2. Demand, Data and Planning Information

Introduction: Why Planning Begins with Demand

Supply chain planning is often described as the coordination of supply, inventory, capacity, transportation, and customer service. Yet each of these activities is shaped by an initial question: what demand is the supply chain expected to satisfy? The answer is not found simply by copying historical sales. Demand must be defined, classified, measured, interpreted, and converted into planning information.

For a logistics manager, demand is both an economic phenomenon and a data problem. Customers may change their purchasing behaviour because of price, promotions, seasonality, competition, weather, regulation, or macroeconomic conditions. At the same time, recorded sales may be distorted by stockouts, delayed transactions, product substitutions, incorrect master data, or changes in the distribution network. A forecast based on such data can be mathematically precise and operationally wrong.

This chapter establishes the information foundation for supply chain planning. It distinguishes independent from dependent demand, examines demand patterns and drivers, and presents a disciplined approach to preparing master data. Particular attention is given to data cleansing, outlier treatment, segmentation, visualization, and data-quality controls. These activities are not merely technical preliminaries. They influence inventory investment, production decisions, supplier commitments, transport capacity, and the credibility of planning decisions.

Learning Objectives

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

1. Distinguish independent, dependent, derived, and intermittent demand.

2. Explain the principal patterns and drivers of demand.

3. Identify the difference between transactional data, master data, and planning parameters.

4. Design a practical process for cleansing and validating supply chain data.

5. Diagnose and treat outliers without removing meaningful business signals.

6. Segment products, customers, and demand streams for planning purposes.

7. Use visualization to reveal demand behaviour and data problems.

8. Establish data-quality controls and governance responsibilities.

9. Evaluate the managerial and ethical consequences of poor planning information.

Demand as a Planning Construct

Demand is the requirement for a product, service, component, or logistics capability during a specified time and under specified conditions. It is not identical to sales. Sales represent transactions recorded by an organization; demand represents what customers wanted or required. If ten units were requested but only six were available, recorded sales may be six while demand was ten. The difference is known as lost sales, unmet demand, or censored demand, depending on the data and analytical context.

A useful planning representation is:

Observed sales = realized demand subject to availability, price, channel, and recording conditions.

This distinction matters because a forecasting system that treats every low-sales period as low demand may perpetuate understocking. Conversely, an exceptional order caused by a one-time project should not automatically become the baseline for future replenishment.

Independent and Dependent Demand

Independent demand originates outside the production or distribution system. It is associated with finished goods, spare parts sold directly to customers, retail items, or services. It is influenced by market behaviour and normally requires forecasting or customer-order analysis.

Dependent demand is derived from the demand for another item. If a finished product requires four wheels, demand for wheels depends on the planned production of finished products. Dependent demand is normally calculated through the bill of materials, production schedule, usage rates, and inventory records rather than forecast independently.

For example, a distributor forecasts demand for 1,000 water pumps. A manufacturer producing each pump with two seals will calculate dependent demand for 2,000 seals, adjusted for scrap, inventory, and planned production. Forecasting seals independently could create inconsistencies between component supply and finished-goods plans.

Some requirements occupy an intermediate position. Packaging, pallets, labels, and transport capacity may be derived from orders but also affected by operating policies and shipment consolidation. Such requirements are often called derived demand. Intermittent demand refers to demand with many zero periods and occasional transactions, as in slow-moving spare parts. It requires methods and performance measures different from those used for fast-moving retail items.

Demand, Orders, Shipments, and Consumption

A planning team must specify which event is being measured. Customer orders indicate intention or commitment; shipments indicate physical dispatch; invoices indicate commercial recognition; point-of-sale data indicate retail movement; and consumption data indicate actual usage. These measures may differ in timing and quantity.

A wholesaler may place a large monthly order even though its customers consume products steadily. If the manufacturer forecasts from wholesale orders without recognizing ordering cycles, it may interpret periodic peaks as true market seasonality. The appropriate data level therefore depends on the planning decision. Distribution-centre replenishment may require shipment and inventory data, while manufacturing capacity planning may benefit from customer consumption data.

Patterns and Drivers of Demand

Demand analysis begins with pattern recognition. A time series can be examined through level, trend, seasonality, cyclicality, variability, and intermittency.

Level is the typical magnitude around which demand varies. Trend is a sustained upward or downward movement. Seasonality is a systematic pattern repeating at a known interval, such as weekly, monthly, or annually. Cycles are longer and less regular fluctuations associated with economic or industry conditions. Variability describes the degree of irregular movement. Intermittency describes the frequency of zero-demand periods.

These features may occur together. A product can have a growing trend, strong seasonal peaks, and intermittent demand. A visual plot is often the first practical way to distinguish them.

Internal and External Drivers

Demand drivers may be classified as internal or external. Internal drivers include price, promotions, assortment, sales-force activity, service levels, product availability, channel policies, and delivery promises. External drivers include competitor actions, customer income, weather, demographic change, regulation, technology, construction activity, and geopolitical conditions.

Promotions require particular care. A promotional spike may represent demand pulled forward from future periods rather than genuinely additional demand. Price changes may alter both quantity and timing. Product launches may have little historical data, while product discontinuations create artificial declines. Cannibalization occurs when a new product takes demand from an existing product within the same portfolio.

Availability is also a demand driver in the observed data. Repeated stockouts suppress sales and create a misleading history. A robust analyst records stockout dates and distinguishes potential demand from fulfilled demand wherever possible.

Planning Information: Three Data Layers

Supply chain planning relies on three related data layers.

Transactional data records events: orders, sales, receipts, shipments, inventory movements, returns, production completions, and purchase orders. It answers questions such as what happened, when, where, and in what quantity.

Master data describes relatively stable entities: item codes, descriptions, units of measure, locations, suppliers, customers, bills of material, lead times, pack sizes, storage conditions, and product hierarchies. It provides the structure within which transactions are interpreted.

Planning parameters convert operational knowledge into decision rules. Examples include safety stock, reorder points, minimum order quantities, order multiples, review periods, service targets, forecast calendars, and lead-time variability.

Master data is not necessarily permanent. Lead times change, suppliers are replaced, products are repackaged, and distribution networks are redesigned. It should therefore be governed as a controlled organizational asset rather than treated as a one-time database upload.

Preparing a Reliable Demand Dataset

A sound preparation process begins by defining the purpose, scope, grain, and time horizon of the dataset. The grain specifies what one record represents—for example, one SKU-location-day or one product-category-month. Mixing grains in one table is a common source of double counting.

A practical workflow contains the following stages:

1. Define business questions. Establish whether the dataset supports forecasting, inventory policy, capacity planning, supplier collaboration, or performance analysis.

2. Inventory data sources. Identify enterprise-resource-planning systems, warehouse systems, point-of-sale feeds, transport platforms, spreadsheets, supplier files, and external data.

3. Create a data dictionary. Document field names, definitions, units, permitted values, owners, and update frequency.

4. Standardize identifiers. Resolve duplicate SKU codes, inconsistent location names, changing customer identifiers, and obsolete records.

5. Validate structural integrity. Check data types, dates, quantities, currencies, units of measure, and relationships between tables.

6. Reconcile totals. Compare extracts with inventory, finance, order-management, and warehouse-control totals.

7. Investigate anomalies. Examine missing values, duplicates, negative quantities, impossible dates, sudden changes, and stockout periods.

8. Approve and document transformations. Every adjustment should be traceable, justified, and reversible where practical.

Cleansing Master Data

Data cleansing is the identification and correction of inaccurate, incomplete, inconsistent, or unusable records. Typical controls include:

• uniqueness checks for item and location codes;
• completeness checks for essential fields such as unit of measure and lead time;
• validity checks against approved code lists;
• range checks for dimensions, weights, prices, and lead times;
• consistency checks between pack size, case quantity, and order multiples;
• referential checks ensuring that each transaction links to a valid item and location;
• duplicate detection using combinations such as description, supplier, size, and barcode.

A missing lead time is not equivalent to a zero lead time. A blank unit of measure is not equivalent to “each.” Such distinctions must be preserved because default values can create serious planning errors. Where a value is estimated, it should be marked as estimated and assigned an owner for confirmation.

Outliers: Error, Event, or Signal?

An outlier is an observation that differs substantially from the prevailing pattern. Statistical rules such as interquartile ranges, standard deviations, or median-based measures can identify candidates, but they cannot determine the business meaning of an observation.

A very large order may be a data-entry error, a genuine project, a promotion, a customer’s forward-buying decision, or a one-time emergency. Each case requires a different treatment:

• correct the record if it is demonstrably erroneous;
• retain and model it if it is likely to recur;
• annotate it as an exceptional event if it is genuine but nonrecurring;
• allocate it across periods if the timing is administrative rather than economic;
• replace it with an informed estimate only when the analytical purpose requires a baseline series.

Blind winsorization or deletion can be dangerous. Removing genuine peaks understates capacity and inventory requirements; retaining errors inflates them. A good practice is to maintain both the raw value and an adjusted analytical value, together with an event code and explanation.

Segmentation for Planning

Not every demand stream deserves the same forecasting method, review frequency, or service policy. Segmentation converts a large portfolio into manageable planning groups.

The classic ABC approach classifies items by annual consumption value, often calculated as annual quantity multiplied by unit cost. A-items require close managerial attention because they represent a large share of value, although they may be few in number. B- and C-items generally receive progressively simpler controls.

ABC alone is insufficient. An inexpensive component can be operationally critical, while a high-value item may be easy to replenish. A second dimension may classify criticality, supply risk, demand variability, or lead-time uncertainty. The resulting matrix supports differentiated policies such as tight review for high-value/high-risk items and simplified rules for low-value/low-risk items.

Demand-pattern segmentation can distinguish smooth, erratic, intermittent, and lumpy demand. Product life-cycle status—introduction, growth, maturity, decline, or discontinuation—also affects method selection. Segmentation should be revisited when assortment, sourcing, or market conditions change.

Visualization as an Analytical Discipline

Visualization is not decoration; it is a method of inquiry. A line chart of demand over time can reveal trend, seasonality, stockouts, calendar effects, and structural breaks. A heat map can show demand by weekday and hour. A scatter plot can relate demand to price or promotion. A histogram can reveal skewness and unusual values. A box plot can compare variability across products or locations.

Useful planning dashboards normally combine demand with explanatory variables. A sales chart without inventory availability may misrepresent demand. A forecast-accuracy chart without volume may give equal importance to trivial and strategically important items. Visualizations should therefore provide context, allow drill-down to transactions, and distinguish actual, adjusted, forecast, and exception periods.

Data-Quality Controls and Governance

Data quality has several dimensions: accuracy, completeness, consistency, timeliness, validity, uniqueness, and fitness for purpose. A dataset can be accurate but too old to support daily replenishment, or complete but inconsistent across systems.

Controls should operate at three levels. Preventive controls restrict invalid entries at source. Detective controls identify problems through validation reports and exception dashboards. Corrective controls assign ownership, resolve the issue, and record the action taken.

Governance requires named data owners, data stewards, approval rules, version control, audit trails, and escalation procedures. A planning team should establish service-level expectations for master-data correction and monitor indicators such as percentage of items with valid lead times, duplicate rate, missing-unit rate, and timeliness of updates.

Technology, Analytics, and AI in Demand Information

Enterprise resource planning, warehouse management, customer relationship management, point-of-sale, and transport systems provide different views of the supply chain. Integration platforms and data warehouses can create a common planning layer, while master-data-management tools control identifiers and hierarchies.

Analytics can automate duplicate detection, anomaly screening, missing-value alerts, and segmentation. Machine-learning systems may identify nonlinear relationships among price, promotion, weather, and demand. However, automation does not eliminate the need for business judgment. An algorithm can identify an unusual observation but may not know that a major customer opened a temporary project site. Models should therefore expose drivers, confidence levels, data lineage, and override reasons.

AI also introduces risks. Biased historical data may reproduce inequitable allocation decisions. Excessive reliance on opaque recommendations may weaken accountability. Access controls are necessary where data includes customer, supplier, or commercially sensitive information. Human review is particularly important for exceptional events and high-impact decisions.

Illustrative Case: Preparing Demand Information for a Spare-Parts Network

Background

A fictional industrial-equipment distributor, Northstar Service Logistics, manages 18,000 spare parts across four regional warehouses. Its planning system produces frequent stockouts for some parts while excess inventory accumulates for others. Management initially attributes the problem to poor forecasting.

Situation and Analysis

An investigation finds that the demand file combines customer orders, inter-warehouse transfers, and warranty replacements. Several parts have duplicate codes following a supplier change. Lead times are missing for 22 percent of items, and zero sales are recorded during periods when the parts were unavailable. A small number of emergency project orders create extreme peaks.

The team defines the planning grain as item-location-week and separates external customer demand from internal transfers. It builds a data dictionary, maps old and new item codes, and marks stockout weeks. Demand is segmented using annual consumption value, criticality, and intermittency. Emergency project orders are retained in the raw dataset but flagged separately for baseline analysis.

Approach and Solution

For fast-moving parts, the team uses cleaned customer-consumption history and reviews promotional or contractual effects. For intermittent parts, it uses an approach suited to sporadic demand and sets service policies based on equipment criticality rather than value alone. Missing lead times are replaced only temporarily with approved estimates and a steward is assigned to verify them with suppliers.

A dashboard displays stockout-adjusted demand, inventory position, lead-time status, duplicate records, and unresolved exceptions. Every adjustment retains the original transaction and an explanation. The planning meeting reviews data exceptions before discussing forecast overrides.

Outcome

The organization does not immediately claim that forecasting accuracy has improved; instead, it first improves the reliability of the planning base. Stockouts are no longer interpreted automatically as low demand, transfers are not counted as external consumption, and critical low-value parts receive appropriate attention. Inventory policy becomes differentiated rather than uniform.

Learning Points

The case demonstrates that forecast performance cannot be separated from data definition. Data preparation is a cross-functional responsibility involving service, procurement, warehousing, finance, IT, and planning. It also shows why the correct response to an anomaly is investigation, not automatic deletion.

Challenges, Risks, and Professional Judgement

The main challenge is that data problems often reflect process problems. If warehouse workers scan late, if sales representatives create duplicate customers, or if procurement does not update supplier lead times, the planning database will reproduce those weaknesses.

Common mistakes include forecasting dependent demand independently, treating sales as demand without checking availability, aggregating incompatible units, ignoring returns, deleting outliers without investigation, using one segmentation scheme for every decision, and applying default values silently. Another error is measuring data quality only through technical completeness. A fully populated field can still contain a wrong value.

Managers must also recognize organizational incentives. A sales team may overstate opportunities, a warehouse may delay recording shortages, and a supplier may report optimistic lead times. Governance should therefore encourage transparent exception reporting rather than punish the discovery of bad data.

Privacy, confidentiality, and regulatory obligations apply when demand data identifies customers, patients, households, or individual purchasing behaviour. Access should follow least-privilege principles, retention should be justified, and personally identifiable information should be anonymized or pseudonymized where it is not required for planning.

Best Practices for Planning Information

Effective organizations generally:

• define demand and data grain before selecting a forecasting method;
• preserve raw data while documenting every transformation;
• separate observed sales from estimated unconstrained demand;
• combine statistical diagnostics with operational investigation;
• segment by value, variability, criticality, and life-cycle context;
• assign data ownership at the source, not only in the planning department;
• use exception-based workflows rather than manual inspection of every record;
• reconcile planning data with financial and physical controls;
• review master data after product, supplier, network, or policy changes;
• measure the quality and usefulness of information continuously.

Chapter Summary

Demand is a planning construct that must be distinguished from recorded sales, orders, shipments, and consumption. Independent demand comes from external market requirements, whereas dependent demand is derived from a parent item, production plan, or bill of material. Demand may display level, trend, seasonality, cycles, variability, and intermittency, and it is shaped by both internal and external drivers.

Reliable planning depends on the integration of transactional data, master data, and planning parameters. Data preparation requires a defined grain, a data dictionary, standardized identifiers, structural validation, reconciliation, anomaly investigation, and documented transformations. Outliers must be interpreted rather than mechanically removed. Segmentation by value, criticality, variability, and life-cycle stage supports differentiated planning policies. Visualization reveals patterns and data defects, while governance ensures accountability, traceability, and continuous quality control. Digital tools and AI can strengthen these activities, but human judgement, ethical safeguards, and business context remain essential.

Key Terms

Independent demand: Demand originating from external customers or markets and normally requiring forecasting.

Dependent demand: Demand calculated from the requirement for another item or production plan.

Derived demand: Demand generated indirectly by another activity, such as packaging or transport requirements.

Intermittent demand: Demand with many zero periods and occasional nonzero transactions.

Observed sales: Recorded transactions, which may be constrained by availability or other conditions.

Unmet demand: Customer requirement that could not be fulfilled.

Master data: Stable descriptive data about items, locations, suppliers, customers, and relationships.

Transactional data: Records of operational events such as orders, shipments, receipts, and sales.

Planning parameters: Values used to operate planning rules, including lead times, safety stocks, and order quantities.

Data grain: The precise level represented by one record in a dataset.

Data cleansing: Identification and correction of inaccurate, incomplete, inconsistent, or unusable data.

Outlier: An observation that differs substantially from the prevailing pattern.

Stockout: A condition in which required inventory is unavailable.

ABC analysis: Segmentation based generally on annual consumption value.

Criticality: The operational consequence of an item’s unavailability.

Seasonality: A recurring demand pattern at a predictable interval.

Data lineage: The documented origin, movement, and transformation history of data.

Data steward: A person responsible for maintaining data quality within an assigned domain.

Review Questions

Short Answer Questions

1. What is the difference between independent and dependent demand?

2. Why may observed sales be lower than actual demand?

3. Define master data and give three examples.

4. What is meant by the grain of a dataset?

5. Name four dimensions of data quality.

Long Answer Questions

1. Explain the major patterns of demand and discuss how each affects planning.

2. Describe a systematic workflow for preparing a demand dataset.

3. Discuss why outliers should be investigated before they are removed or adjusted.

4. Explain how ABC, criticality, variability, and life-cycle segmentation can be combined.

5. Examine the role of data governance in supply chain planning.

Analytical / Application Questions

1. A product shows low sales during three months, but inventory records indicate repeated stockouts. How should the planning team interpret and prepare the demand history?

2. A manufacturer forecasts finished-product demand and separately forecasts its components. Analyze the risks and propose a better planning approach.

3. A large order causes a demand spike, but the customer confirms that it is a one-time infrastructure project. Explain how the event should be represented in raw data, baseline analysis, and future planning.

22. Multiple Choice Questions with Answers

Question 1

Which situation most clearly represents dependent demand?

A. Customer purchases of finished bicycles

B. Retail demand for bottled water

C. Demand for tires calculated from the planned production of vehicles

D. Orders for replacement laptops from corporate customers

Correct Answer: C. Demand for tires calculated from the planned production of vehicles

Question 2

Why can recorded sales understate true demand?

A. Because all sales records include safety stock

B. Because stockouts may prevent some customer requirements from being fulfilled

C. Because dependent demand is always added to sales

D. Because seasonal demand is removed from transaction files

Correct Answer: B. Because stockouts may prevent some customer requirements from being fulfilled

Question 3

A dataset contains one record per SKU-location-day, but monthly totals are appended without a flag. What is the principal risk?

A. Improved seasonality measurement

B. Reduced lead-time variability

C. Double counting caused by mixed data grain

D. Automatic correction of missing values

Correct Answer: C. Double counting caused by mixed data grain

Question 4

What is the most appropriate first response to an unusually large order?

A. Delete it immediately

B. Replace it with the median without investigation

C. Investigate whether it is an error, recurring event, or one-time requirement

D. Use it as the permanent forecast baseline

Correct Answer: C. Investigate whether it is an error, recurring event, or one-time requirement

Question 5

Which item would most likely require a specialized intermittent-demand approach?

A. A grocery item sold every hour

B. A spare part with many zero-demand weeks

C. A promotional item with predictable daily sales

D. A product with stable monthly consumption

Correct Answer: B. A spare part with many zero-demand weeks

Question 6

Which example is master data rather than transactional data?

A. A shipment completed yesterday

B. A customer order entered this morning

C. The approved unit of measure for an item

D. A warehouse receipt posted last week

Correct Answer: C. The approved unit of measure for an item

Question 7

What does ABC analysis primarily classify?

A. Items by annual consumption value

B. Suppliers by geographic distance

C. Customers by payment behaviour

D. Orders by delivery sequence

Correct Answer: A. Items by annual consumption value

Question 8

A planner removes all high-demand observations using a statistical outlier rule. What is the main danger?

A. The dataset will contain too many identifiers

B. Genuine peaks and capacity requirements may be understated

C. Dependent demand will be calculated automatically

D. Data lineage will become unnecessary

Correct Answer: B. Genuine peaks and capacity requirements may be understated

Question 9

Which control checks whether every transaction is linked to a valid item code?

A. Referential integrity check

B. Seasonality check

C. Promotion effectiveness check

D. Forecast bias check

Correct Answer: A. Referential integrity check

Question 10

Why should an AI anomaly-detection result normally be reviewed by a business expert?

A. Algorithms cannot process numerical data

B. Business context is needed to distinguish errors from genuine exceptional events

C. Human review removes the need for data governance

D. Anomaly detection only applies to master data

Correct Answer: B. Business context is needed to distinguish errors from genuine exceptional events