be more useful for capacity planning than a less accurate forecast for thousands of individual stock-keeping units (SKUs). Conversely, an aggregate forecast may be insufficient for short-term replenishment.

S&OP is a cross-functional management process through which demand, supply, inventory, finance and resources are reconciled. It is normally conducted at an aggregate level, such as product family, market segment or plant, over a medium-term horizon. Modern organisations increasingly extend S&OP into integrated business planning, where strategic, financial and operational plans are aligned in one decision system.

Learning Objectives

After completing this chapter, the learner should be able to:

1. Explain the purpose, scope and limitations of demand forecasting.
2. Distinguish qualitative, time-series and causal forecasting methods.
3. Apply moving-average, exponential-smoothing, trend and seasonal methods.
4. Select and interpret forecast-accuracy measures.
5. Explain uncertainty, bias, forecast aggregation and demand sensing.
6. Design the stages, roles and outputs of an S&OP process.
7. Reconcile demand, supply, finance, inventory and workforce requirements.
8. Evaluate technology, governance and ethical issues in forecasting and S&OP.

1. The Nature and Role of Demand Forecasting

Demand is the quantity customers are willing and able to purchase under specified conditions. Sales, by contrast, are orders actually received or transactions completed. The difference matters: a stockout may produce low sales even when underlying demand was high. Forecasting should therefore use the best available representation of true demand, including lost-sales estimates, promotions, cancellations and exceptional events.

Forecasts may be classified by:

• Time horizon: short term (hours to weeks), medium term (months to approximately two years), and long term (multiple years).
• Level of aggregation: SKU, product family, plant, region, market or enterprise.
• Purpose: replenishment, workforce planning, capacity planning, budgeting, strategic investment or network design.
• Information source: judgmental, historical, causal, market-based or hybrid.

A useful forecasting process begins by defining the decision to be supported. The required horizon, detail, frequency and accuracy are consequences of that decision. For example, a weekly production schedule requires a different forecast from a five-year capacity-investment decision.

Forecasting should be integrated with planning rather than treated as an isolated analytical exercise. The planner must ask:

• What demand is expected?
• What demand is commercially desired?
• What demand can the supply system serve?
• What financial result follows?
• What risks and assumptions must management accept?

2. Demand Patterns and Data Preparation

Historical demand commonly contains four components:

1. Level: the underlying average magnitude.
2. Trend: a persistent upward or downward movement.
3. Seasonality: a repeating pattern linked to calendar or operating cycles.
4. Irregular or random variation: unpredictable fluctuations.

A time series may be represented conceptually using an additive model:

where 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 irregular error. The additive form is most appropriate when seasonal variation is approximately constant in absolute units.

A multiplicative representation is:

where the seasonal effect is proportional to the level. Multiplicative models are appropriate when seasonal fluctuations grow as demand grows. The choice should be based on evidence, not convenience.

Before modelling, data should be examined for:

• missing periods and duplicate records;
• changes in product definitions or pack sizes;
• stockouts, rationing and backorders;
• promotions, price changes and competitor actions;
• one-time contracts or abnormal events;
• calendar effects, working days and holidays;
• product launches, phase-outs and substitutions.

A common error is to smooth away an important business event. An unusual sales peak caused by a one-time promotion should not automatically be treated as a permanent increase in baseline demand. Equally, treating a stockout as zero demand can cause systematic underforecasting.

3. Qualitative Forecasting Methods

Qualitative methods are essential when historical data are limited, unstable or structurally irrelevant. They are especially useful for new products, major market changes, technology transitions and long-range scenarios.

Executive opinion and consensus panels

Senior managers combine market, strategic and operational knowledge. This method is fast but vulnerable to hierarchy, optimism and political influence. A documented evidence base and independent estimates reduce groupthink.

Sales-force composite

Sales staff estimate demand in their territories. They often possess current customer information, but may understate demand to create easier targets or overstate it to secure inventory. Incentives and calibration are therefore important.

Customer and market research

Surveys, interviews, conjoint studies and test marketing can reveal preferences and purchase intentions. Stated intention is not identical to actual behaviour, so research results should be combined with observed data.

Delphi method

A panel of experts provides anonymous forecasts through several rounds. A facilitator summarises responses and allows revision without exposing individual identities. Delphi is useful where expertise is dispersed and open discussion might be dominated by senior participants.

Scenario analysis

Managers construct internally consistent futures, such as high growth, constrained supply and recession. Scenarios are not probability-free substitutes for forecasts; they are tools for testing resilience and preparing contingent actions.

Qualitative methods are improved by recording assumptions, separating facts from opinions, using structured aggregation and reviewing forecast performance after outcomes are known.

4. Quantitative Forecasting Methods

4.1 Naïve forecast

The simplest benchmark sets next-period demand equal to the most recent observed demand:

Here, F_t+1 is the forecast for the next period and A_t is actual demand in the current period. The method is useful for stable series and as a benchmark. A seasonal naïve method uses demand from the corresponding previous season:

where m is the seasonal cycle length and h is the forecast horizon.

4.2 Moving average

A simple moving average of n periods is:

It gives equal weight to the most recent n observations. A larger n produces greater smoothing but slower response to change. It is suitable when demand has no strong trend or seasonality.

Worked example. Demand over the last four months was 920, 980, 1,020 and 1,080 units. The three-month moving-average forecast for month 5 is:

The result may be rounded to 1,027 units where whole-unit planning is required. The model does not infer causation; it simply smooths recent observations.

A weighted moving average assigns weights that sum to one:

with w_i ≥ 0 and ∑w_i = 1. Larger weights are commonly given to recent observations.

4.3 Exponential smoothing

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

where 0 < α ≤ 1 is the smoothing constant. Equivalently:

The term A_t − F_t is the forecast error for period t. A high α responds quickly to changes but retains less smoothing; a low value is more stable but slower to react. Single exponential smoothing is not appropriate by itself for strong trend or seasonality.

Worked example. Suppose F_t = 1,000 units, A_t = 1,080 units and α = 0.30. Then:

4.4 Trend-adjusted smoothing

When a series has trend, Holt’s method estimates level and trend separately:

Here, L_t is the level, T_t is the trend, β is the trend-smoothing constant, and h is the number of periods ahead. The parameters satisfy 0 < α ≤ 1 and 0 ≤ β ≤ 1. Initial values must be specified, and the method should be validated against holdout data.

4.5 Seasonal forecasting

For seasonal data, Holt–Winters methods estimate level, trend and seasonal indices. In a multiplicative seasonal model, a common structure is:

where S is the seasonal index and m is the seasonal period. Seasonal indices are generally normalised so their average over one complete cycle is 1.00. Multiplicative seasonality is unsuitable when demand can be zero or when seasonal amplitude is constant in units; additive treatment may then be preferable.

4.6 Causal and regression methods

Causal models relate demand to explanatory variables such as price, advertising, income, weather or economic activity. A simple linear model is:

where Y_t is demand, X_t is an explanatory variable, a is the intercept, b is the estimated change in demand per unit change in X_t, and ε_t is unexplained error. Multiple regression extends the model to several predictors.

Correlation does not establish causation. A regression forecast is only as credible as the assumptions that future explanatory variables are available, sufficiently stable and related to demand in the same way as in the estimation period. Multicollinearity, omitted variables, changing customer behaviour and endogeneity can undermine interpretation.

5. Forecast Accuracy, Bias and Uncertainty

Forecast accuracy should be measured on the decision-relevant horizon and aggregation level. No single metric is universally best.

Forecast error is commonly defined as:

Mean Absolute Deviation (MAD) is:

It is expressed in demand units and is easy to interpret. Mean Squared Error (MSE) is:

MSE penalises large errors more heavily. Root Mean Squared Error (RMSE) is:

Mean Absolute Percentage Error (MAPE) is:

MAPE becomes undefined when actual demand is zero and unstable when actual demand is close to zero. Weighted MAPE, sometimes called WAPE, is:

It gives greater influence to high-volume items but can conceal poor performance on low-volume products.

Worked accuracy example. Actual demand for three periods is 100, 120 and 80 units; forecasts are 90, 130 and 100 units. Errors are 10, −10 and −20 units. Therefore:

• MAD = (10 + 10 + 20) ÷ 3 = 13.33 units.
• MSE = (100 + 100 + 400) ÷ 3 = 200 units².
• RMSE = √200 ≈ 14.14 units.
• WAPE = 40 ÷ 300 × 100 = 13.33%.

Bias indicates systematic over- or underforecasting. Mean Error is:

In the example, ME = (10 − 10 − 20) ÷ 3 = −6.67 units, indicating underforecasting under the convention e_t = A_t − F_t. A tracking signal may be defined as:

where RSFE = ∑e_t is running sum of forecast errors. Organisations should define sign conventions consistently; otherwise bias reports can be misinterpreted.

Accuracy metrics must not be used to reward or punish teams without considering controllable causes. A planner may appear inaccurate because a promotion was changed after the forecast or because supply restrictions suppressed observed sales.

6. Managing Seasonality, Intermittent Demand and Uncertainty

Seasonality should be distinguished from trend and from one-off events. Useful practices include seasonal plots, year-on-year comparisons, decomposition, calendar adjustment and separate modelling of promotions. Forecasts should be reconciled across levels: SKU forecasts should generally aggregate to the product-family or market forecast unless a documented override exists.

Intermittent demand contains many zero periods and occasional non-zero demand. Standard MAPE and ordinary smoothing can perform poorly. Croston-type approaches separately estimate demand size and inter-arrival interval, although variants differ in bias and response. For critical items, planners should also consider service-level requirements, obsolescence risk and supplier lead time.

Uncertainty cannot be eliminated. It can be represented using prediction intervals, scenarios, probability distributions, safety stock, flexible capacity and postponement. A forecast is a central estimate; the operating plan should also identify a plausible range and triggers for action.

For independent random demand during a fixed lead time, a simplified safety-stock relationship is:

where SS is safety stock in units, z is the standard-normal service factor and σ_L is the standard deviation of demand during lead time. This formula assumes a specified demand distribution, reliable lead time and a service measure linked to the chosen z. If both demand and lead time vary, a more complete variance model is required; the simple formula should not be used uncritically.

Forecasting and inventory decisions are linked but not identical. Reducing forecast error may reduce inventory, but inventory also depends on lead time, service targets, supply reliability, order constraints and the cost of shortage.

7. Sales and Operations Planning

S&OP is a recurring, cross-functional process, usually monthly, that aligns a single set of assumptions for demand and supply with financial objectives. It operates at an aggregate level and bridges strategic intent with execution.

A robust S&OP cycle commonly includes the following stages:

1. Data gathering and review: collect sales, inventory, orders, forecasts, capacity, supply performance and financial data.
2. Demand review: create the unconstrained demand plan, incorporating statistical forecasts, sales intelligence, market events and customer commitments.
3. Supply review: test whether production, suppliers, labour, equipment, inventory and logistics can support demand.
4. Pre-S&OP reconciliation: evaluate alternatives, trade-offs, scenarios and financial consequences.
5. Executive S&OP: senior management resolves exceptions and approves a feasible plan.
6. Implementation and feedback: communicate decisions, monitor assumptions and update the next cycle.

Figure: The recurring S&OP cycle linking evidence, decisions and implementation.

The demand plan should distinguish baseline statistical demand, sales overrides, promotional demand, customer commitments and unconstrained market demand. The supply plan should show capacity limits, bottlenecks, supplier constraints, inventory policies, overtime, subcontracting and feasible alternatives. The executive decision is not necessarily to accept all demand; it may involve price changes, allocation, lead-time changes, capacity investment or deliberate backlog.

Aggregate planning choices

S&OP frequently evaluates three broad strategies:

• Level strategy: maintain a relatively stable production rate and use inventory or backlog to absorb demand variation.
• Chase strategy: vary production or workforce to follow demand.
• Hybrid strategy: combine stable core capacity with overtime, temporary labour, subcontracting, inventory or backlog.

The appropriate choice depends on labour regulations, hiring and release costs, inventory carrying costs, customer service, product shelf life, skill scarcity and strategic priorities. A mathematically cheap plan may be operationally or ethically unacceptable.

Financial and workforce reconciliation

Each S&OP alternative should be translated into revenue, material cost, labour cost, conversion cost, inventory investment, overtime, subcontracting, logistics cost and contribution margin. Finance should participate before executive approval rather than merely validating the plan afterward.

Workforce planning includes required hours, available regular hours, overtime, training, absenteeism, shifts and critical skills. If demand is stated in product units, conversion to labour requirements requires valid standards. For example, if 4,000 units require 0.25 labour-hours each, required labour is 1,000 hours. If one worker provides 160 productive hours per month, the theoretical requirement is 6.25 workers; actual staffing must account for absence, training, legal limits and skill mix.

8. A Practical S&OP Decision Framework

A disciplined meeting should focus on exceptions rather than reading every report. A useful decision record contains:

The organisation should maintain one agreed version of demand and supply data while allowing transparent scenario versions. Overrides should be logged with reasons and later evaluated. A forecast override that consistently improves accuracy is useful; one that introduces bias should be challenged.

9. Technology, AI and Governance

Enterprise resource planning systems, advanced planning and scheduling tools, cloud analytics and collaborative platforms can automate data extraction, hierarchy aggregation, exception alerts and scenario evaluation. Machine-learning models may detect nonlinear relationships, web signals, weather effects or customer behaviour changes. However, algorithmic complexity does not guarantee better decisions.

AI-supported forecasting requires:

• reliable and traceable source data;
• controls for leakage, where future information accidentally enters model training;
• validation using time-ordered holdout periods;
• monitoring for drift after market conditions change;
• explainable overrides and human accountability;
• protection of customer and employee data;
• controls against discriminatory allocation or service decisions.

Human judgment remains important where structural breaks, regulation, geopolitical events or strategic launches are not represented in historical data. The best practice is usually a governed combination of statistical baseline, documented commercial intelligence and scenario analysis.

10. Illustrative Case: Meridian Home Appliances

Background

Meridian Home Appliances is an illustrative manufacturer of countertop food processors. Demand is strongly seasonal, peaking before a major holiday. The company sells through retailers and online channels. Its monthly S&OP process has become contentious because Sales frequently raises the forecast while Operations reports insufficient assembly capacity.

Problem

The previous year’s sales data included a large retailer promotion that will not be repeated. In addition, two months contained stockouts. The demand analyst used raw sales as demand, producing a low baseline for the stockout months and an inflated seasonal peak from the promotion. The result was excessive finished-goods inventory after the peak and emergency overtime before it.

Analysis

The team first corrected the history. Stockout months were adjusted using confirmed backorders and retailer estimates. The one-time promotion was separated from baseline demand. Demand was forecast at product-family level and then disaggregated to models using current channel mix. A seasonal model was compared with a seasonal naïve benchmark using rolling-origin validation. WAPE and bias were reviewed by channel, not only in aggregate.

The supply review identified an assembly bottleneck and a six-week supplier lead time for a motor. Three options were costed:

1. build inventory early using regular shifts;
2. use limited overtime near the peak;
3. subcontract part of assembly and postpone final configuration.

Finance calculated the cash effect of each option. Sales agreed to a constrained-demand plan rather than promising all requested retailer volume. Marketing moved the promotion to a period with available capacity.

Approach and outcome

The executive team approved a hybrid plan: stable production in normal months, early motor purchasing, limited pre-build of common components, and controlled overtime during the peak. A weekly exception review was introduced for the six weeks before the season. Forecast overrides required a reason code and were assessed after the season.

The plan reduced peak overtime and prevented a repeat of post-season excess inventory. The most important improvement was not a particular algorithm. It was the separation of true demand from observed sales, the use of a common data set, and explicit reconciliation of commercial ambition with physical and financial constraints.

Learning points

• Sales data are not always equivalent to demand.
• Promotions must be modelled explicitly.
• Forecast accuracy should be evaluated at the level of the decision.
• S&OP is a decision process, not merely a meeting or software module.
• A feasible, financially evaluated plan is more valuable than an unconstrained forecast.

11. Challenges and Limitations

Forecasts are vulnerable to structural breaks, new-product uncertainty, weak data, changing prices, competitor actions and behavioural responses to the forecast itself. Aggregation can hide important local shortages; disaggregation can create noisy and unstable SKU forecasts. Forecast competition between departments may encourage sandbagging, optimism or unrecorded overrides.

There is also a danger of confusing precision with accuracy. A forecast stated as 10,437 units may appear sophisticated even when uncertainty is several thousand units. Managers should communicate ranges, confidence, assumptions and decision triggers. Forecasts should be judged by business outcomes and learning, not by numerical elegance alone.

Ethically, allocation decisions should not unfairly disadvantage smaller customers, vulnerable groups or regions without transparent justification. Employee scheduling should respect legal working-hour limits, safety and dignity. Customer data used for forecasting must comply with applicable privacy and data-protection requirements. Governance should specify who owns the forecast, who may override it, how conflicts are resolved and how decisions are audited.

Best-Practice Checklist

1. Define the decision, horizon and level of aggregation before choosing a method.
2. Establish a clean, documented demand history.
3. Use a simple benchmark before adopting a complex model.
4. Separate baseline demand, promotions, events and supply constraints.
5. Measure both accuracy and bias using appropriate metrics.
6. Validate models on periods not used for fitting.
7. Reconcile forecasts across product and organisational hierarchies.
8. Use scenarios and ranges to represent uncertainty.
9. Make S&OP cross-functional, evidence-based and decision-oriented.
10. Record assumptions, overrides, owners and review triggers.
11. Evaluate financial, capacity, inventory and workforce consequences together.
12. Monitor models and governance continuously after implementation.

Chapter Summary

Demand forecasting provides structured estimates of future customer requirements, but it does not remove uncertainty. Forecasts must be designed for a specific decision and may be qualitative, time-series, causal or hybrid. Naïve methods provide useful benchmarks; moving averages smooth demand; exponential smoothing updates forecasts from recent errors; trend and seasonal methods represent systematic movement; and causal models relate demand to explanatory variables.

Forecast error can be assessed using MAD, MSE, RMSE, MAPE, WAPE and bias measures, with careful attention to zero demand, aggregation and sign conventions. Seasonality, promotions, stockouts and intermittent demand require deliberate treatment. Safety stock and scenario planning help convert uncertainty into operational protection.

S&OP integrates demand, supply, finance, inventory, capacity and workforce decisions through a recurring cycle of data review, demand review, supply review, reconciliation and executive decision. Effective S&OP creates a feasible and financially coherent plan rather than accepting either unconstrained sales ambition or isolated operational limits. Technology and AI can improve speed and insight, but require high-quality data, validation, explainability, privacy and human accountability.

Key Terms

Aggregate forecast: A forecast prepared for a product family, market, region or other grouped level.

Bias: Systematic tendency to overforecast or underforecast.

Causal forecast: A forecast based on relationships between demand and explanatory variables.

Demand: Quantity customers are willing and able to purchase under specified conditions.

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

Forecast error: Difference between actual demand and forecast demand under a stated sign convention.

Holt’s method: Exponential-smoothing method that models level and trend.

Integrated business planning: Extended planning that aligns operational, financial and strategic decisions.

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

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

MAPE: Mean Absolute Percentage Error, an average percentage error measure that has problems with zero actuals.

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

Prediction interval: Range intended to contain a future observation with a stated level of confidence.

S&OP: Cross-functional process for reconciling demand, supply and financial plans.

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

Seasonality: Repeating demand pattern associated with a calendar or operating cycle.

Tracking signal: A ratio used to identify persistent forecast bias.

Unconstrained demand: Demand expected by the market before supply limitations are applied.

WAPE: Weighted Absolute Percentage Error, total absolute error divided by total actual demand.

Review Questions

5 Short Answer Questions

1. Distinguish demand from sales and explain why the distinction matters in forecasting.
2. State two conditions under which a moving average is useful.
3. What is the role of the smoothing constant in exponential smoothing?
4. Why can MAPE be unsuitable for intermittent demand?
5. Identify the major stages of a typical S&OP cycle.

5 Long Answer Questions

1. Compare qualitative, time-series and causal forecasting methods, including their strengths and limitations.
2. Explain additive and multiplicative representations of seasonality and discuss when each is appropriate.
3. Describe how forecast accuracy and bias should be measured and interpreted in a production organisation.
4. Explain how S&OP reconciles demand, supply, finance, inventory and workforce requirements.
5. Discuss the governance, ethical and data-quality risks associated with AI-supported forecasting.

3 Analytical / Application Questions

1. A product recorded demands of 600, 660, 720 and 750 units in four consecutive months. Calculate the three-month moving-average forecast for the next month and explain one limitation of the result.
2. A planner has a forecast of 2,000 units, but Sales expects a promotion to add 20% and Operations can supply only 2,250 units. Develop two S&OP options and discuss their service, cost and inventory implications.
3. A business has repeated stockouts during peak season and excess inventory afterward. Diagnose at least four possible forecasting or S&OP causes and propose a corrective process.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes demand from sales?

A. Demand is always lower than sales.

B. Sales are observed transactions, while demand may include unmet requirements.

C. Demand refers only to production capacity.

D. Sales are forecasts and demand is historical data.

Correct Answer: B. Sales are observed transactions, while demand may include unmet requirements.

Question 2

Using a three-period moving average, what is the forecast when the last three demands are 900, 1,000 and 1,100 units?

A. 950 units

B. 1,000 units

C. 1,050 units

D. 1,100 units

Correct Answer: B. 1,000 units

Question 3

In exponential smoothing, what does a larger value of α generally do?

A. Makes the forecast less responsive to recent changes

B. Eliminates all random variation

C. Makes the forecast more responsive to recent observations

D. Adds a seasonal cycle automatically

Correct Answer: C. Makes the forecast more responsive to recent observations

Question 4

Which metric is expressed in the same units as demand and averages absolute errors?

A. MAD

B. MAPE

C. MSE

D. WAPE

Correct Answer: A. MAD

Question 5

Why can MAPE be problematic when actual demand is zero?

A. It becomes negative

B. It cannot be calculated because division by zero occurs

C. It always equals 100%

D. It ignores forecast errors

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

Question 6

Which S&OP stage normally tests whether capacity, labour and suppliers can support the demand plan?

A. Demand review

B. Supply review

C. Data gathering only

D. Market research

Correct Answer: B. Supply review

Question 7

What is the primary purpose of safety stock?

A. To remove all forecast error

B. To increase the smoothing constant

C. To protect service against uncertainty

D. To replace capacity planning

Correct Answer: C. To protect service against uncertainty

Question 8

A forecast error is defined as e_t = A_t − F_t. If actual demand is 500 and the forecast is 540, what is the error?

A. −40 units

B. 40 units

C. 1,040 units

D. 0.08 units

Correct Answer: A. −40 units

Question 9

Which situation most strongly supports using a qualitative method?

A. A stable product with ten years of clean history

B. A new product with no historical demand

C. A repetitive component with reliable weekly data

D. A mature item with established seasonality

Correct Answer: B. A new product with no historical demand

Question 10

What is a central characteristic of executive S&OP?

A. It approves a cross-functional plan and resolves major trade-offs.

B. It produces only a statistical forecast.

C. It excludes finance from planning decisions.

D. It replaces all short-term scheduling.

Correct Answer: A. It approves a cross-functional plan and resolves major trade-offs.