4. Demand Forecasting and Sales and Operations Planning

Introduction and Context

Supply chains are coordinated around expectations of future demand. A manufacturer decides what materials to purchase, how much capacity to reserve, and when to produce only after forming some view of what customers are likely to require. Retailers determine inventory, transport requirements, and store replenishment in the same way. Demand forecasting therefore provides the informational foundation for many logistics and production decisions.

A forecast is not a promise about the future. It is a structured estimate based on historical evidence, market intelligence, managerial judgement, and assumptions about relevant conditions. Forecasts are inevitably imperfect because demand is affected by competitors, economic conditions, promotions, weather, technological change, regulations, and customer behaviour. The managerial task is not to eliminate uncertainty, but to measure it, respond to it, and design a supply system that remains effective when forecasts are wrong.

Sales and Operations Planning (S&OP), also called Integrated Business Planning in more advanced organisations, converts forecasts into coordinated decisions. It brings together sales, marketing, operations, procurement, finance, and senior management to reconcile expected demand with production, inventory, labour, supplier, and financial capacity. Forecasting answers, “What may customers require?” S&OP addresses, “How will the organisation respond, and what trade-offs are necessary?”

Learning Objectives

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

1. Explain the role of forecasting in logistics and supply chain decisions.

2. Distinguish qualitative, quantitative, causal, time-series, and collaborative forecasting approaches.

3. Select appropriate forecasting methods for different demand conditions.

4. Calculate and interpret forecast-error measures.

5. Explain demand variability, bias, safety stock, and the bullwhip effect.

6. Describe the stages and governance requirements of an S&OP process.

7. Evaluate capacity, inventory, sourcing, and demand-shaping alternatives.

8. Use digital analytics and AI responsibly in forecasting and planning.

9. Diagnose common forecasting and S&OP failures through applied analysis.

Demand Forecasting as a Managerial Process

Demand forecasting begins by defining the forecasting object. This includes the product or service, customer or market segment, geographic location, time horizon, and unit of measure. A forecast at the stock-keeping-unit and daily level may be useful for replenishment, whereas a product-family and monthly forecast is more suitable for aggregate capacity planning. Forecasting at an inappropriate level of detail often creates apparent precision without decision value.

The planning horizon is also important. Short-term forecasts support scheduling and replenishment; medium-term forecasts support workforce, inventory, procurement, and capacity decisions; long-term forecasts support facility location, technology investment, and network design. Forecast uncertainty generally increases with the length of the horizon, although aggregation across products or locations can reduce random variation.

Demand should be separated conceptually from sales. Sales are observed transactions and may be constrained by stockouts, delivery failures, credit restrictions, or limited selling effort. Demand is the quantity customers would have purchased under unconstrained availability. If a product was unavailable for two weeks, recorded sales understate underlying demand. Planners should therefore identify lost sales, backorders, substitutions, and promotional effects before modelling future demand.

Important demand patterns include:

• Level: the typical magnitude around which demand varies.
• Trend: a persistent upward or downward movement.
• Seasonality: a predictable pattern linked to calendar, climate, or institutional cycles.
• Cycles: longer-term movements associated with economic or industry conditions.
• Random variation: irregular fluctuations that cannot be predicted reliably.
• Intermittency: many zero-demand periods interrupted by occasional transactions.

Forecasting is most effective when the pattern, data quality, decision purpose, and cost of error are considered together.

Forecasting Approaches and Models

Qualitative Forecasting

Qualitative methods are valuable when historical data are limited, the product is new, or structural change makes past relationships unreliable. They rely on expert knowledge, market research, customer consultation, sales-force estimates, and structured managerial judgement.

The Delphi method obtains opinions from a panel of experts through several anonymous rounds. A facilitator summarises responses and seeks revised estimates until a reasoned consensus or range is developed. Its anonymity reduces the influence of hierarchy, although results remain dependent on the expertise and assumptions of the panel. Sales-force composite forecasting uses estimates from personnel close to customers. It may provide valuable local information but can be distorted by incentives, optimism, or conservative targets.

Market surveys and test marketing gather information directly from potential customers. Scenario analysis is particularly useful for strategic planning because it explores several plausible futures rather than producing one supposedly precise number. Qualitative forecasts should be documented: the assumptions, evidence, participants, and confidence level should be visible to later reviewers.

Time-Series Forecasting

Time-series methods use the historical sequence of demand. A naïve forecast sets the next period equal to the most recent actual demand. Although simple, it is a useful benchmark. A moving average calculates the mean of demand over a selected number of recent periods:

Moving average = (sum of demand in the selected periods) / (number of periods).

A short window responds rapidly to change but is more sensitive to random variation. A long window is smoother but may react slowly to a trend. A weighted moving average assigns greater importance to selected periods, commonly the most recent observations.

Exponential smoothing updates the previous forecast using the latest error:

New forecast = Previous forecast + α(Actual demand − Previous forecast),

where α, the smoothing constant, lies between 0 and 1. A high α responds quickly to new information; a low α produces a more stable forecast. Trend-adjusted methods, such as Holt’s method, estimate both level and trend. Seasonal models add or multiply seasonal factors to reflect recurring patterns. The method should match the data rather than being selected solely because it is mathematically sophisticated.

Intermittent demand requires special treatment. Standard averages may generate excessive inventory or unstable replenishment signals when demand contains many zeros. Croston-type methods and related approaches estimate demand size and the interval between demand events separately. Even then, managerial judgement is needed because a zero may mean either genuine absence of demand or an unavailable item.

Causal and Econometric Forecasting

Causal models explain demand using drivers such as price, income, advertising expenditure, temperature, construction activity, or competitor actions. Regression analysis can estimate the relationship between demand and these variables. A model may be written as:

Demand = a + b1(Price) + b2(Advertising) + b3(Income) + error.

Causal models are useful for scenario analysis, price decisions, and promotion planning, but correlation does not establish causation. Relationships may change, variables may be strongly related to one another, and future values of the explanatory variables must themselves be estimated.

Collaborative Forecasting

Collaborative Planning, Forecasting and Replenishment (CPFR) shares information among supply-chain partners. Manufacturers, distributors, and retailers may jointly review sales, inventory, promotions, and market events. Collaboration can reduce duplicated forecasts and improve visibility, but it requires common definitions, data governance, trust, and agreement about ownership of decisions. Sharing a forecast without sharing assumptions rarely produces genuine collaboration.

Forecast Error and Uncertainty

A forecast should be evaluated by comparing forecast values with actual outcomes. The error may be defined as:

Error = Actual demand − Forecast demand.

Mean Absolute Deviation (MAD) is the average of the absolute errors and expresses typical error in the original unit. Mean Squared Error gives greater weight to large errors. Root Mean Squared Error returns the result to the original unit and is useful when large misses are particularly costly. Mean Absolute Percentage Error (MAPE) expresses error as a percentage, but it is unsuitable when actual demand is zero or very small. Weighted MAPE can prevent low-volume items from dominating the evaluation.

Forecast bias measures systematic over- or under-forecasting. A positive cumulative error, using the definition above, indicates that actual demand has generally exceeded the forecast. Tracking Signal is commonly calculated as cumulative forecast error divided by MAD. A persistent signal beyond an organisation’s control limits suggests bias or a change in the demand process. The exact control limits should be defined by the organisation and used as prompts for investigation, not as automatic proof that a model is invalid.

Accuracy should be assessed at the level at which decisions are made. An aggregate forecast may appear accurate because positive and negative errors cancel, while individual products remain seriously misaligned. Forecast value added analysis compares a baseline statistical forecast with subsequent adjustments. If managerial overrides do not improve accuracy or reduce relevant business risk, the adjustment process should be redesigned.

Forecast error has economic consequences. Under-forecasting can cause stockouts, lost sales, expedited freight, overtime, and customer dissatisfaction. Over-forecasting can cause excess inventory, obsolescence, discounting, and working-capital pressure. The preferred forecast is therefore not always the one with the lowest statistical error; it is the one that supports the best decision under the organisation’s cost and service priorities.

Managing Variability in the Supply Chain

Demand variability is only one source of uncertainty. Lead-time variability, supplier reliability, production yield, transportation disruptions, and data delays also affect service. Safety stock protects against uncertainty, but it does not compensate for poor planning indefinitely. Its level should reflect desired service, demand variability, lead time, review frequency, and replenishment policy.

Aggregation often reduces relative variability. Forecasting a product family may be more reliable than forecasting each individual item, and regional demand may be more stable than demand at one store. However, aggregate plans must eventually be translated into feasible product and location decisions. Postponement, flexible manufacturing, multiple sourcing, capacity buffers, and postponing final configuration can also reduce the cost of forecast error.

The bullwhip effect occurs when order variability becomes greater as one moves upstream in a supply chain. It can be amplified by forecast updating, order batching, price promotions, shortage gaming, long lead times, and poor information sharing. More frequent replenishment, stable pricing, smaller order quantities, shared demand data, and coordinated planning can reduce amplification. A downstream order should not be treated automatically as independent market demand.

Sales and Operations Planning: From Forecast to Decision

S&OP is a recurring, cross-functional process that aligns demand, supply, and financial plans at an aggregate level. It is not simply a forecasting meeting and not merely a production schedule. Its value lies in making trade-offs explicit and securing organisational commitment to one feasible plan.

A typical monthly cycle contains the following stages.

Data gathering and analytics. Recent sales, inventory, orders, forecasts, capacity, supplier performance, promotions, and major exceptions are collected. The previous plan is evaluated, and assumptions are documented.

Demand review. Sales and marketing prepare an unconstrained demand plan. New products, promotions, customer intelligence, lost sales, competitive actions, and market changes are considered. The team distinguishes genuine demand changes from temporary noise.

Supply review. Operations, procurement, logistics, and engineering test whether the demand plan can be supported. They examine capacity, labour, materials, suppliers, inventory, lead times, transportation, and bottlenecks. Alternative supply responses are quantified.

Integrated reconciliation. Demand and supply alternatives are compared with financial objectives. Options may include overtime, subcontracting, inventory build, product substitution, allocation rules, revised promotions, price changes, or renegotiated customer dates. Each option should show service, cost, working capital, risk, and implementation implications.

Executive S&OP. Senior managers resolve issues that cannot be settled at functional level. They approve a single operating plan, define exceptions, assign accountability, and communicate decisions to detailed planning processes.

S&OP is normally conducted at product-family, market, or business-unit level. The master production schedule and material requirements planning later translate the agreed plan into item-level and time-phased requirements. Thus, S&OP provides direction and constraints rather than replacing detailed operational planning.

Capacity and Demand Balancing Choices

When expected demand exceeds available capacity, managers can increase capacity through overtime, additional shifts, temporary labour, subcontracting, equipment investment, or supplier development. They can also influence demand through pricing, promotion timing, allocation, product substitution, delivery-date negotiation, or prioritisation of strategic customers. When demand is weak, the organisation may reduce output, build selected inventory, schedule maintenance, promote products, or reassign labour.

These choices involve trade-offs. Building inventory can protect future service but increases holding cost and obsolescence risk. Subcontracting provides flexibility but may reduce control and expose proprietary knowledge. Overtime responds quickly but may cause fatigue and quality problems. Price incentives can shift demand but may train customers to wait for discounts. A sound S&OP process evaluates alternatives across the entire supply chain rather than optimising one function.

Digital Tools, Analytics, and Artificial Intelligence

Enterprise resource planning systems provide transaction data for orders, inventory, procurement, and production. Advanced planning systems can combine forecasts with constraints and simulate alternative plans. Dashboards support exception management by highlighting unusual demand, declining forecast accuracy, supplier delays, and capacity overloads.

Machine-learning models can detect nonlinear relationships, identify demand signals from large datasets, and improve forecasts for selected products. Possible inputs include web searches, point-of-sale data, weather, promotions, and lead-time information. However, more data and more complex algorithms do not automatically create better decisions. Problems of data leakage, biased historical data, changing customer behaviour, poor explainability, cybersecurity, and inappropriate automation must be managed. Human planners remain responsible for validating assumptions, interpreting exceptional events, and approving actions.

AI governance should include clear data ownership, access controls, model monitoring, version records, performance thresholds, and escalation procedures. Forecast outputs should be presented with uncertainty ranges where possible. A single point estimate can encourage false confidence and conceal the need for contingency planning.

Illustrative Case: Meridian Appliances

Background

Meridian Appliances, a fictional mid-sized manufacturer, produces three families of domestic water heaters. Retailer orders had become highly variable, inventory had increased, and production frequently required emergency overtime. Sales blamed production shortages, while operations argued that sales promotions were not communicated early enough.

Situation and Analysis

An initial review found that the organisation used retailer orders as its forecast. Retailers often placed large orders during monthly promotions and little or nothing in the following month. The order pattern therefore exaggerated end-consumer demand. The company also evaluated accuracy only at total-company level, where errors cancelled each other. At product level, the premium model was consistently over-forecast and the standard model under-forecast.

The planning team created a baseline forecast using seasonal exponential smoothing and separately recorded promotion uplift. It measured MAD and bias by product family and compared the forecast with point-of-sale information supplied by two major retailers. A capacity review showed that the painting process, rather than final assembly, was the principal bottleneck.

Approach and Solution

Meridian introduced a monthly S&OP cycle. Sales submitted a market-informed demand plan, including promotion dates and expected uplift. Operations prepared constrained supply scenarios: normal production, overtime at the bottleneck, and limited subcontracting. Finance evaluated inventory and margin implications. The executive meeting approved a plan that shifted some promotions away from peak months, reserved bottleneck capacity for the standard model, and established allocation rules for shortages.

The company also reduced order batching by moving selected retailers toward more frequent replenishment. A dashboard monitored forecast bias, stockouts, inventory age, and capacity utilisation. Overrides to the statistical forecast required a written assumption and were reviewed after the event.

Outcome

The principal improvement was not a claim of perfect forecasting. Rather, the organisation gained a shared view of demand, identified the true bottleneck, and made trade-offs before problems became urgent. Inventory and emergency overtime were reduced in the illustrative scenario, while service became more predictable. The case demonstrates that better coordination and transparency can create value even when uncertainty remains.

Learning Points

1. Orders are not always equivalent to final demand.

2. Forecast evaluation should be performed at the level relevant to decisions.

3. Promotions must be integrated into both demand and capacity planning.

4. S&OP creates value by reconciling functional perspectives.

5. A forecast is useful only when connected to feasible actions.

Challenges, Governance, and Professional Practice

Common forecasting mistakes include using poor-quality data, ignoring stockouts, selecting a model without examining demand patterns, overreacting to one unusual observation, and allowing unchallenged managerial optimism. Another error is confusing forecast accuracy with business performance. A highly accurate forecast may still support a poor decision if capacity, inventory, or customer profitability is ignored.

S&OP can fail when meetings are dominated by functional conflict, senior leaders do not make decisions, data definitions differ, plans are changed informally, or participants focus on detailed exceptions rather than aggregate trade-offs. Effective governance defines the calendar, decision rights, escalation thresholds, performance measures, and ownership of assumptions.

Ethically, forecasts must not be manipulated to create favourable performance reports, justify inappropriate inventory, or shift blame to another function. Sales incentives should not encourage deliberate under-forecasting or order loading. Customer and point-of-sale data must be handled according to contractual, privacy, and cybersecurity requirements. Planners should communicate uncertainty honestly, especially when forecasts affect employment, supplier commitments, public services, or critical products.

Current Developments and Best Practices

Contemporary planning increasingly combines near-real-time demand sensing with traditional monthly planning. Digital twins and scenario engines allow managers to test disruptions, capacity changes, and alternative sourcing before implementation. Collaborative platforms connect suppliers, manufacturers, and customers, while control towers provide cross-network visibility.

Nevertheless, technology should strengthen planning discipline rather than replace it. Strong practice includes the following:

• define the decision, horizon, product hierarchy, and forecast owner;
• clean and reconcile data before modelling;
• use a simple benchmark before adopting a complex model;
• separate baseline demand from promotion and event effects;
• measure bias as well as accuracy;
• express uncertainty and prepare contingency actions;
• review overrides and learn from forecast errors;
• link demand, supply, financial, and service measures;
• use S&OP to make decisions, not merely exchange reports; and
• maintain clear accountability for assumptions and execution.

Chapter Summary

Demand forecasting estimates future customer requirements so that organisations can plan inventory, capacity, procurement, production, and logistics. Qualitative methods are valuable where data are limited or structural change is substantial; time-series methods exploit historical patterns; causal models explain demand through external drivers; and collaborative approaches improve visibility across supply-chain partners. Forecast performance must be assessed through error, bias, and business consequences rather than a single accuracy percentage.

Demand variability interacts with lead-time and supply uncertainty. Safety stock, aggregation, postponement, flexible capacity, and information sharing help manage these conditions, while poor coordination can create the bullwhip effect. S&OP integrates demand, supply, and finance through a recurring process of data review, demand review, supply review, reconciliation, and executive decision-making. Digital systems, advanced analytics, and AI can enhance forecasting, but governance, explainability, data quality, and human accountability remain essential.

Key Terms

Forecast: A structured estimate of future demand or another future condition.

Demand: The quantity customers would purchase under specified conditions and availability.

Time series: A sequence of observations arranged in time order.

Seasonality: A recurring demand pattern associated with calendar or environmental cycles.

Trend: A sustained upward or downward movement in demand.

Moving average: A forecast based on the average of recent observations.

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

Causal forecasting: Forecasting based on explanatory variables that influence demand.

Qualitative forecasting: Forecasting based primarily on expert judgement and market information.

CPFR: Collaborative Planning, Forecasting and Replenishment among supply-chain partners.

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

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

Forecast bias: A persistent tendency to over-forecast or under-forecast.

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

Bullwhip effect: Amplification of demand variability as information moves upstream.

S&OP: A cross-functional process aligning demand, supply, and financial plans.

Integrated Business Planning: An extended S&OP approach incorporating broader strategic and financial integration.

Demand sensing: Use of recent signals to update short-term demand expectations.

Postponement: Delaying final product configuration or commitment until demand is clearer.

Forecast value added: Evaluation of whether each forecasting step improves the baseline forecast.

Review Questions

Short Answer Questions

1. Why is recorded sales data not always equivalent to true demand?

2. Distinguish between seasonality and trend.

3. What does forecast bias indicate?

4. Why can MAPE be problematic for intermittent demand?

5. What is the purpose of executive S&OP?

Long Answer Questions

1. Compare qualitative, time-series, causal, and collaborative forecasting approaches, including situations in which each is appropriate.

2. Explain how forecast error should be measured and interpreted in a supply-chain organisation.

3. Describe the stages of a monthly S&OP cycle and the responsibilities of major functions.

4. Discuss the causes of the bullwhip effect and methods for reducing it.

5. Evaluate the role of AI and advanced analytics in demand forecasting, including governance risks.

Analytical / Application Questions

1. A product has demand of 100, 120, and 140 units in the last three months. Calculate the three-period moving-average forecast for the next month and explain one limitation of this method.

2. A company repeatedly over-forecasts a premium product while under-forecasting a standard product. Diagnose possible causes and recommend changes to its forecasting and S&OP process.

3. Demand is expected to exceed a bottleneck process by 15 percent during a promotion. Compare overtime, subcontracting, promotion rescheduling, and customer allocation as possible responses, considering cost, service, risk, and implementation speed.

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes demand from sales?

A. Demand is always lower than sales.

B. Sales represent transactions, while demand may include purchases prevented by stockouts or other constraints.

C. Demand is measured only through financial records.

D. Sales include only forecast quantities.

Correct Answer: B. Sales represent transactions, while demand may include purchases prevented by stockouts or other constraints.

Question 2

A forecasting method gives greater weight to recent observations and updates the forecast using the latest error. Which method is this?

A. Exponential smoothing

B. Delphi analysis

C. Simple regression

D. Market surveying

Correct Answer: A. Exponential smoothing

Question 3

Why should MAPE be used cautiously for intermittent demand?

A. It cannot compare percentages.

B. It gives no information about actual demand.

C. Zero or very small actual values can make the percentage error undefined or misleading.

D. It always overstates forecast accuracy.

Correct Answer: C. Zero or very small actual values can make the percentage error undefined or misleading.

Question 4

A persistent positive cumulative error, where error is defined as actual demand minus forecast demand, most likely indicates:

A. Systematic under-forecasting

B. Systematic over-forecasting

C. Perfect forecast neutrality

D. An absence of demand variability

Correct Answer: A. Systematic under-forecasting

Question 5

Which activity belongs primarily to the supply review in S&OP?

A. Estimating competitor advertising alone

B. Testing whether labour, materials, suppliers, and capacity can support the demand plan

C. Replacing all statistical forecasts with sales opinions

D. Approving individual customer invoices

Correct Answer: B. Testing whether labour, materials, suppliers, and capacity can support the demand plan

Question 6

Retailer orders fluctuate sharply because customers place large orders during promotions and none immediately afterward. This situation can contribute most directly to:

A. The bullwhip effect

B. A reduction in all lead times

C. Perfect demand visibility

D. Elimination of safety stock

Correct Answer: A. The bullwhip effect

Question 7

Which response best reflects sound S&OP practice when demand exceeds a bottleneck?

A. Increase every product forecast without checking capacity.

B. Select among quantified alternatives such as overtime, subcontracting, allocation, or promotion rescheduling.

C. Allow each function to create an independent plan.

D. Ignore financial and service consequences.

Correct Answer: B. Select among quantified alternatives such as overtime, subcontracting, allocation, or promotion rescheduling.

Question 8

What is the principal purpose of forecast value added analysis?

A. To determine whether successive forecasting steps improve a baseline forecast

B. To calculate warehouse rental cost

C. To set the legal price of a product

D. To replace all managerial judgement automatically

Correct Answer: A. To determine whether successive forecasting steps improve a baseline forecast

Question 9

Which is an appropriate governance practice for AI-supported forecasting?

A. Accepting model outputs without monitoring

B. Removing human review from exceptional events

C. Maintaining data ownership, model monitoring, access controls, and escalation procedures

D. Using only the most complex available algorithm

Correct Answer: C. Maintaining data ownership, model monitoring, access controls, and escalation procedures

Question 10

Why is S&OP conducted commonly at product-family or business-unit level rather than only at SKU level?

A. Aggregate planning supports cross-functional capacity and financial trade-offs before detailed scheduling.

B. SKU data are never useful.

C. Product families eliminate all uncertainty.

D. Detailed planning is unrelated to S&OP.

Correct Answer: A. Aggregate planning supports cross-functional capacity and financial trade-offs before detailed scheduling.