Demand Forecasting and Integrated Planning

Introduction

Supply chains must make decisions before the future is known. A manufacturer must decide how much to purchase, produce, store and distribute even though future customer demand is uncertain. Demand forecasting provides a structured estimate of future requirements, while integrated planning connects that estimate with capacity, inventory, procurement, finance and customer service decisions.

Forecasting is not the same as prediction with certainty. It is a disciplined process for reducing uncertainty and improving decisions. A good forecast may still be wrong; a poor planning process can fail even when the forecast is accurate. Consequently, modern supply chain management combines statistical forecasting with managerial judgement, demand sensing, scenario analysis and cross-functional coordination.

This chapter examines qualitative and quantitative forecasting, forecast error, demand variability, aggregate planning, sales and operations planning (S&OP), integrated business planning (IBP) and collaborative planning. It also considers technology, governance and the practical limits of forecasting models.

Learning Objectives

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

1. Explain the purpose, scope and limitations of demand forecasting.
2. Select appropriate qualitative and quantitative forecasting methods.
3. Calculate and interpret common forecast-error measures.
4. Distinguish trend, seasonality, cycles, random variation and demand variability.
5. Apply aggregate-planning strategies to balance demand and resources.
6. Explain the roles of S&OP, IBP and collaborative planning.
7. Evaluate forecasting and planning decisions in a practical supply-chain setting.
8. Identify governance, ethical and technological issues affecting forecast quality.

1. Foundations of Demand Forecasting

1.1 Demand, sales and the forecast horizon

Demand is the quantity customers are willing and able to purchase under specified conditions. Sales are the transactions actually recorded. The two may differ because of stockouts, capacity constraints, price changes, delivery delays or order cancellations. A sales history therefore does not always represent unconstrained demand.

Forecasts are commonly classified by horizon:

The appropriate horizon should cover the longest relevant lead time plus the time needed to respond. A forecast for a product with a six-month procurement lead time cannot be replaced by a weekly forecast alone.

1.2 Components of a demand series

A time series may contain several components:

• Level: the underlying average demand.
• Trend: sustained upward or downward movement.
• Seasonality: a repeating pattern linked to calendar or operational cycles.
• Cycle: longer-term fluctuation associated with economic or industry conditions.
• Random variation: irregular movement that cannot be reliably anticipated.
• Event effects: temporary changes caused by promotions, strikes, regulation, weather or product launches.

Separating these components helps planners choose methods and avoid treating a one-time event as a permanent trend.

1.3 Forecasting principles

Several principles guide professional forecasting:

1. Forecasts should be generated at the level at which decisions are made.
2. Aggregate forecasts are normally more accurate than forecasts for individual items.
3. Forecast accuracy should be measured against an agreed baseline.
4. Bias is different from random error and requires separate monitoring.
5. A forecast should include assumptions, not merely a numerical output.
6. Forecasts should be updated when relevant information changes, but excessive adjustment can create instability.
7. Forecast ownership must be clear, while inputs should be cross-functional.

2. Qualitative Forecasting

Qualitative methods are especially useful when historical data are limited, unreliable or no longer representative. They rely on structured knowledge rather than purely numerical patterns.

2.1 Executive opinion

Senior managers develop a forecast from market knowledge, strategic plans and experience. This method is fast and useful for new products, but it can be affected by hierarchy, optimism, anchoring and personal incentives. It should therefore be supported by documented assumptions and independent challenge.

2.2 Sales-force composite

Sales representatives estimate demand in their territories, and the estimates are consolidated. Sales staff may understand customers closely, but may also understate demand to obtain easier targets or overstate it to secure inventory. Training, historical calibration and management review are important controls.

2.3 Customer surveys and market research

Surveys, interviews, conjoint studies, test marketing and panel data can provide information about customer intentions. Stated intention is not identical to actual purchasing behaviour, so results should be combined with behavioural data where possible.

2.4 Delphi method

The Delphi method obtains forecasts from a panel of experts through several anonymous rounds. A facilitator summarises the responses after each round, allowing participants to reconsider their views without direct social pressure. It is useful for long-term technology and market forecasts, though it can be time-consuming and depends on panel quality.

2.5 Scenario planning

Scenario planning does not produce one supposedly precise forecast. It develops coherent alternatives such as high-growth, constrained-supply and recession scenarios. Each scenario should identify assumptions, indicators and contingent actions.

Qualitative methods are strongest when structural change is expected. They are weakest when opinions are collected informally, when dissent is suppressed or when political targets are presented as forecasts.

3. Quantitative Forecasting

Quantitative methods use historical observations and measurable explanatory variables. They are not automatically objective: the choice of data, aggregation, exclusions and parameters involves judgement.

3.1 Naïve forecast

A simple benchmark sets the next-period forecast equal to the latest actual observation:

where F_t+1 is the forecast for period t+1 and A_t is actual demand in period t. It is appropriate as a baseline and can perform well for stable series. A seasonal naïve method uses demand from the corresponding previous season:

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

3.2 Moving average

A simple moving average uses the mean of the most recent n observations:

All observations receive equal weight. A larger n smooths random variation but responds more slowly to genuine changes.

Example. Demand for the last three months was 100, 120 and 110 units. The three-month moving-average forecast is:

F_t+1 = (100 + 120 + 110) ÷ 3 = 110 units.

3.3 Weighted moving average

A weighted moving average assigns different weights:

The weights must sum to 1. A higher weight on recent demand increases responsiveness but may increase sensitivity to noise.

3.4 Exponential smoothing

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

Here 0 ≤ α ≤ 1 is the smoothing constant, A_t is actual demand and F_t is the previous forecast. A high α reacts quickly; a low value produces greater smoothing. The method assumes no systematic trend or seasonality unless extensions are used.

A useful equivalent form is:

The term A_t − F_t is the forecast error for period t.

3.5 Trend and seasonal methods

When trend exists, methods such as Holt’s trend-adjusted exponential smoothing can be used. When both trend and seasonality exist, Holt–Winters methods may be appropriate. Additive seasonality assumes that seasonal effects are approximately constant in units; multiplicative seasonality assumes that seasonal effects vary in proportion to the level.

Method selection should be based on out-of-sample performance and operational suitability, not on complexity alone.

3.6 Causal and regression models

A causal model relates demand to explanatory variables such as price, advertising, income, temperature or construction activity. A simple linear model is:

where Y is demand, X is an explanatory variable, β₀ is the intercept, β₁ is the estimated effect per unit of X, and ε is unexplained error. The relationship should not be interpreted as causal merely because two variables move together. Business knowledge, appropriate data and diagnostic testing are required.

4. Forecast Error and Accuracy

A forecast error is commonly defined as:

Positive error means actual demand exceeded the forecast; negative error means the forecast was higher than actual demand. Some organisations use the opposite sign, so the convention must be stated.

4.1 Common measures

Here n is the number of observations. Percentage measures require non-zero actual demand. MAPE can be unstable for low-volume or intermittent items and is unsuitable when A_t = 0. Alternatives include weighted absolute percentage error or scale-dependent measures such as MAE.

Worked example. Actual demand over three periods is 100, 120 and 80 units; forecasts are 90, 130 and 100 units. Errors are 10, −10 and −20 units. Absolute errors total 40, so:

MAE = 40 ÷ 3 = 13.33 units.

Squared errors total 100 + 100 + 400 = 600, so:

MSE = 600 ÷ 3 = 200 units², and RMSE = √200 ≈ 14.14 units.

Absolute percentage errors are 10%, 8.33% and 25%; therefore MAPE ≈ 14.44%. MPE is (10% − 8.33% − 25%) ÷ 3 ≈ −7.78%, indicating overall over-forecasting.

4.2 Bias and tracking signal

Bias is persistent directional error. A common cumulative-error measure is:

A tracking signal may be defined as:

where MAD is mean absolute deviation, often equivalent in practice to MAE. The acceptable range is organisation-specific. A sustained positive value suggests under-forecasting; a sustained negative value suggests over-forecasting. A single threshold should not be treated as a universal law.

Accuracy metrics should be segmented by product, customer, horizon and value. A portfolio can have good aggregate accuracy while failing on strategically important items.

5. Managing Demand Variability

Demand variability arises from customer behaviour, promotions, product substitution, economic conditions, weather, order batching and information delays. Variability propagates upstream as the bullwhip effect: small changes in end demand may produce larger fluctuations in replenishment orders.

Managers can reduce variability by sharing point-of-sale data, stabilising prices, reducing lead times, limiting unnecessary order batching, improving forecast discipline and coordinating promotions. Variability cannot always be eliminated; it must also be buffered through inventory, capacity, time or information.

Forecast uncertainty should be distinguished from supply uncertainty. A forecast distribution describes possible demand outcomes, whereas safety stock protects against uncertainty during replenishment. In a simplified independent-demand setting, cycle-service safety stock may be represented as:

where SS is safety stock, z is the selected service-factor, and σ_L is the standard deviation of demand during lead time, measured in units. This formula assumes that lead-time demand is approximately normally distributed and that the relevant standard deviation is correctly estimated. If demand and lead-time variability are both material, a more general variance expression is required; planners should not use the simplified formula automatically.

6. Aggregate Planning

Aggregate planning converts medium-term demand expectations into a feasible plan for product families, labour, capacity, inventory and subcontracting. It uses aggregated units because detailed SKU-level decisions are usually too unstable at this horizon.

The principal strategies are:

• Level strategy: maintain a relatively stable workforce and production rate, using inventory or backorders to absorb demand changes.
• Chase strategy: vary production or workforce to follow demand.
• Mixed strategy: combine regular production with overtime, inventory, temporary labour, subcontracting or controlled backlog.

A simplified planning cost model is:

where TC is total planning cost; P regular production; I inventory; B backorders; O overtime; S subcontracting; L labour changes; and the corresponding C terms are unit costs. A plan must also satisfy a flow balance such as:

Definitions and signs must be adapted if backorders are represented as a separate variable. The equation expresses that ending available stock equals beginning stock plus supply minus demand, with backlog treatment explicitly specified.

Aggregate planning is not merely a cost-minimisation exercise. Excessive overtime may damage safety and quality; layoffs may destroy scarce skills; subcontracting may expose intellectual property; and backorders may reduce customer trust. Service, resilience, sustainability and workforce considerations should be included in evaluation.

7. Sales and Operations Planning

S&OP is a recurring cross-functional process that reconciles demand and supply at an aggregate level. It normally includes:

1. Data gathering and statistical baseline: collect actuals, inventory, capacity, orders and forecast history.
2. Demand review: evaluate the unconstrained demand plan, market intelligence, promotions and risks.
3. Supply review: test production, procurement, labour, logistics and supplier constraints.
4. Reconciliation: evaluate gaps, scenarios, trade-offs and financial consequences.
5. Executive review: approve decisions, priorities, assumptions and accountability.

Effective S&OP has one agreed set of numbers, a defined calendar, decision rights and escalation rules. It should not become a meeting in which departments defend incompatible spreadsheets.

S&OP usually operates monthly and at product-family or business-unit level. It links operational plans to financial plans but may not fully integrate product development, strategic investment, risk and portfolio decisions.

8. Integrated Business Planning and Collaborative Planning

IBP extends S&OP by integrating strategic, commercial, operational and financial planning. It may include new-product introductions, capital expenditure, market portfolios, cash implications, risk scenarios and long-range capacity. The distinction between S&OP and IBP varies across organisations; IBP should therefore be defined by scope and decision rights rather than by terminology.

Collaborative planning involves multiple firms sharing information and coordinating decisions. Important approaches include:

• CPFR (Collaborative Planning, Forecasting and Replenishment): partners jointly develop forecasts and replenishment plans.
• Vendor-managed inventory: the supplier monitors agreed inventory information and manages replenishment within defined limits.
• Collaborative promotion planning: trading partners coordinate promotional timing, volume and replenishment capacity.
• Joint business planning: partners align objectives, investments, risks and performance measures.

Collaboration requires data standards, confidentiality controls, agreed ownership and trust. Sharing forecasts without sharing assumptions, inventory policies or promotional calendars may create only the appearance of collaboration.

9. Technology, AI and Governance

Enterprise resource planning systems, advanced planning systems, cloud platforms, point-of-sale feeds and control towers can improve visibility and calculation speed. Machine-learning methods may detect nonlinear relationships, demand signals and anomalies. They are most valuable when data are sufficiently rich and decisions are repeatable.

AI does not remove the need for domain expertise. Risks include biased historical data, leakage of future information into training data, unstable behaviour after market changes, opaque recommendations and automation of incorrect master data. Forecast models should be validated using time-based holdout periods, monitored after deployment and compared with simple baselines.

Governance should address:

• data ownership and quality standards;
• version control and audit trails;
• human approval of high-impact decisions;
• privacy and commercial confidentiality;
• explainability and model-performance monitoring;
• cybersecurity and access rights;
• clear distinction between a statistical forecast and a management override.

A management override should have a recorded reason, author, date, expected effect and later review. Overrides that consistently improve accuracy should inform model redesign; overrides that merely shift targets should not be treated as superior forecasts.

10. Illustrative Case: Meridian Foods

Background

Meridian Foods, a fictional chilled-meals producer, supplies supermarkets through regional distribution centres. Demand rises during winter and around selected promotional events. The company frequently experiences stockouts during promotions and excess finished goods afterward.

Problem

Sales staff increase forecasts before promotions, but marketing does not consistently communicate event timing or expected uplift. Production responds with overtime, while procurement orders ingredients using a different spreadsheet. The result is high expediting cost, variable service and avoidable waste.

Analysis

The planning team separates baseline demand from promotion demand and identifies three years of weekly history. It finds that promotional weeks have a different pattern from ordinary weeks and that stockouts caused recorded sales to understate demand. Forecast performance is measured using MAE and bias rather than MAPE alone because some meal variants have low volume.

The team compares a seasonal naïve forecast, a weighted moving average and an exponential-smoothing model. The statistical forecast becomes the baseline. Marketing supplies a documented promotion calendar; sales provides customer intelligence; operations confirms capacity; finance translates alternatives into margin and working-capital effects.

Approach and solution

Meridian introduces a monthly S&OP cycle and a weekly short-term exception review. For each promotion, the team records expected uplift, start and end dates, customer commitments and cancellation risk. Production uses a mixed aggregate strategy: stable base production, limited pre-build for reliable promotions and approved overtime for short peaks. Procurement receives a time-phased requirement rather than an informal email.

The company also establishes a forecast-value-added review. Overrides are retained only when their rationale is documented and their subsequent performance is better than the baseline. Inventory and waste targets are considered together with service targets.

Outcome

After two planning cycles, the organisation has not eliminated uncertainty, but it has reduced unexplained overrides, improved promotion readiness and made capacity conflicts visible earlier. The principal benefit is not a perfect forecast; it is faster, more coordinated and more accountable decision-making.

Learning points

1. Recorded sales may understate true demand when stockouts occur.
2. Promotion forecasts require event information, not only historical averages.
3. Cross-functional planning is more valuable than isolated forecast sophistication.
4. Forecast accuracy, service, waste and financial outcomes must be reviewed together.

11. Best Practices and Limitations

Best practice begins with a clean demand history, explicit treatment of stockouts and exceptional events, stable product and customer hierarchies, and a documented forecast calendar. Methods should be selected through back-testing and evaluated at the decision-relevant horizon. Forecasts should be reconciled across levels so that family totals and item forecasts are not contradictory.

Forecasting remains limited by structural breaks, new products, intermittent demand, competitor actions, poor data and unknown shocks. No historical model can reliably infer an event that has no historical analogue. Scenario planning, expert judgement and resilience measures are therefore complements to statistical forecasting, not signs of methodological failure.

Managers should resist false precision. A forecast of 10,237 units may communicate more certainty than the evidence supports. Ranges, scenarios and confidence information are often more useful for capacity and risk decisions than a single number.

Chapter Summary

Demand forecasting estimates future requirements so that supply-chain resources can be planned. Qualitative methods use expert and market knowledge and are particularly useful for new products and structural change. Quantitative methods include naïve forecasts, moving averages, exponential smoothing, seasonal methods and causal models. Method choice should reflect data characteristics, decision horizon and operational purpose.

Forecast error is the difference between actual and forecast demand. MAE, MSE, RMSE, MAPE, MPE, cumulative forecast error and tracking signals provide different information; bias and random error must be distinguished. Demand variability can be reduced through information sharing, stable processes and shorter lead times, while remaining uncertainty can be buffered with inventory, capacity, time or information.

Aggregate planning balances demand with production, labour, inventory, overtime, subcontracting and backlog. S&OP creates a cross-functional process for reconciling demand and supply. IBP extends this integration to finance, strategy, portfolio and risk. Collaborative planning aligns supply-chain partners through shared information and coordinated decisions. Technology and AI can strengthen forecasting, but require sound data, validation, human oversight and governance.

Key Terms

Demand forecast: A structured estimate of future customer requirements.

Forecast horizon: The future period covered by a forecast.

Level: The underlying average value of a demand series.

Trend: Sustained upward or downward movement in demand.

Seasonality: Repeating variation associated with a calendar or operating cycle.

Demand variability: Fluctuation in customer requirements over time.

Naïve forecast: A forecast based on a recent or corresponding previous observation.

Moving average: The mean of a specified number of recent observations.

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

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

Bias: Persistent directional forecast error.

MAE: Mean absolute error, measured in demand units.

MAPE: Mean absolute percentage error.

Aggregate planning: Medium-term planning for product families and resources.

Level strategy: A strategy using relatively stable production and workforce levels.

Chase strategy: A strategy that adjusts production or workforce to demand.

S&OP: Sales and operations planning that reconciles aggregate demand and supply.

IBP: Integrated business planning linking operational, financial and strategic plans.

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

Management override: A documented human adjustment to a statistical forecast.

Review Questions

Short Answer Questions

1. Distinguish demand from recorded sales.
2. What is the difference between seasonality and trend?
3. State the formula for mean absolute error and define its variables.
4. Why can MAPE be unsuitable for intermittent demand?
5. What is the primary purpose of an S&OP process?

Long Answer Questions

1. Compare qualitative forecasting methods and explain when each may be appropriate.
2. Explain moving averages, weighted moving averages and exponential smoothing, including their responsiveness and limitations.
3. Discuss how demand variability produces the bullwhip effect and identify methods for reducing it.
4. Compare level, chase and mixed aggregate-planning strategies using cost, service, workforce and resilience criteria.
5. Explain how IBP extends S&OP and identify the governance requirements for effective implementation.

Analytical / Application Questions

1. Actual demand for three periods is 150, 180 and 120 units, while forecasts are 160, 170 and 140 units. Calculate MAE, RMSE and MPE, and interpret the direction of bias.
2. A distributor has highly seasonal demand and a four-month supplier lead time. Recommend a forecasting and planning approach, specifying the role of seasonal history, scenarios, safety stock and supplier collaboration.
3. A company’s statistical forecast is repeatedly changed by sales managers before each S&OP meeting. Design a governance process to determine whether the overrides add forecast value or merely convert sales targets into forecasts.

Multiple Choice Questions with Answers

Question 1

Which component represents a repeating demand pattern associated with a calendar cycle?

A. Trend

B. Seasonality

C. Random variation

D. Forecast bias

Correct Answer: B. Seasonality

Question 2

Demand in the last three months was 90, 120 and 150 units. What is the three-month moving-average forecast for the next month?

A. 110 units

B. 120 units

C. 130 units

D. 360 units

Correct Answer: B. 120 units

Question 3

If actual demand is 500 units and the forecast is 460 units, using e_t = A_t − F_t, what is the forecast error?

A. −40 units

B. 40 units

C. 460 units

D. 960 units

Correct Answer: B. 40 units

Question 4

Which measure is expressed in the same physical units as demand and calculates the average absolute error?

A. MAE

B. MAPE

C. MPE

D. MSE

Correct Answer: A. MAE

Question 5

A sustained negative cumulative forecast error generally indicates that the organisation is:

A. Under-forecasting demand

B. Over-forecasting demand

C. Using seasonal data

D. Reducing lead time

Correct Answer: B. Over-forecasting demand

Question 6

Which aggregate-planning strategy keeps production relatively stable and uses inventory or backlog to absorb demand changes?

A. Chase strategy

B. Level strategy

C. Pure subcontracting strategy

D. Zero-inventory strategy

Correct Answer: B. Level strategy

Question 7

What is the main purpose of S&OP?

A. To replace all operational schedules

B. To reconcile aggregate demand and supply across functions

C. To calculate only transportation costs

D. To eliminate all uncertainty

Correct Answer: B. To reconcile aggregate demand and supply across functions

Question 8

Which statement best describes IBP?

A. It is limited to warehouse slotting

B. It extends planning by linking operational, financial and strategic decisions

C. It uses only qualitative forecasts

D. It prevents managers from reviewing forecasts

Correct Answer: B. It extends planning by linking operational, financial and strategic decisions

Question 9

Why should MAPE be used cautiously for demand series containing zero actual values?

A. It always produces a negative result

B. Its percentage calculation becomes undefined or unstable

C. It measures squared rather than absolute error

D. It ignores the forecast value

Correct Answer: B. Its percentage calculation becomes undefined or unstable

Question 10

Which practice best supports responsible use of AI forecasting?

A. Automatically accept every model output

B. Delete historical exceptions without investigation

C. Use time-based validation, monitoring and documented human oversight

D. Replace all domain experts with the algorithm

Correct Answer: C. Use time-based validation, monitoring and documented human oversight