Demand Forecasting and Integrated Planning

Introduction

Demand forecasting is the disciplined estimation of future customer requirements. In logistics and supply chain management, it is not an exercise in predicting the future with certainty; it is a process for reducing uncertainty sufficiently to make better decisions about procurement, inventory, capacity, transport, labour and customer service. Forecasts connect market information with operational plans. A poor forecast can create excess inventory, obsolescence and working-capital pressure, while an optimistic forecast can produce stockouts, expediting and lost sales.

Modern planning extends beyond a statistical forecast. Organisations combine historical demand, market intelligence, promotional information, operational constraints and partner data through sales and operations planning (S&OP), integrated business planning (IBP), demand sensing and collaborative planning, forecasting and replenishment (CPFR). The central managerial challenge is to balance responsiveness against cost and stability.

Learning Objectives

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

1. Distinguish qualitative, quantitative and causal forecasting approaches.
2. Select suitable forecasting methods and recognise their assumptions.
3. Calculate and interpret forecast-error measures.
4. Explain demand sensing, aggregate planning, S&OP and IBP.
5. Analyse capacity, inventory and workforce trade-offs.
6. Describe collaborative planning and CPFR.
7. Design governance practices that improve forecast quality and planning integration.

1. The Role and Nature of Forecasting

A forecast is an estimate of a future variable, usually demand over a specified time horizon, product level and location. The forecast horizon should match the decision: a daily transport plan may require a short horizon, whereas a plant investment decision may require several years. Forecasts should therefore be defined by four dimensions: what is being forecast, where, when and in what unit.

Demand may be classified as:

• Independent demand: demand originating outside the organisation, such as customer purchases.
• Dependent demand: demand derived from another item, such as components required for a finished product.
• Stable demand: relatively consistent level and variation.
• Seasonal demand: recurring variation linked to calendar, climate or social cycles.
• Intermittent demand: many zero-demand periods interrupted by non-zero transactions.
• Lumpy demand: intermittent demand with highly variable order quantities.

A forecast is not the same as a target, budget or sales commitment. A forecast describes what is likely to happen; a target describes what management wants to happen; a budget authorises resources. Confusing these purposes can cause biased forecasts and conceal risk.

Forecasting is subject to irreducible uncertainty. A useful planning system therefore produces a forecast, a confidence or risk assessment, and a response plan. Forecast accuracy should be evaluated at the level at which decisions are made. Errors may cancel when aggregated across products, but they may remain severe for individual stock-keeping units (SKUs).

2. Forecasting Approaches

2.1 Qualitative forecasting

Qualitative methods are appropriate when historical data are limited, the product is new, structural change is expected, or expert knowledge contains information not yet represented in the data. Major techniques include:

• Executive opinion: senior managers develop a forecast through discussion. It is fast but vulnerable to hierarchy and optimism.
• Sales-force composite: sales representatives estimate customer or territory demand. It captures local information but may be distorted by incentives.
• Market research: surveys, interviews, conjoint analysis and test marketing assess potential demand. Stated intentions may not equal actual purchases.
• Delphi method: experts respond anonymously in successive rounds, receiving an aggregated summary after each round. Anonymity reduces domination by senior participants.
• Analogy: demand is inferred from a comparable product, market or historical launch.

Qualitative judgements should be documented: source, assumptions, confidence, expected timing and evidence. Forecast overrides should be traceable rather than treated as unquestionable expertise.

2.2 Quantitative time-series forecasting

Time-series methods assume that useful information about future demand is contained in the pattern of past observations. Components often considered are level, trend, seasonality and irregular variation. A decomposition may be additive when seasonal variation is approximately constant in absolute units, or multiplicative when seasonal variation grows with the level.

#### Naïve method

The simplest forecast for the next period equals the latest actual observation:

Here, F_t+1 is the forecast for period t+1, and A_t is actual demand in period t. It is a useful benchmark and can perform well for stable, random-walk-like demand.

#### Moving average

For a simple moving average of k periods:

All observations receive equal weight. Increasing k smooths noise but makes the forecast less responsive to a genuine change in level.

#### Weighted moving average

The weights w_i are non-negative and sum to one. Larger weights can be assigned to recent observations when responsiveness is important.

#### Simple exponential smoothing

α is the smoothing constant. A high value responds rapidly to new observations; a low value produces greater smoothing. Simple exponential smoothing is suitable where demand has no systematic trend or seasonality. Its forecast error is e_t = A_t − F_t.

#### Trend-adjusted smoothing

When a trend exists, level and trend can be updated separately. One common Holt formulation is:

L_t is the estimated level, T_t the trend, m the number of periods ahead, and β the trend-smoothing constant. Initial values and parameter selection affect results.

#### Seasonal forecasting

A seasonal index compares a period’s typical demand with the overall level. Under a multiplicative model:

Seasonal indices should normally average to one across a complete seasonal cycle. They require enough cycles to distinguish seasonality from unusual events.

#### Causal and regression models

Causal forecasting relates demand to explanatory variables such as price, income, advertising, weather or economic activity. A linear model may be written as:

Here, y_t is demand, x_jt are predictors, β_j are estimated coefficients and ε_t is unexplained error. The model is useful only when relationships are theoretically plausible, data are sufficiently reliable, and future predictor values are available. Correlation does not establish causation. Regression assumptions such as appropriate functional form, independent or suitably modelled errors, and stable relationships must be assessed.

Intermittent demand requires special treatment. A conventional average can generate frequent false signals. Croston-type methods separately estimate demand size and the interval between non-zero demands; variants are useful but can be biased and do not automatically solve inventory-service decisions. The forecasting method should reflect the transaction process, not merely the visual appearance of a chart.

3. Forecast Evaluation and Error Governance

No single error measure is universally best. The choice depends on scale, aggregation and managerial purpose.

The definitions assume n comparable observations and consistent units. For MAPE, A_t must not be zero. A forecast should be evaluated against a naïve benchmark and over a rolling, decision-relevant window. Accuracy and bias are different: errors can have a low average absolute size while consistently being positive or negative.

Worked example. Suppose actual demand over four weeks is 100, 120, 80 and 100 units, while forecasts are 90, 130, 100 and 90 units. Errors e_t=A_t−F_t are 10, −10, −20 and 10. Thus:

• ME=(10−10−20+10)÷4=−2.5 units, indicating slight over-forecasting.
• MAE=(10+10+20+10)÷4=12.5 units.
• MSE=(100+100+400+100)÷4=175 units².
• RMSE=√175≈13.23 units.

MAPE is 100×(0.10+0.0833+0.25+0.10)÷4≈13.33%. The measures do not say whether the forecast is acceptable; that judgement depends on service requirements, cost and benchmark performance.

Forecast evaluation should also examine forecast value added: whether a human override or a new model improves decisions relative to the existing process. Excessive overrides, especially those that do not improve accuracy or bias, indicate weak governance.

4. Demand Sensing and Digital Forecasting

Demand sensing uses near-real-time signals to update a near-term forecast. Signals may include point-of-sale data, orders, shipment data, inventory positions, promotions, web traffic, weather, mobility and market events. It differs from conventional forecasting mainly in data latency, granularity and response speed. It does not eliminate uncertainty or replace medium- and long-term planning.

A robust architecture separates:

1. Data layer: validated internal and external data, product and location hierarchies, timestamps and master data.
2. Analytical layer: statistical models, causal variables, anomaly detection and scenario analysis.
3. Decision layer: replenishment, allocation, capacity and exception management.
4. Governance layer: ownership, access control, model monitoring, audit trails and override rules.

Machine-learning models can identify nonlinear patterns and interactions, but they may overfit, reproduce historical bias or fail under unprecedented conditions. Explainability, drift monitoring and human review are particularly important for high-value or safety-critical decisions. Data privacy, commercial confidentiality and cybersecurity must be protected when customer or partner data are shared.

5. Aggregate Planning and Capacity Balancing

Aggregate planning converts demand forecasts into a medium-term plan for product families, workforce, inventory, subcontracting and capacity. It usually covers several months to roughly two years, depending on the industry. The plan balances cost and service rather than scheduling every individual order.

Typical strategies are:

• Level strategy: maintain a stable workforce and production rate, using inventory or backlog to absorb demand variation.
• Chase strategy: vary production or workforce to follow demand.
• Mixed strategy: combine regular time, overtime, inventory, subcontracting and controlled backlog.

A simplified aggregate-planning objective is:

Subject to material balance:

Here, I_t is ending inventory, P_t regular production, S_t subcontracted or additional supply, D_t demand and B_t backlog adjustment, with consistent units. Capacity constraints may include:

The equations are simplified: real models may include setup times, multiple resources, perishability, minimum batch sizes, labour laws and transportation constraints. A mathematically cheapest plan may be operationally infeasible or commercially unacceptable.

Capacity balancing compares required capacity with available capacity. Managers can alter demand through pricing, promotion timing and allocation, or alter supply through shifts, overtime, subcontracting, inventory prebuild and postponement. The correct response depends on cost, lead time, quality, resilience and strategic capability.

6. S&OP and Integrated Business Planning

S&OP is a cross-functional monthly or periodic process that reconciles a demand plan with a supply plan and presents alternatives to management. A typical cycle includes data preparation, statistical baseline, demand review, supply review, reconciliation and executive decisions. Participants commonly include sales, marketing, operations, procurement, finance and logistics.

IBP extends S&OP by integrating financial plans, strategic objectives, product portfolios, risk, capital and scenarios more explicitly. The distinction is not universal across organisations; some use the terms interchangeably. The essential principle is one integrated set of assumptions and decisions rather than disconnected departmental forecasts.

A mature process has:

• one agreed planning hierarchy and calendar;
• explicit assumptions and scenario versions;
• documented ownership of forecast, supply and financial consequences;
• exception-based management rather than meeting every SKU equally;
• decisions recorded with owner and due date;
• feedback from execution into the next cycle.

Consensus does not mean averaging every opinion. It means selecting a decision after transparent challenge of assumptions, constraints and trade-offs. Finance should translate operational plans into revenue, margin, cash and working-capital consequences.

7. Collaborative Planning, Forecasting and Replenishment

CPFR is a structured approach in which supply-chain partners share information, develop aligned forecasts and coordinate replenishment. A simplified cycle is:

Figure: A simplified collaborative planning, forecasting and replenishment cycle.

CPFR can reduce duplicated safety stock and improve promotional readiness, but collaboration creates costs and risks. Partners must agree on data definitions, forecast ownership, confidentiality, dispute resolution, service measures and the treatment of promotional uplift. Information sharing should comply with competition law; exchanging sensitive information about competitors or future pricing can create legal risk. Access should be limited to legitimate planning purposes.

8. Inventory, Service and Forecast Uncertainty

Forecast uncertainty affects safety stock and replenishment. If demand during lead time is approximately normally distributed and lead time is constant, a simplified reorder point is:

ROP is reorder point, μ_LT mean demand during lead time, σ_LT standard deviation of lead-time demand, and z the standard-normal factor associated with the desired cycle-service probability. This formula assumes an appropriate distribution, reliable estimates, independent or properly aggregated variation, and no major structural break. If both period demand and lead time vary independently:

where L is mean lead time, σ_D standard deviation of demand per period, μ_D mean demand per period and σ_L standard deviation of lead time. The formula is an approximation and omits a cross-variation term under the stated independence assumption. Service targets should be differentiated by item criticality and economics rather than maximised indiscriminately.

9. Illustrative Case: Meridian Home Appliances

Background. Meridian, a fictional regional appliance distributor, experienced frequent stockouts of air conditioners during hot periods and excess stock after the season. Sales teams issued separate forecasts, suppliers received irregular orders and finance could not reconcile the inventory plan with cash targets.

Problem. Historical monthly demand was seasonal, promotions caused spikes, and retailer point-of-sale data arrived with a two-week delay. A single annual growth percentage was being applied to all SKUs. Forecast overrides were undocumented.

Analysis. The supply-chain team segmented products by volume, margin, criticality and demand intermittency. It established a baseline seasonal model for high-volume items, used intermittent-demand methods for service parts, and introduced a naïve benchmark. Forecast error was measured with WAPE at family level, MAE at SKU level and bias by channel. The team found that promotions were recorded as sales events but not as planned causal drivers, causing the model to treat temporary peaks as permanent growth.

Approach. Meridian created a monthly S&OP cycle. Marketing supplied a promotion calendar; sales submitted evidence-based overrides; operations checked supplier and warehouse capacity; finance translated scenarios into cash and margin. Retailers shared weekly point-of-sale data under a data-sharing agreement. A CPFR pilot aligned promotional forecasts and replenishment for two major retailers. Near-term demand sensing updated the next eight weeks, while the seasonal model governed the medium-term plan.

Outcome. In the illustrative scenario, stockouts declined, obsolete seasonal inventory fell and emergency freight became less frequent. The main improvement did not come from a sophisticated algorithm alone. It came from better data, explicit promotion assumptions, shared accountability and capacity decisions made before demand materialised.

Learning points. Forecasting must be segmented; partner data are valuable only when definitions and incentives are aligned; and S&OP is a decision process, not merely a forecast meeting.

10. Challenges, Ethics and Best Practices

Important challenges include structural breaks, supply-constrained sales data, the bullwhip effect, forecast gaming, missing data, inconsistent product hierarchies and excessive model complexity. When supply is constrained, observed sales underestimate true demand; lost-sales estimates should be separated from actual sales wherever possible.

Ethical governance requires avoiding deliberate inflation of forecasts to secure capacity or budgets, protecting customer data, explaining automated decisions and ensuring that service reductions do not unfairly disadvantage vulnerable communities. AI systems should be monitored for bias, drift and false precision. A forecast expressed as a single number can create unjustified confidence; ranges, scenarios and assumptions are more honest.

Best practice is to:

1. Define the decision, horizon, unit and forecast owner.
2. Establish clean master data and a common calendar.
3. Use a transparent statistical baseline before overrides.
4. Segment by demand pattern, value and criticality.
5. Measure both accuracy and bias against a benchmark.
6. Record reasons and results of overrides.
7. Integrate demand, supply, finance and capacity in S&OP/IBP.
8. Use collaborative data selectively, with governance agreements.
9. Plan scenarios for promotions, disruption and structural change.
10. Review whether forecast improvements actually improve service, cost and cash.

Chapter Summary

Demand forecasting estimates future requirements to support inventory, capacity, procurement, transport and financial decisions. Qualitative methods are valuable when data are scarce or structural change is expected; quantitative time-series and causal methods are powerful when data and assumptions are appropriate. Naïve, moving-average, exponential-smoothing, seasonal and regression models each have distinct applicability conditions.

Forecast quality must be assessed through measures such as bias, MAE, RMSE, MAPE and WAPE, with care for zero demand, scale and aggregation. Demand sensing uses more timely signals but does not replace strategic planning. Aggregate planning balances demand with workforce, production, inventory, overtime and subcontracting decisions. S&OP reconciles functional plans, while IBP connects them to finance, strategy and risk. CPFR extends integration across organisational boundaries. Effective practice combines analytics with governance, collaboration, ethical data use and explicit capacity and scenario decisions.

Key Terms

Forecast: An estimate of a future variable over a defined horizon.

Independent demand: Demand originating from external customers.

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

Qualitative forecasting: Forecasting based primarily on judgement, expertise or market research.

Time-series forecasting: Forecasting based on historical patterns over time.

Moving average: An average of recent observations used to smooth demand.

Exponential smoothing: A method assigning declining weights to older observations.

Seasonality: Systematic demand variation recurring at a known calendar frequency.

Causal forecasting: Forecasting demand from explanatory variables.

Forecast bias: Persistent tendency to over- or under-forecast.

MAE: Mean absolute error, measured in demand units.

RMSE: Root mean squared error, which gives extra weight to large errors.

Demand sensing: Use of recent, granular signals to update short-term demand estimates.

Aggregate planning: Medium-term balancing of demand, resources and capacity.

Level strategy: A plan maintaining relatively stable production or workforce.

Chase strategy: A plan varying resources to follow demand.

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

IBP: Integrated business planning linking operations with financial and strategic plans.

CPFR: Collaborative planning, forecasting and replenishment among supply-chain partners.

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

Review Questions

Short Answer Questions

1. Distinguish a forecast from a target and a budget.
2. State two situations in which qualitative forecasting is appropriate.
3. What is the effect of increasing the smoothing constant in simple exponential smoothing?
4. Why can MAPE be unsuitable when actual demand is zero or very small?
5. Identify two differences between S&OP and demand sensing.

Long Answer Questions

1. Explain the assumptions, strengths and limitations of moving averages, exponential smoothing and causal regression.
2. Discuss how aggregate planning balances level, chase and mixed strategies.
3. Explain the process and governance requirements of an effective S&OP or IBP cycle.
4. Evaluate the benefits and risks of CPFR for a manufacturer and a retailer.
5. Discuss how AI and near-real-time data can improve forecasting while creating governance risks.

Analytical / Application Questions

1. A product has actual weekly demand of 200, 240 and 220 units. Using a three-period moving average, calculate the forecast for week four and explain one limitation of the method.
2. A firm has a mean lead-time demand of 1,000 units, standard deviation of 120 units and a cycle-service factor of 1.65. Calculate the simplified reorder point and explain the assumptions behind it.
3. A retailer repeatedly increases supplier orders above its own forecast because of fear of shortage. Analyse how this may contribute to the bullwhip effect and recommend three collaborative controls.

Multiple Choice Questions with Answers

Question 1

Which method is most appropriate as a transparent benchmark for many short-term demand series?

A. Naïve forecasting

B. Delphi forecasting

C. Executive opinion

D. Conjoint analysis

Correct Answer: A. Naïve forecasting

Question 2

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

A. Less responsive to recent observations

B. More responsive to recent observations

C. Independent of actual demand

D. Unable to produce a forecast

Correct Answer: B. More responsive to recent observations

Question 3

If actual demand is 100 and the forecast is 90, using e_t=A_t−F_t, the error is:

A. −10 units

B. 0 units

C. 10 units

D. 190 units

Correct Answer: C. 10 units

Question 4

Which measure is expressed in the same units as demand and penalises large errors more than MAE?

A. MAPE

B. RMSE

C. Mean error

D. WAPE

Correct Answer: B. RMSE

Question 5

Demand sensing primarily differs from conventional medium-term forecasting through its use of:

A. Only annual budgets

B. Near-real-time and granular signals

C. No external data

D. Fixed production capacity

Correct Answer: B. Near-real-time and granular signals

Question 6

A level aggregate-planning strategy generally relies on which mechanism to absorb demand variation?

A. Stable resources with inventory or backlog changes

B. Eliminating all inventory

C. Changing workforce every day

D. Ignoring capacity constraints

Correct Answer: A. Stable resources with inventory or backlog changes

Question 7

The principal purpose of S&OP is to:

A. Replace all statistical models

B. Reconcile demand, supply and cross-functional decisions

C. Calculate only transport rates

D. Set individual employee wages

Correct Answer: B. Reconcile demand, supply and cross-functional decisions

Question 8

In the simplified reorder-point formula ROP=μ_LT+zσ_LT, zσ_LT represents:

A. Average order quantity

B. Safety-stock allowance for uncertainty

C. Annual purchasing cost

D. Forecast bias only

Correct Answer: B. Safety-stock allowance for uncertainty

Question 9

Which condition makes ordinary MAPE problematic?

A. All actual values are positive and large

B. Actual demand includes zero values

C. Forecasts are recorded monthly

D. The data have a product hierarchy

Correct Answer: B. Actual demand includes zero values

Question 10

A central requirement for effective CPFR is:

A. Partners must conceal all demand information

B. Shared data definitions, responsibilities and governance

C. Only the retailer may change forecasts

D. Replenishment must ignore promotions

Correct Answer: B. Shared data definitions, responsibilities and governance