Demand Forecasting and Supply Planning

Introduction

Demand forecasting is the disciplined process of estimating future customer requirements using historical information, market knowledge, statistical methods and managerial judgement. Supply planning converts those estimates into decisions about procurement, production, inventory, workforce, transport and capacity. Together, forecasting and supply planning help an organisation balance customer service, operating cost and resource utilisation.

Forecasts are not promises. They are estimates made under uncertainty. A professional supply-chain manager therefore asks not only, “What is the forecast?” but also, “How reliable is it?”, “What assumptions support it?”, “How will uncertainty be managed?” and “What decisions should change if demand is higher or lower than expected?”

A forecast may be prepared at several levels: product, product family, customer, region, channel or total business. It may also cover different time horizons. Operational forecasts support replenishment and scheduling; medium-term forecasts support aggregate planning and budgeting; long-term forecasts support capacity, network and investment decisions.

Learning Objectives

After studying this chapter, you should be able to:

1. Explain the purpose, horizons and hierarchy of demand forecasts.
2. Distinguish qualitative and quantitative forecasting methods.
3. Apply moving-average, weighted-average and exponential-smoothing methods.
4. Recognise trend, seasonality, cycles, random variation and demand intermittency.
5. Measure forecast error using appropriate performance indicators.
6. Explain how forecasts are translated into capacity, aggregate plans and inventory decisions.
7. Describe the role of sales and operations planning (S&OP) and integrated business planning (IBP).
8. Evaluate forecasting limitations, governance requirements and the responsible use of technology.

1. The Nature of Demand in Supply Chains

Demand is the quantity of a product or service that customers are willing and able to purchase during a specified period. It differs from sales: sales may be constrained by stockouts, production limits, credit restrictions or distribution failures. If a retailer sells 800 units because only 800 were available, actual sales do not necessarily represent true demand.

Forecasting should therefore distinguish:

• Independent demand: demand originating outside the organisation, such as customer demand for finished goods.
• Dependent demand: demand calculated from the requirement for another item, such as components needed for a production plan.
• Observed demand: recorded orders or sales, which may be distorted by promotions, shortages or cancellations.
• Unconstrained demand: estimated demand assuming adequate availability and no artificial restrictions.

A demand pattern commonly contains the following components:

The distinction between seasonality and cycles is important. Seasonality repeats at a known interval, whereas cycles may have variable length and are often associated with economic or industry conditions.

2. Forecasting Horizons and Aggregation

Forecast horizons should match the decision being supported.

Forecast accuracy generally decreases as the horizon increases. Accuracy also tends to improve when demand is aggregated. Demand for one SKU may be volatile, while the total demand for a product family is relatively stable. This is known as the portfolio effect.

Forecasts should be generated at the level at which decisions are made, but reconciled across levels. For example, forecasts for individual warehouses should be consistent with the regional and company-wide forecast. This process is called forecast hierarchy reconciliation.

3. Qualitative Forecasting Methods

Qualitative methods are particularly useful when historical data are limited, the product is new, market conditions have changed, or expert knowledge contains information not yet visible in the data.

3.1 Executive opinion

Senior managers provide a collective estimate. It is fast and useful for strategic decisions, but power relationships, optimism and anchoring can distort the result. A structured process should require explicit assumptions and evidence.

3.2 Sales-force composite

Sales representatives estimate demand in their territories, and the estimates are consolidated. Sales staff may understand local customers well, but they may also understate demand to obtain easier targets or overstate it to secure inventory.

3.3 Customer and market research

Surveys, interviews, test marketing and conjoint studies can reveal customer intentions and preferences. Stated intention, however, may differ from actual purchase behaviour.

3.4 Delphi method

A panel of experts submits forecasts independently. A facilitator summarises the responses and circulates them for further rounds until a reasoned convergence is achieved. Anonymity reduces the influence of dominant personalities.

3.5 Analogy and scenario analysis

A new product may be compared with a similar product, adjusted for price, market size and distribution. Scenario analysis develops alternative futures, such as base, high-demand and low-demand cases. Scenarios are especially valuable when structural uncertainty is high.

Qualitative judgement should not be treated as a substitute for analysis. A sound practice is to document the adjustment, its rationale, its owner and the date on which it should be reviewed.

4. Quantitative Forecasting Methods

Quantitative methods use numerical observations and explicit rules. They are most useful when reliable historical data exist and the future is expected to have some relationship with the past.

4.1 Naïve forecast

The simplest method uses the most recent actual value as the next forecast:

Here, F_t+1 is the forecast for period t+1, and A_t is actual demand in period t. The method is inexpensive and provides a useful benchmark. It is often reasonable for stable demand, but it reacts strongly to unusual observations.

For seasonal demand, a seasonal naïve method uses the value from the same season in the previous cycle:

where s is the number of periods in one seasonal cycle. This is applicable when the seasonal pattern is reasonably stable.

4.2 Simple moving average

A moving average uses the most recent n observations:

n is the number of observations, all measured in the same units and time interval. A larger n smooths random variation but responds more slowly to genuine changes. A moving average is most appropriate when there is no strong trend or seasonality.

Worked example. Weekly demand for the last four weeks was 120, 140, 130 and 150 units. The four-week moving-average forecast for week 5 is:

4.3 Weighted moving average

More recent observations may receive greater importance:

The weights must satisfy w_i ≥ 0 and ∑w_i = 1. If weights do not sum to one, the result is not in the same demand units unless a suitable scaling adjustment is made.

Suppose the last three demands were 100, 120 and 140 units, and weights of 0.2, 0.3 and 0.5 are assigned from oldest to newest. The forecast is:

4.4 Exponential smoothing

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

Equivalently:

α is the smoothing constant, with 0 < α < 1. A high value reacts rapidly to changes; a low value produces greater smoothing. Simple exponential smoothing assumes no systematic trend or seasonality.

If A_t = 140, F_t = 126 and α = 0.3, then:

The initial forecast must be specified, for example as the first actual observation or an initial average. Different initialisation choices can affect early results.

4.5 Trend and seasonality

When trend exists, methods such as Holt’s double exponential smoothing can be used. One common additive form is:

L_t is the level, T_t is the trend, m is the number of periods ahead, and α and β are smoothing constants between zero and one. The additive form is suitable when seasonal fluctuations are approximately constant in size. Multiplicative seasonal models are more suitable when seasonal variation grows with the level of demand.

A common decomposition expresses demand as:

for an additive model, where T_t is trend, S_t is seasonal effect and R_t is residual variation. In a multiplicative model:

The additive model is appropriate when seasonal differences remain roughly constant; the multiplicative model is appropriate when seasonal ratios remain roughly constant. Decomposition requires sufficient history and stable definitions of the seasons.

4.6 Causal and regression forecasting

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

where y is demand, x is an explanatory variable, a is the intercept, b is the estimated change in demand for a one-unit change in x, and ε is unexplained error. The relationship should not be interpreted as causal merely because two variables are correlated. Business logic, timing and possible confounding factors must be examined.

5. Forecast Error and Accuracy

For period t, forecast error can be defined as:

A positive error means actual demand exceeded the forecast; a negative error means the forecast exceeded actual demand. Organisations must use one consistent sign convention.

Important measures include:

In these formulas, n is the number of evaluated periods. MAPE is undefined when A_t = 0 and unstable when actual demand is very small. WAPE is often more useful for portfolios, but it can hide poor performance on low-volume items. No single metric is universally best.

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

The negative ME indicates overall over-forecasting. MAE indicates the average absolute deviation, without allowing positive and negative errors to cancel.

Forecast evaluation must use a holdout period or rolling-origin testing where possible. Evaluating a model only on the data used to build it can produce an overly favourable result. Comparisons should cover the same products, periods, demand units and data exclusions.

6. Variability, Safety Stock and Service

Demand variability creates a risk that actual demand during replenishment lead time will exceed available stock. Safety stock is inventory held to protect against uncertainty. In a simplified continuous-review system with approximately independent demand and lead-time variation, a common reorder-point expression is:

where ROP is reorder point in units, μ_D is average demand per time unit, L is lead time in the same time unit, and SS is safety stock.

If demand varies but lead time is constant, a common approximation is:

where σ_D is the standard deviation of demand per time unit and z is the standard-normal service factor associated with the chosen cycle-service probability. This formula assumes independent demand across periods, stable variation and a fixed lead time. It should not be applied uncritically to intermittent demand, correlated demand or highly variable lead times.

If both demand and lead time vary and are independent, an often-used approximation for the standard deviation of lead-time demand is:

and:

Here σ_L is the standard deviation of lead time. The independence assumption is important; correlation between demand and lead time requires a more complete variance model.

A cycle service level is the probability of avoiding a stockout during a replenishment cycle. A fill rate is the proportion of demand supplied immediately from available stock. They are different measures and should not be used interchangeably.

7. From Forecast to Supply Plan

A forecast becomes useful only when translated into coordinated action. The conversion normally follows this logic:

1. Create a baseline demand forecast.
2. Add validated information about promotions, contracts, market events and product changes.
3. Review assumptions and identify risks.
4. Aggregate demand into product families and time buckets.
5. Compare demand with available capacity and supply.
6. Select responses such as overtime, subcontracting, inventory build, alternative sourcing, backlog or demand shaping.
7. Approve a feasible plan and monitor deviations.

Capacity planning

Capacity is the maximum output a resource can provide during a specified period under stated conditions. Design capacity is the theoretical maximum; effective capacity reflects maintenance, setups, breaks, product mix and operating policies. A useful measure is:

Utilisation is not the same as efficiency. A plant can have high utilisation but poor performance if it produces the wrong mix or experiences excessive rework.

Capacity may be adjusted through overtime, temporary labour, subcontracting, cross-training, inventory, product prioritisation, alternate equipment or capital investment. Decisions should consider cost, lead time, quality, labour rules and supplier risk.

Aggregate planning

Aggregate planning develops a medium-term plan for product families rather than individual stock-keeping units. It balances forecast demand with workforce, production rate, inventory and backlog. Common strategies include:

• 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.
• Mixed strategy: combine stable core capacity with overtime, temporary labour, inventory or subcontracting.

A simplified planning balance is:

If backorders are permitted, ending inventory may be negative and represents backlog. The identity applies over a consistent time bucket and requires all quantities to use the same unit definition.

Master production scheduling and material requirements

The aggregate plan is disaggregated into a master production schedule (MPS), which specifies what finished products are to be produced and when. Material requirements planning (MRP) then explodes the MPS through the bill of materials, offsets requirements by lead times and produces planned orders for components and materials. The accuracy of MRP depends on bills of material, inventory records, lead times, lot-sizing rules and the quality of the MPS.

8. S&OP and Integrated Business Planning

Sales and operations planning is a cross-functional process that reconciles demand, supply and financial implications. A typical monthly cycle includes:

1. Data and forecast review: establish a common baseline.
2. Demand review: incorporate commercial intelligence, promotions and customer changes.
3. Supply review: test capacity, materials, labour, suppliers and logistics.
4. Pre-S&OP reconciliation: evaluate alternatives and financial effects.
5. Executive review: approve decisions, priorities and risks.
6. Execution and monitoring: track performance and revise when assumptions change.

IBP extends this logic by integrating strategic plans, financial plans, product portfolios, risk and investment decisions. The objective is not merely to produce a forecast; it is to create one agreed set of assumptions and decisions across functions.

Good S&OP governance defines data ownership, decision rights, escalation thresholds, meeting calendars and performance measures. It also prevents local optimisation—for example, sales promising demand that operations cannot supply, or purchasing reducing unit price while increasing total inventory and obsolescence.

9. Illustrative Case: Forecasting Seasonal Demand for a Distributor

Background

This illustrative case concerns Northstar Tools, a distributor of portable water pumps. Demand rises sharply before the rainy season. The company has limited warehouse space and a supplier lead time of six weeks.

Problem

Sales representatives proposed a large inventory build based on expected rainfall. The statistical forecast, based on recent monthly demand, was lower. Last year the company experienced stockouts during the peak season, but it also carried obsolete stock afterward.

Analysis

The planning team separated normal demand from a one-time emergency order in the previous year. It plotted monthly demand over three years and identified a repeating seasonal pattern. It calculated a seasonal index for each month and compared a seasonal naïve forecast, a moving-average forecast and a judgementally adjusted forecast. Performance was assessed using WAPE and bias, not MAPE alone, because several months had very low demand.

The supply team then modelled three scenarios: low rainfall, expected rainfall and high rainfall. For each scenario it checked supplier capacity, warehouse space, cash requirements and transport availability. The high-demand scenario could not be met entirely through normal replenishment because the six-week lead time exceeded the remaining planning window.

Approach and solution

Northstar approved a base plan using the statistical seasonal forecast, with a controlled uplift supported by weather information and customer commitments. It reserved supplier capacity for an additional production lot, arranged a temporary overflow warehouse and classified pumps into service-critical and non-critical categories. A weekly exception review was established during the season. The plan also included a post-season markdown and redeployment policy for residual stock.

Outcome

The company reduced the risk of peak-season stockouts without adopting the full sales-force estimate. It improved transparency by documenting the assumptions behind the uplift and assigning responsibility for monitoring rainfall, customer orders and supplier confirmations.

Learning points

The case demonstrates that forecasting and planning are connected but not identical. A statistical model provided a disciplined baseline; judgement added current market information; scenario planning exposed capacity limits; and governance converted uncertainty into explicit decisions. The best forecast is not necessarily the most complex model, but the one that is reliable, explainable and linked to action.

10. Technology, AI and Data Governance

Enterprise resource planning systems, advanced planning systems and business-intelligence tools can automate data collection, forecast generation, exception alerts and scenario analysis. Machine-learning models may identify nonlinear relationships across prices, promotions, weather and customer behaviour. They are most valuable when data are sufficiently rich and decisions are repeatable.

Technology does not remove the need for professional judgement. Risks include data leakage, changing customer behaviour, biased training data, unexplained model outputs, excessive model complexity and false confidence in precise-looking forecasts. AI-generated forecasts should be benchmarked against naïve methods, tested out of sample and monitored for drift.

Data governance should address:

• consistent product, customer and calendar definitions;
• traceable data sources and version control;
• controls over manual overrides;
• access security and privacy;
• documented model assumptions;
• audit trails for forecast changes; and
• human accountability for consequential supply decisions.

Ethically, organisations should avoid manipulating forecasts to shift blame or secure favourable performance measures. Forecasts influence employment, supplier commitments, customer allocation and investment; therefore, transparency and fair treatment matter.

11. Challenges and Best Practices

Forecasting is difficult when products are new, demand is intermittent, promotions create demand shifts, markets are disrupted or data are contaminated by stockouts. Common errors include using sales as demand without adjustment, overreacting to one unusual observation, ignoring lead-time changes, measuring accuracy only at aggregate level and allowing unrecorded overrides.

Recommended practices are:

1. Define the purpose, horizon, unit and level of every forecast.
2. Clean and document data, including stockouts, returns and promotions.
3. Establish a simple benchmark before adopting a sophisticated method.
4. Match the method to the demand pattern and data availability.
5. Measure both accuracy and bias at relevant aggregation levels.
6. Use rolling evaluation and monitor forecast stability.
7. Separate baseline statistical forecasts from documented business adjustments.
8. Use scenarios where structural uncertainty is greater than statistical noise.
9. Translate forecast uncertainty into capacity, inventory and supplier decisions.
10. Review assumptions through a cross-functional S&OP process.

Chapter Summary

Demand forecasting estimates future requirements, while supply planning converts those estimates into feasible decisions about capacity, inventory, production, procurement and logistics. Forecasts may be qualitative or quantitative. Naïve, moving-average, weighted-average and exponential-smoothing methods are useful for stable demand, while trend, seasonal and causal methods are required when patterns or drivers are more complex.

Forecast error must be measured consistently. Mean error identifies bias; MAE and RMSE express error in units; percentage measures support comparison but have limitations when actual demand is zero or very small. Demand variability and lead-time uncertainty influence safety stock and service performance. Forecasts should be aggregated appropriately and reconciled across the forecast hierarchy.

Aggregate planning, the MPS and MRP connect demand to resources and materials. S&OP and IBP provide cross-functional governance by reconciling commercial, operational and financial plans. Technology and AI can improve analysis, but only when supported by reliable data, transparent assumptions, monitoring and human accountability. Effective forecasting is therefore not simply a mathematical exercise; it is an organisational process for making better decisions under uncertainty.

Key Terms

Demand forecast: An estimate of future customer requirements.

Independent demand: Demand originating from external customers.

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

Demand variability: Fluctuation in demand over time.

Seasonality: A repeating demand pattern linked to a regular calendar interval.

Trend: A sustained upward or downward movement in demand.

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

Moving average: An average of recent observations used as a forecast.

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

Causal forecasting: Forecasting that relates demand to explanatory variables.

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

Bias: Persistent tendency to over-forecast or under-forecast.

MAE: Mean absolute error, expressed in the original demand units.

WAPE: Weighted absolute percentage error for aggregated demand.

Safety stock: Inventory held to protect against uncertainty.

Reorder point: Inventory position at which replenishment is triggered.

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

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

IBP: Integrated business planning linking operations, strategy and finance.

Forecast hierarchy: Levels of forecasts from detailed items to total business.

Review Questions

5 Short Answer Questions

1. Distinguish independent demand from dependent demand.
2. What is the difference between seasonality and a cycle?
3. State one advantage and one limitation of a moving-average forecast.
4. What does a negative mean forecast error indicate when error is defined as actual minus forecast?
5. Why should a statistical baseline and a business judgement adjustment be documented separately?

5 Long Answer Questions

1. Explain the major qualitative forecasting methods and discuss situations in which each may be appropriate.
2. Describe simple exponential smoothing, define its variables and explain the effect of changing the smoothing constant.
3. Compare ME, MAE, RMSE, MAPE and WAPE, including the limitations of percentage measures.
4. Explain how a demand forecast is translated into an aggregate plan, master production schedule and material requirements plan.
5. Describe the stages of an S&OP cycle and explain how IBP extends the process.

3 Analytical / Application Questions

1. A company has weekly demands of 200, 220, 180 and 240 units. Calculate the four-week moving-average forecast for the next week. Explain whether this method is suitable if demand has a strong seasonal pattern.
2. Actual demand in three months is 500, 600 and 400 units, while forecasts are 450, 650 and 500 units. Calculate mean error and mean absolute error using error = actual − forecast, and interpret the results.
3. A distributor faces an uncertain rainy season, a six-week supplier lead time and limited warehouse capacity. Design a forecasting and supply-planning process that combines a statistical baseline, scenarios, capacity checks, safety-stock decisions and S&OP governance.

Multiple Choice Questions with Answers

Question 1

Which statement best describes independent demand?

A. Demand calculated from a bill of materials

B. Demand originating from external customers

C. Demand caused only by internal production schedules

D. Demand for maintenance components determined by machine usage

Correct Answer: B. Demand originating from external customers

Question 2

Weekly demand for the last three weeks was 100, 120 and 140 units. Using weights 0.2, 0.3 and 0.5 from oldest to newest, what is the weighted forecast?

A. 116 units

B. 120 units

C. 126 units

D. 132 units

Correct Answer: C. 126 units

Question 3

In simple exponential smoothing, what happens when the smoothing constant α is increased?

A. The forecast reacts more strongly to recent observations

B. The forecast ignores the latest observation

C. The forecast becomes permanently equal to the historical average

D. The forecast can no longer be expressed in demand units

Correct Answer: A. The forecast reacts more strongly to recent observations

Question 4

If actual demand is 100 units and forecast demand is 120 units, using error = actual − forecast, what is the forecast error?

A. 20 units

B. −20 units

C. 100 units

D. 120 units

Correct Answer: B. −20 units

Question 5

Which accuracy measure is expressed in the original demand units and does not allow positive and negative errors to cancel?

A. Mean absolute error

B. Mean error

C. Mean percentage error

D. Bias ratio

Correct Answer: A. Mean absolute error

Question 6

Why can MAPE be problematic for intermittent demand?

A. It always overstates forecasts

B. It cannot be calculated when actual demand is zero

C. It requires demand to have a trend

D. It is measured in kilograms rather than percentages

Correct Answer: B. It cannot be calculated when actual demand is zero

Question 7

Which strategy maintains a relatively stable production rate and uses inventory or backlog to absorb demand changes?

A. Chase strategy

B. Level strategy

C. Zero-inventory strategy

D. Spot-buying strategy

Correct Answer: B. Level strategy

Question 8

What is the primary purpose of S&OP?

A. To replace all operational forecasts with sales targets

B. To reconcile demand, supply and financial implications across functions

C. To calculate only component requirements

D. To eliminate all demand uncertainty

Correct Answer: B. To reconcile demand, supply and financial implications across functions

Question 9

If average demand is 50 units per day, lead time is 4 days and safety stock is 30 units, what is the reorder point?

A. 80 units

B. 170 units

C. 200 units

D. 230 units

Correct Answer: D. 230 units

Question 10

Which practice is most appropriate when a forecast is influenced by a promotion?

A. Ignore the promotion because historical data are always sufficient

B. Document the promotion as an explicit forecast assumption

C. Replace all actual demand with the highest previous demand

D. Use MAPE without checking for stockouts

Correct Answer: B. Document the promotion as an explicit forecast assumption