Demand Management and Forecast Integration

Introduction and Context

Master Production Scheduling (MPS) translates aggregate production intentions into time-phased decisions for specific end items, configurations, or product families. Its quality depends fundamentally on the quality and governance of demand information. Demand management is therefore not merely forecasting. It is the disciplined process of identifying demand sources, assessing their credibility, integrating them into a coherent demand plan, and communicating the resulting requirements to production, procurement, distribution, and customer-facing functions.

An MPS commonly receives demand from several sources: statistical forecasts, confirmed customer orders, promotional uplifts, interplant requirements, service parts, safety-stock policies, and occasionally strategic or contractual commitments. These sources differ in certainty, timing, granularity, and ownership. If they are combined without clear rules, the organization may double count demand, inflate the schedule, or conceal genuine capacity requirements.

The central managerial challenge is to preserve both responsiveness and realism. Customer orders must be respected, but forecasts are needed for capacity and material planning before orders arrive. Promotions may create temporary peaks, while service parts may remain significant after a product has been discontinued. A sound demand-management process integrates these signals while distinguishing actual commitments from assumptions.

Learning Objectives

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

1. Explain the role of demand management in MPS.
2. Classify and integrate forecasts, orders, promotions, interplant demand, and service-part requirements.
3. Apply demand consumption rules without double counting.
4. Segment and aggregate demand at appropriate planning levels.
5. Disaggregate aggregate demand into product, location, and time requirements.
6. Calculate and interpret forecast-error measures.
7. Design governance controls for exceptions, overrides, and data quality.
8. Evaluate the managerial implications of demand uncertainty for the MPS.

1. Demand as an MPS Input

Demand is a requirement for a product, component, capacity resource, or service during a specified time interval. In MPS, the relevant question is not simply “How much might customers want?” but “What quantity should the master schedule treat as the requirement for each item and period?”

Demand signals can be classified according to certainty and origin:

The same physical requirement may appear in more than one source. For example, a customer order may already be included in the forecast, or an interplant order may have been generated from a downstream plant’s forecast. Integration therefore requires an explicit ownership and consumption policy.

Independent and Dependent Demand

Independent demand originates outside the production system, typically from customers or markets. Finished products and replacement parts usually have independent demand. Dependent demand is calculated from the need for another item, such as components derived from a bill of material. The MPS should generally be driven by independent demand; dependent requirements should then be generated through material requirements planning (MRP). Treating component forecasts as additional independent demand can create serious overstatement.

Demand Time Buckets

Demand is recorded in time buckets—days, weeks, or months. The bucket must match the decision horizon and operational lead times. Monthly buckets may be suitable for aggregate planning but can conceal a short-term capacity peak. Weekly buckets are common for MPS because they balance detail with stability. A demand date should be defined consistently: order-entry date, requested delivery date, confirmed delivery date, or production requirement date. Mixing these definitions produces misleading comparisons.

2. Forecast Integration and Demand Consumption

The Basic Consumption Principle

A forecast represents expected demand not yet observed as an order. Once an actual order is received, it should normally consume the forecast for the corresponding item, location, and time window. The purpose is to prevent the same requirement from being counted twice.

A simple period-level rule is:

where D_t is the demand used for planning in period t, F_t is the forecast, and O_t is the confirmed-order quantity. This rule is appropriate when orders are expected to be part of the forecast and the larger of the two values is a reasonable representation of total demand.

An alternative is explicit forecast consumption:

Here, C_t is the amount of forecast already consumed by earlier orders, R_t is residual forecast, and D_t is total planned demand. The equations assume that orders and forecasts are expressed in the same units, item definition, location, and time basis.

For a simple period in which orders are completely matched against the forecast, the rule becomes:

For example, if the forecast is 1,000 units and actual orders are 700 units, planned demand is 1,000 units, not 1,700. If orders are 1,250 units, planned demand is 1,250 units, assuming no separate promotion or strategic uplift.

Time-Phased and Order-Profile Consumption

Consumption should not always be restricted to the exact calendar period. A customer order scheduled for week 4 may have been represented by a forecast spread across weeks 3–5. A consumption window can therefore be defined, such as the forecast period containing the requested date plus adjacent periods. The window must be governed rather than adjusted informally.

Common rules include:

• exact-period consumption for short-cycle products;
• backward consumption from the requested delivery date within a defined frozen window;
• forward consumption when demand is delayed but remains valid;
• family-level consumption when forecasts exist only at family level;
• location-level consumption when the forecast is geographically aggregated.

Consumption requires a traceable link between the order and the forecast quantity it replaces. Without this link, planners cannot distinguish new demand from a change in timing, product mix, or customer identity.

Avoiding Double Counting

Double counting commonly occurs when:

1. orders are added to a forecast that already included them;
2. promotional uplift is added although it is embedded in the statistical forecast;
3. interplant orders are added to a forecast that already reflects downstream requirements;
4. service-part demand is included both in a replacement forecast and in an installed-base model;
5. aggregate family demand is added to detailed SKU demand;
6. dependent component requirements are forecast independently and then generated again by MRP.

A practical control is a demand lineage record. For every planning quantity, the system should identify source, owner, time bucket, item level, location, version, and whether the quantity is incremental or replacing another quantity.

3. Demand Segmentation and Aggregation

Segmentation

A single forecasting method rarely performs well across all products. Demand should be segmented according to characteristics relevant to the decision. Useful dimensions include:

• volume or annual usage;
• demand variability;
• intermittency and frequency of zero-demand periods;
• product life-cycle stage;
• customer criticality;
• margin or strategic importance;
• lead time and supply risk;
• forecastability;
• promotion sensitivity.

An ABC–XYZ approach is widely used. ABC classifies items by annual consumption value, while XYZ classifies them by variability or predictability. An AX item has high value and relatively stable demand; a CZ item has low value and irregular demand. These categories should influence review frequency, forecast method, safety-stock policy, and the degree of managerial attention.

Segmentation is not a substitute for judgment. A low-volume service part may be operationally critical even if its financial value is low. Conversely, a high-value item with a short life cycle may require a market-intelligence process rather than a stable time-series model.

Aggregation

Aggregation combines demand across products, locations, customers, or time periods. It reduces noise and supports capacity decisions. Forecast accuracy often improves at higher aggregation because positive and negative errors partially offset one another.

A valid aggregation requires consistent units and a stable mapping. If products have different sizes, equivalent units may be used, but the conversion factor must be documented. If the conversion for product i is w_i, aggregate equivalent demand is:

where q_i is the quantity of product i, w_i is its approved equivalent-unit factor, and Q is total equivalent demand. The formula is meaningful only when the weighting basis—machine hours, standard cost, volume, or another measure—is appropriate to the decision.

Aggregation can occur vertically or horizontally. Vertical aggregation combines SKUs into a product family; horizontal aggregation combines locations or customers. The selected level should match the production decision. Capacity may be planned at family level, while final assembly and material availability may require SKU-level demand.

Disaggregation

Disaggregation converts an aggregate plan into detailed requirements. Suppose aggregate family demand for period t is Q_t, and product i has a mix proportion p_i,t. Then:

where q_i,t is detailed demand and the mix proportions satisfy:

If Q_t = 10,000 units and the expected mix is 40% model A, 35% model B, and 25% model C, the disaggregated quantities are 4,000, 3,500, and 2,500 units. Rounding must be reconciled so that the detailed total remains 10,000 units.

Historical mix should not be used mechanically when promotions, price changes, supply constraints, or product substitution have altered customer choice. Scenario-based mix assumptions are preferable when uncertainty is material.

4. Promotions, Cannibalization, and Market Events

Promotional demand is usually incremental, but not always. A promotion may:

• create genuinely new demand;
• shift demand forward from later periods;
• shift demand between products;
• shift demand between channels;
• cannibalize a regular product;
• increase returns or service requirements.

A promotion record should specify baseline demand, incremental uplift, duration, product and channel scope, expected substitution, and post-promotion effects. A useful conceptual decomposition is:

where B_t is baseline demand, I_t is incremental demand, S_t is demand shifted into period t, and K_t is cannibalized or displaced demand. The variables are planning components, not necessarily independently measurable accounting quantities. Their use requires a documented business assumption.

Promotions should be integrated through a controlled override or event-based model. A planner should not simply replace the statistical forecast with an unsupported percentage increase. Post-event review is essential: actual uplift, timing, product mix, stock-outs, and residual demand should be compared with assumptions.

5. Interplant and Service-Part Demand

Interplant Demand

Interplant demand occurs when one plant supplies another plant, distribution center, or regional operation. The supplying plant should not automatically add an interplant order to a market forecast if the downstream requirement has already been represented there.

The preferred hierarchy is:

1. identify the ultimate external demand or independent requirement;
2. determine which facility owns the forecast;
3. convert the requirement into a transfer demand only where replenishment is needed;
4. account for transit time, minimum shipment quantities, and transfer losses;
5. prevent the receiving location’s forecast and the supplying location’s transfer order from being treated as two independent market demands.

Interplant requirements may legitimately be additive when they serve different purposes, such as safety-stock repositioning, pipeline fill, or a strategic inventory build. These additions should be explicitly coded rather than hidden in an order quantity.

Service Parts

Service-part demand has distinctive features: intermittent occurrence, long tails, installed-base dependence, product obsolescence, and high consequences of stockout. Forecasting may use failure rates, installed population, warranty records, repair history, and engineering changes.

A simple installed-base model is:

where E_t is expected part demand in period t, N_t is the relevant installed base, r_t is the expected failure or replacement rate per unit, and a_t is the average number of parts required per event. This is an expectation, not a guarantee; it is most useful when the installed base and rate assumptions are credible.

Service-part demand should be separated from new-equipment demand at the planning level, even where the same physical component is used. Separate ownership improves visibility of warranty, obsolescence, emergency orders, and service-level commitments.

6. Forecast-Error Analysis

Forecast evaluation should support decisions, not merely produce attractive statistics. Let actual demand in period t be A_t, and the forecast made for that period be F_t. Forecast error is commonly defined as:

A positive error under this convention means that actual demand exceeded the forecast.

Common Measures

Mean error, or bias, is:

Mean absolute error is:

Root mean squared error is:

Mean absolute percentage error is:

MAPE is undefined when A_t = 0 and unstable when actual demand is close to zero. For intermittent demand, MAE, RMSE, scaled errors, or measures based on forecast-observation pairs are often more appropriate. Weighted measures may be used when high-volume or high-value items deserve greater influence.

Worked Example

Suppose actual demand over three weeks is 100, 120, and 80 units, while forecasts are 90, 130, and 100 units. Errors are 10, −10, and −20 units. Therefore:

• ME = (10 − 10 − 20)/3 = −6.67 units;
• MAE = (10 + 10 + 20)/3 = 13.33 units;
• RMSE = √(100 + 100 + 400)/3 = √200 ≈ 14.14 units.

The negative mean error indicates overall overforecasting under the stated convention. MAE describes typical absolute deviation, while RMSE penalizes the larger error in week 3 more heavily.

Error metrics must be calculated on comparable forecast vintages. Comparing a forecast made one day before demand with one made three months earlier is not a fair test. Bias should also be investigated by segment, location, horizon, promotion status, and planner override.

7. Governance, Overrides, and Technology

Forecast integration is a cross-functional governance problem. Sales, marketing, operations, finance, supply chain, and service functions may hold different information and incentives. A demand-review process should define:

• the demand owner for each source;
• the forecast horizon and frozen zone;
• rules for overrides and approval thresholds;
• the definition of an order, forecast, promotion, and transfer;
• the calendar and time-zone convention;
• escalation rules for exceptional demand;
• audit trails for changed quantities;
• performance measures that distinguish forecast quality from supply failure.

A forecast should not be judged as inaccurate when the product was unavailable and observed sales were constrained. In such cases, recorded shipments may underestimate unconstrained demand. Lost-sales estimates and stockout flags are important explanatory variables.

Enterprise resource planning, advanced planning systems, customer relationship management, point-of-sale data, and machine-learning tools can improve integration. AI may identify nonlinear patterns, event effects, and product relationships, but it does not eliminate the need for data governance. Risks include data leakage, unstable models, opaque overrides, biased historical records, and false confidence in highly granular forecasts. Human review is particularly important for launches, discontinuations, promotions, crises, and structural market changes.

8. Illustrative Case: Orion Appliances

Background

Orion Appliances manufactures three refrigerator models in one plant and supplies two regional distribution centers. Its MPS team receives a statistical forecast from the planning system, confirmed retailer orders, promotional plans from marketing, interplant transfer orders, and service-part requirements.

Problem

For the next month, the system shows 12,000 units of family demand. Retailer orders total 7,500 units. Marketing has entered a 2,000-unit promotion uplift, but the statistical forecast already included the promotion campaign. In addition, the eastern distribution center has submitted a 1,500-unit transfer order based on the same retailer orders. The service organization has requested 300 compressor units.

A preliminary spreadsheet adds all quantities: 12,000 + 7,500 + 2,000 + 1,500 + 300. This produces 23,300 units, but the total is conceptually invalid because it mixes finished-appliance demand, duplicate transfer demand, and service-part demand.

Analysis

The 12,000-unit forecast includes the planned promotion. The 7,500 retailer orders consume part of that forecast rather than being added to it. Since the eastern transfer order is generated from the retailer orders, it is a logistics expression of existing demand, not new market demand. The 300 compressor units are a separate service-part requirement and should be planned under the service-parts item structure.

Assume the 12,000-unit family forecast is the accepted baseline and retailer orders are within the forecast quantity. Finished-appliance demand for the MPS is therefore 12,000 units, subject to capacity and mix review. Service-part demand is 300 compressor units, planned separately. If a documented review concludes that retailer orders exceed the forecast for the family, the excess would be added as an exception; it would not be added automatically in full.

Approach and Outcome

The team creates a demand-lineage table identifying forecast, order, promotion, transfer, and service-part sources. The promotion is retained as an event attribute rather than an additional quantity. The transfer order is linked to the downstream order and excluded from market-demand aggregation. Model mix is updated using the promotion-specific mix rather than the historical average. The MPS is released for 12,000 finished appliances, with a separate 300-unit service-part plan and an exception review for any order quantity outside the forecast-consumption window.

The outcome is a lower but more credible schedule. Capacity is still reviewed because the promotion changes the mix and may create a bottleneck in final testing. The case demonstrates that preventing double counting does not mean ignoring demand; it means representing each requirement once, at the correct planning level.

Learning Points

1. Demand sources must be linked to their underlying economic requirement.
2. Promotions can alter mix even when they do not increase total demand.
3. Interplant orders may be logistical duplicates of external demand.
4. Service parts require separate segmentation and ownership.
5. A credible MPS depends on traceability as much as on forecast accuracy.

9. Best-Practice Implementation Framework

A practical implementation can follow seven steps:

1. Define the demand architecture. Document items, locations, calendars, ownership, and source systems.
2. Classify demand. Separate independent, dependent, promotional, interplant, and service-part requirements.
3. Select the planning level. Decide where demand is forecast, where it is consumed, and where it is disaggregated.
4. Apply consumption rules. Use item, location, and time-window matching with auditable exceptions.
5. Review events and overrides. Record assumptions, responsible managers, and expiry dates.
6. Measure performance. Use bias and absolute-error measures appropriate to demand type and forecast horizon.
7. Close the learning loop. Compare assumptions with outcomes and update segmentation, mix, and governance rules.

The purpose is not to create a perfect forecast. It is to create a transparent, decision-relevant demand signal that the MPS can convert into feasible production commitments.

Challenges and Limitations

Demand integration remains difficult when data are delayed, product definitions change, customers place blanket orders, or supply shortages censor observed demand. Forecasts may also contain organizational bias: sales may overstate opportunities, finance may impose targets, and operations may smooth demand to fit capacity. Aggregation can hide mix risk, while disaggregation can create false precision. Error measures can reward a model that predicts average demand but fails during important events.

Managers should therefore interpret forecast outputs alongside market intelligence, capacity constraints, inventory position, lead times, and service consequences. A numerically sophisticated model cannot compensate for unclear ownership or duplicated data.

Chapter Summary

Demand management provides the foundation for a reliable Master Production Schedule. It integrates forecasts, customer orders, promotions, interplant requirements, and service parts while preserving the distinction between independent and dependent demand. Forecast consumption prevents orders already represented in a forecast from being counted again. Segmentation aligns methods and attention with volume, variability, criticality, and life-cycle conditions. Aggregation supports capacity planning, whereas disaggregation converts family or regional plans into executable item-level requirements.

Promotional and interplant demand must be traced to determine whether they are incremental or merely different representations of existing demand. Service parts require distinct models because installed-base size, failure rates, intermittency, and service commitments shape their requirements. Forecast-error measures such as bias, MAE, RMSE, and MAPE are useful only when their assumptions and limitations are understood. Finally, governance, auditability, cross-functional ownership, and responsible technology use are essential to prevent double counting and improve MPS quality.

Key Terms

Demand management: The coordinated process of identifying, evaluating, integrating, and governing demand signals.

Independent demand: Demand originating from external customers or markets.

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

Master Production Schedule: A time-phased plan for producing specific items or product families.

Forecast consumption: The replacement of forecast quantities by, or offsetting of, actual orders within a defined rule.

Demand lineage: Traceability showing the source, owner, level, timing, and treatment of a demand quantity.

Demand segmentation: Classification of demand according to characteristics such as value, variability, or criticality.

ABC analysis: Classification based commonly on annual usage value.

XYZ analysis: Classification based on demand variability or predictability.

Aggregation: Combining demand across products, locations, customers, or periods.

Disaggregation: Converting aggregate demand into detailed requirements.

Promotion uplift: Demand attributed to a marketing event beyond an established baseline.

Cannibalization: Loss of demand for one product caused by another product or promotion.

Interplant demand: Requirement for movement or replenishment between facilities.

Service-part demand: Requirement for replacement or maintenance parts.

Forecast bias: Systematic tendency to overforecast or underforecast.

MAE: Mean absolute error between actual and forecast demand.

RMSE: Root mean squared forecast error.

Frozen zone: Near-term scheduling period in which changes are restricted.

Double counting: Including the same underlying requirement more than once.

Review Questions

Short Answer Questions

1. Why should confirmed customer orders normally consume, rather than be added to, an existing forecast?
2. Distinguish independent demand from dependent demand.
3. What is the purpose of demand segmentation?
4. State one limitation of MAPE.
5. Why should service-part demand usually be managed separately from new-product demand?

Long Answer Questions

1. Explain how a demand-lineage framework can prevent double counting across forecasts, orders, promotions, and interplant requirements.
2. Discuss the managerial purposes of aggregation and disaggregation in an MPS process.
3. Compare MAE, RMSE, and bias as forecast-performance measures, including their interpretation and limitations.
4. Explain how promotions can create incremental demand, time shifting, product substitution, and cannibalization.
5. Design a cross-functional governance process for forecast overrides and demand review.

Analytical / Application Questions

1. A product has a forecast of 4,000 units and confirmed orders of 3,200 units. The orders are known to be included in the forecast. What quantity should normally enter the MPS, and what additional information would justify a different quantity?
2. A family forecast is 18,000 units. Expected mix is 50% model A, 30% model B, and 20% model C. Calculate the disaggregated quantities and explain how you would handle a later mix change caused by a promotion.
3. A company adds customer orders, promotional uplift, and interplant transfers to its statistical forecast. Develop a diagnostic procedure to determine which quantities are genuinely incremental and which are duplicate representations.

Multiple Choice Questions with Answers

Question 1

A forecast for a product is 1,000 units and confirmed orders are 700 units. The orders are known to be included in the forecast. What quantity should normally be used for planning?

A. 300 units

B. 700 units

C. 1,000 units

D. 1,700 units

Correct Answer: C. 1,000 units

Question 2

Which demand type is normally generated from a bill of material after the MPS is established?

A. Independent demand

B. Dependent demand

C. Promotional demand

D. Interplant demand

Correct Answer: B. Dependent demand

Question 3

What is the principal purpose of forecast consumption?

A. To increase the forecast systematically

B. To prevent the same requirement being counted twice

C. To eliminate all forecast uncertainty

D. To replace the MPS with customer orders

Correct Answer: B. To prevent the same requirement being counted twice

Question 4

A family forecast is 10,000 units and model B represents 35% of the expected mix. What is model B’s disaggregated quantity?

A. 350 units

B. 3,000 units

C. 3,500 units

D. 6,500 units

Correct Answer: C. 3,500 units

Question 5

Under the convention e_t = A_t − F_t, a negative mean error indicates that the organization has generally:

A. Underforecast demand

B. Overforecast demand

C. Eliminated forecast error

D. Measured only promotional demand

Correct Answer: B. Overforecast demand

Question 6

Which measure is especially problematic when actual demand is zero?

A. MAE

B. RMSE

C. Mean error

D. MAPE

Correct Answer: D. MAPE

Question 7

An interplant transfer order is generated entirely from retailer orders already included in the market forecast. How should it normally be treated?

A. As new external demand

B. As a duplicate representation of existing demand

C. As dependent component demand

D. As service-part demand

Correct Answer: B. As a duplicate representation of existing demand

Question 8

In an installed-base service-parts model, expected demand is calculated as installed base × failure rate × parts per event. What does the result represent?

A. A guaranteed order quantity

B. An expected requirement based on assumptions

C. A capacity limit

D. A forecast-error measure

Correct Answer: B. An expected requirement based on assumptions

Question 9

Which classification combines value-based and variability-based segmentation?

A. ABC–XYZ

B. FIFO–LIFO

C. EOQ–ROP

D. MRP–ERP

Correct Answer: A. ABC–XYZ

Question 10

Why can aggregation improve forecast accuracy?

A. It guarantees that promotions are predicted correctly

B. It removes the need for disaggregation

C. Positive and negative errors may partially offset

D. It converts independent demand into dependent demand

Correct Answer: C. Positive and negative errors may partially offset