Network Data, Cost-to-Serve and Economic Analysis

Introduction

Distribution-network configuration is only as sound as the data and economics on which it rests. A network model may contain sophisticated optimisation algorithms, yet produce poor decisions if demand is overstated, transport flows are duplicated, capacities are unavailable, taxes are misclassified, or service requirements are treated as averages rather than constraints. The purpose of this chapter is to establish an auditable economic and data foundation for network design.

A distribution network connects supply points, plants, ports, distribution centres, cross-docks, customers and reverse-logistics locations. Configuration decisions determine which facilities exist, which customers they serve, how products flow, how much inventory is held, and what service promise can be achieved. These decisions are usually long-lived and involve both fixed and variable costs. Consequently, a network model must represent operational reality while remaining sufficiently transparent for managerial challenge.

The central analytical principle is that network cost must be evaluated at the level of the relevant decision. A low-cost transport lane may increase inventory, handling, taxes or customer penalties. Conversely, a more expensive facility may reduce lead time, improve resilience and enable revenue that is not visible in a narrow freight calculation.

Learning objectives

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

1. Design a structured network data model covering demand, products, locations, flows, capacities and service levels.
2. Distinguish landed cost, cost-to-serve, capital expenditure and operating cost.
3. Calculate inventory carrying cost, taxes and total network cost using consistent units and assumptions.
4. Evaluate trade-offs among cost, service, capacity, risk and working capital.
5. Assess data quality, uncertainty, model boundaries and governance controls.
6. Interpret economic-analysis results for investment and configuration decisions.

1. The network data model

A network model is a representation of entities, relationships, attributes and constraints. It should be designed as a controlled decision model rather than as a collection of spreadsheets. The basic dimensions are shown below.

1.1 Demand representation

Demand should be recorded at the finest level needed for the decision. A monthly national total may be sufficient for a preliminary footprint study but inadequate for delivery-frequency, route, inventory or service analysis. A useful demand record identifies customer or zone, SKU or product family, time period, quantity, units, value and demand status.

Observed sales are not always equal to unconstrained demand. Stockouts, allocation rules, promotions, lost customers and substitutions can cause recorded sales to understate actual demand. Conversely, pipeline orders or promotional forecasts may overstate sustainable demand. Analysts should distinguish:

• historical shipments;
• customer orders;
• unconstrained demand estimates;
• approved forecasts; and
• scenario demand.

Seasonality and variability matter. For period demand D_t, a simple average over n periods is:

where D̄ is average demand per period and D_t is demand in period t. The formula assumes comparable periods and a consistent unit of measure. If demand is strongly seasonal, the average alone is insufficient; seasonal indices or monthly profiles should be retained.

1.2 Flow and material balance

For each product and period, the model should conserve material. If I^begin_t and I^end_t are beginning and ending inventory, Qⁱn_t is inbound quantity and Q^out_t is outbound quantity, then:

This identity applies when losses, returns and transformations are either absent or represented separately. Shrinkage, scrap, returns and product conversion must not be hidden inside an unexplained balancing adjustment.

A flow record normally includes origin, destination, product, period, quantity, mode, shipment frequency, distance, transit time and responsible party. The model should avoid double counting: a supplier-to-warehouse movement and warehouse-to-customer movement are separate legs, while a consolidated shipment must not also be counted as its individual components unless the accounting structure explicitly requires it.

1.3 Locations, capacities and service levels

Locations should be coded consistently and classified by function: supplier, plant, port, regional distribution centre, fulfilment centre, customer, return centre or cross-dock. Each location requires a time zone, calendar, operating hours, country, tax jurisdiction and valid coordinates. Geocoding errors can distort both distance and service calculations.

Capacity is multidimensional. A facility may have adequate annual throughput but insufficient peak-week labour, dock doors, cold storage, hazardous-goods space or order-picking capacity. Capacity fields should therefore distinguish:

• storage capacity, usually pallets, cases, cubic metres or square metres;
• throughput capacity, such as cases per day;
• handling capacity, such as lines per labour hour;
• dock and yard capacity;
• transport or supplier capacity; and
• peak and ordinary capacity.

Service levels must be operationally defined. “95% service” might mean order fill rate, on-time delivery, case fill, cycle-service probability or product availability. These are not interchangeable. A service record should specify the metric, measurement window, customer segment, product scope and penalty or consequence of failure.

2. Cost architecture

2.1 Landed cost

Landed cost is the cost of acquiring and placing a product at a specified receiving point. A general expression is:

where:

• LC = landed cost per unit;
• P = purchase or production cost per unit;
• F_in = inbound freight and transport accessorials per unit;
• H_in = origin and destination handling per unit;
• D = duties and customs charges per unit;
• T = non-recoverable taxes per unit;
• I = insurance per unit; and
• O = other directly attributable costs per unit.

The formula assumes every component is measured on the same unit basis and that recoverable taxes are excluded from economic cost. In accounting practice, treatment may differ by jurisdiction and inventory policy; the model must therefore state whether it is calculating cash cost, accounting inventory cost or managerial economic cost.

If freight is quoted for a shipment rather than a unit, allocation may be based on weight, volume, value, pallets or activity. The chosen basis should reflect the cost driver. Allocating a container charge by unit can be misleading when products differ substantially in cube or weight.

2.2 Cost-to-serve

Cost-to-serve measures the cost of fulfilling a customer, channel, product or order after the relevant supply and distribution activities are included. A useful activity-based expression is:

where CTS_j is cost-to-serve for customer or segment j, and the terms represent procurement, inbound, warehousing, outbound, inventory, order-processing, returns and special-service costs. The formula is a managerial framework, not a universal accounting standard. Shared costs should be allocated only when the allocation has decision relevance; otherwise they should be displayed separately.

Contribution after service cost can be stated as:

where R_j is revenue and COGS_j is cost of goods sold. A customer with high revenue may generate low or negative contribution after expedited deliveries, small-order handling, returns and special packaging are recognised.

2.3 Operating cost and capital expenditure

Operating expenditure (OPEX) is incurred to run the network. It includes labour, rent, utilities, maintenance, information systems, security, routine handling, transport, insurance and management overhead. Capital expenditure (CAPEX) creates or improves an asset, such as land, buildings, racking, automation, material-handling equipment or systems implementation.

A project’s annual economic cost should not compare one-time CAPEX with one year of OPEX without a time basis. A simplified annualised capital charge is:

where ACC is annual capital charge and r is the annual capital-recovery or hurdle rate. This is a simplified screening method and does not replace discounted-cash-flow analysis. If asset life and residual value are known, an annuity-equivalent annual cost can be calculated as:

where N is asset life in years and RV is end-of-life residual value. The formula assumes constant annual payments, a constant discount rate and cash flows at period end. Taxes, inflation, maintenance escalation and depreciation effects require additional treatment.

Total annual network cost may be structured as:

The categories must be mutually exclusive and collectively appropriate to the decision. A cost included in transport must not also appear in facility handling.

2.4 Inventory carrying cost

Inventory carrying cost represents the economic cost of holding inventory. For average inventory value V_avg and annual carrying-rate h,:

The carrying rate may include cost of capital, insurance, shrinkage, obsolescence, storage and taxes on inventory. If these items are separately included elsewhere, they must be removed from h to prevent double counting.

For average quantity Q_avg and unit value v,:

If demand is stable, replenishment is instantaneous and cycle stock is the only variable, a simplified cycle-stock quantity is Q/2, where Q is order quantity. Safety stock, pipeline stock, seasonal stock and display stock must be added when relevant. The familiar economic order quantity under restrictive assumptions is:

where D is annual demand in units, S is ordering cost per order and H is annual holding cost per unit. It assumes constant demand, known parameters, instantaneous replenishment, no quantity discounts and no stockouts. It is not a substitute for a network inventory model when lead times, service constraints or multi-echelon effects dominate.

2.5 Taxes and trade costs

Taxes require careful classification. Customs duty may be calculated as a percentage of a customs value:

Import VAT or sales tax may be recoverable and therefore excluded from economic cost, although it still affects cash flow and working capital. Excise taxes, non-recoverable environmental levies and brokerage charges may be true costs. Incoterms determine which party bears freight, insurance, import clearance and risk, but they do not by themselves determine accounting treatment.

A cross-border model should document product classification, country of origin, valuation basis, tariff rate, tax recovery, currency, exchange-rate date and compliance assumptions. Trade rules change, and a model should include a review date rather than treating tax parameters as permanent facts.

3. Economic trade-offs and decision analysis

Network decisions are multi-objective. The lowest freight solution may require more inventory; the lowest facility-cost solution may increase delivery distance; a single national centre may reduce fixed cost but increase disruption exposure. Important trade-offs include:

• transport cost versus facility count;
• inventory investment versus lead time;
• fixed capacity versus flexibility;
• service speed versus shipment consolidation;
• tax efficiency versus operational complexity;
• automation CAPEX versus labour OPEX; and
• efficiency versus resilience.

A scenario comparison should show at least total cost, service attainment, capacity utilisation, inventory value, carbon or other externalities where relevant, and implementation risk. A cost difference should be interpreted with its uncertainty interval, not as an exact truth.

For project comparison, net present value is:

where I₀ is initial investment, CF_t is net cash flow in year t, r is the discount rate, N is project life and RV is terminal residual value. The calculation assumes cash flows and the discount rate are defined consistently, normally in nominal or real terms but not mixed. A positive NPV indicates value creation relative to the chosen discount rate, not automatic operational superiority.

Break-even volume between two alternatives can be found when fixed and variable costs are linear:

where F₁ and F₂ are fixed annual costs, c₁ and c₂ are variable costs per unit, and V_BE is the volume at which total costs are equal. The denominator must be non-zero, and the result is meaningful only within the feasible capacity range.

4. Data quality, assumptions and governance

Data quality has at least six dimensions: accuracy, completeness, consistency, timeliness, uniqueness and validity. A precise-looking model can still be weak if it contains old rates or systematically excludes difficult customers. Data profiling should identify missing values, outliers, duplicate locations, impossible coordinates, negative quantities, inconsistent units and unexplained reconciliation gaps.

A practical data-quality register should include field, owner, source, definition, unit, refresh date, quality score, transformation and known limitation. Assumptions should be written in an assumption log with a rationale, sensitivity, approval status and expiry date. Important assumptions include forecast growth, facility availability, labour rates, exchange rates, service targets, tax recovery and residual value.

Reconciliation controls are essential. Total demand should reconcile to customer or product totals; inbound and outbound flows should reconcile to inventory movements; lane costs should reconcile to invoices or rate cards; and modelled facility throughput should reconcile to operational records. Version control should preserve the source data, transformations, scenario parameters and results.

Sensitivity and scenario analysis should test variables that could reverse the decision. Common tests include demand growth, fuel price, wage rates, exchange rates, duty rates, service requirements, capacity loss, inventory carrying rate and facility start date. Monte Carlo simulation may be appropriate when several uncertain variables interact, but it is only as credible as the distributions and correlations supplied.

5. Worked example: comparing two distribution configurations

An illustrative manufacturer must serve 100,000 units annually. Unit product value is $50. Configuration A uses one national distribution centre; Configuration B uses two regional centres.

Assume a carrying rate of 20% and use a 10% annual capital charge for screening. Inventory value for A is:

18,000 × $50 = $900,000.

Thus, annual inventory carrying cost is $900,000 × 0.20 = $180,000. For B, inventory value is 25,000 × $50 = $1,250,000, giving carrying cost of $250,000.

Annual operating cost excluding capital charge is therefore:

• A: $1,200,000 + $700,000 + $300,000 + $180,000 = $2,380,000.
• B: $900,000 + $1,050,000 + $420,000 + $250,000 = $2,620,000.

Capital charges are $400,000 for A and $650,000 for B. Screening economic cost becomes:

• A: $2,780,000 per year.
• B: $3,270,000 per year.

B therefore costs $490,000 more annually under the stated assumptions, but improves service attainment by four percentage points. Whether that premium is justified depends on the value of additional service, contractual penalties, customer retention, strategic positioning and risk reduction. The result is not “A is correct”; it is “B requires at least $490,000 of annual quantified or strategically accepted benefit to offset its cost under this screening method.”

The example also illustrates why transport cost alone is inadequate. B saves $300,000 in transport but incurs $350,000 more facility and handling cost and $70,000 more inventory cost before capital is considered.

6. Illustrative case: Northstar Consumer Products

Background. Northstar Consumer Products imports household goods through one port and serves six metropolitan markets. It operates one central warehouse and is considering a second regional facility.

Problem. The central warehouse reports high transport expenditure and 91% on-time delivery, below the company’s 96% target. A preliminary proposal claims that a regional facility will reduce annual cost by 8%, but the proposal excludes inventory, returns and implementation costs.

Analysis. The project team first created a data dictionary and reconciled twelve months of orders with shipment records. It discovered that 6% of recorded demand had been constrained by stockouts, 14% of freight invoices contained accessorial charges not present in the rate table, and customer addresses included 3.5% duplicate or obsolete records. The team then modelled demand by market and SKU family, peak-week capacity, delivery frequency, safety stock, reverse logistics and import-tax treatment.

The regional option reduced outbound kilometres and improved projected on-time delivery to 97%. However, it required additional safety stock, duplicated management systems and a higher fixed-cost base. Sensitivity analysis showed that the regional facility was economically preferable only if annual demand exceeded 1.18 million units, or if the value of improved service exceeded approximately $0.42 per shipped unit. Below that threshold, the central option remained cheaper after inventory and capital costs.

Approach and solution. Northstar adopted a staged decision. It first introduced a cross-dock service in the regional market using existing third-party capacity. This created service improvement with limited capital commitment. A permanent facility would be reconsidered after six months of measured demand, service and cost data.

Outcome. The staged approach reduced decision risk, corrected the cost baseline and made the expansion trigger explicit. It also prevented a misleading conclusion based solely on transport savings.

Learning points. Data cleansing can change the apparent economics; service has an economic value that must be made explicit; reversible options can be preferable under uncertainty; and every recommendation should identify the assumptions that could reverse it.

7. Technology, ethics and contemporary practice

Modern network analysis commonly combines ERP, warehouse-management, transport-management, telematics, geospatial systems and cloud analytics. Digital twins can test facility, inventory and service scenarios, while machine learning can improve demand forecasts or anomaly detection. These tools do not eliminate the need for definitions, controls and managerial judgement. An algorithm trained on biased historical allocations may reproduce poor service to smaller customers; a forecast model may hide uncertainty behind a single number.

Governance requires traceability from source transaction to decision output. Access controls should protect commercially sensitive customer, price and supplier data. Personal data in delivery records should be minimised, retained only as necessary and processed in accordance with applicable privacy law. Automated recommendations should be reviewable, and material assumptions should be explainable to decision-makers and affected stakeholders.

Contemporary analysis increasingly includes carbon, labour conditions, disruption exposure and circular flows. These may be monetised, constrained or reported as separate dimensions. They should not be added casually to financial cost without a stated valuation method. A robust model makes trade-offs visible rather than presenting an artificial single optimum.

Best-practice checklist

1. Define the decision, scope, time horizon and cost perspective before collecting data.
2. Use a data dictionary with owners, units and refresh dates.
3. Reconcile demand, flows, inventory and costs independently.
4. Separate fixed, variable, capital, working-capital and non-recoverable tax effects.
5. Model peak capacity and service definitions, not only annual averages.
6. Prevent double counting through a documented cost architecture.
7. Record assumptions and test the variables most likely to change the decision.
8. Present service, risk and sustainability alongside cost.
9. Prefer staged or reversible commitments when uncertainty is high.
10. Preserve an auditable trail from source data to recommendation.

Chapter Summary

A distribution-network model requires integrated data on demand, products, locations, flows, capacities, service promises and costs. Material-balance equations and reconciliation controls protect the model from structural errors. Landed cost captures the cost of placing product at a receiving point, whereas cost-to-serve extends analysis to customer-specific fulfilment activities. Operating cost, capital expenditure, inventory carrying cost and taxes must be classified consistently and evaluated over a common time horizon.

Economic analysis should expose trade-offs rather than optimise freight in isolation. NPV, annualised capital cost and break-even analysis are useful when their assumptions are respected. Data quality, tax interpretation, service definitions and uncertainty can materially alter a recommendation. Sound practice therefore combines quantitative modelling, sensitivity analysis, governance, ethical data use and staged implementation where appropriate.

Key Terms

Demand: Required or observed quantity of product over a specified period.

Constrained demand: Demand not fulfilled because of stock, capacity or policy limitations.

Material balance: Equality between beginning inventory plus inflows and ending inventory plus outflows.

Landed cost: Cost of acquiring and placing product at a specified receiving location.

Cost-to-serve: Cost of activities required to fulfil a customer, channel or segment.

CAPEX: Expenditure that creates or improves a long-lived asset.

OPEX: Operating expenditure incurred to run the network.

Inventory carrying cost: Economic cost of holding inventory over time.

Safety stock: Inventory held to protect against demand or supply uncertainty.

Service level: Defined measure of fulfilment or delivery performance.

Capacity: Maximum feasible volume under specified resource and time conditions.

Fixed cost: Cost that does not vary directly with short-term volume within a relevant range.

Variable cost: Cost that changes with activity or volume.

Customs duty: Border charge calculated according to classification, value or quantity.

Incoterm: Standard trade term allocating specified delivery responsibilities and risks.

NPV: Present value of project cash flows net of initial investment.

Break-even volume: Volume at which two alternatives have equal total cost.

Sensitivity analysis: Examination of result changes when assumptions vary.

Data lineage: Traceable path from source data through transformations to outputs.

Cost allocation: Assignment of shared cost to products, customers or activities using a stated basis.

Review Questions

Short Answer Questions

1. Why should unconstrained demand be distinguished from historical shipments?
2. State the material-balance identity for a product and period.
3. What is the difference between landed cost and cost-to-serve?
4. Why can recoverable tax be excluded from economic cost but remain relevant to cash flow?
5. Name four dimensions of data quality.

Long Answer Questions

1. Explain how demand, flow, location, capacity and service data combine to form a network model.
2. Discuss the principal trade-offs between a centralised and decentralised distribution network.
3. Explain how inventory carrying cost should be calculated and identify the risks of double counting.
4. Evaluate the role of NPV and annualised capital cost in distribution-network decisions.
5. Describe a governance process for ensuring that a network model remains auditable and decision-relevant.

Analytical / Application Questions

1. A proposed regional facility reduces transport cost by $250,000 annually but increases facility OPEX by $180,000, inventory carrying cost by $45,000 and annual capital charge by $90,000. Calculate the annual economic-cost change and state the minimum annual benefit required from improved service.
2. A customer generates $2,000,000 revenue, $1,300,000 COGS and $520,000 cost-to-serve. Calculate contribution after service cost and explain whether revenue alone supports retaining the account.
3. Two facilities have fixed annual costs of $800,000 and $1,250,000 and variable costs of $6.50 and $4.25 per unit, respectively. Calculate the break-even volume and discuss one reason why the result might be unreliable in practice.

Multiple Choice Questions with Answers

Question 1

Which data element is most directly required to test a facility’s peak-week feasibility?

A. Annual average sales only

B. Peak-period throughput and resource capacity

C. Supplier brand name

D. Customer payment terms

Correct Answer: B. Peak-period throughput and resource capacity

Question 2

A product has a purchase cost of $40, inbound freight of $5, duty of $3 and non-recoverable tax of $2 per unit. What is its landed cost, excluding other components?

A. $43

B. $45

C. $48

D. $50

Correct Answer: D. $50

Question 3

If average inventory value is $600,000 and the annual carrying rate is 18%, what is annual inventory carrying cost?

A. $10,800

B. $108,000

C. $333,333

D. $600,018

Correct Answer: B. $108,000

Question 4

Which statement best distinguishes cost-to-serve from landed cost?

A. Cost-to-serve includes fulfilment activities associated with a customer or segment

B. Landed cost always includes customer returns and promotions

C. Cost-to-serve excludes outbound transportation

D. Landed cost is calculated only for services

Correct Answer: A. Cost-to-serve includes fulfilment activities associated with a customer or segment

Question 5

A network has annual CAPEX of $5,000,000 for screening and uses a 10% capital charge. What is the simplified annual capital charge?

A. $50,000

B. $500,000

C. $5,000,000

D. $50,000,000

Correct Answer: B. $500,000

Question 6

Which condition is an assumption of the basic economic order quantity formula?

A. Demand is completely unknown

B. Replenishment is instantaneous

C. Unlimited stockouts are required

D. Every product has a different currency

Correct Answer: B. Replenishment is instantaneous

Question 7

Why must service-level terminology be explicitly defined in a network model?

A. All service measures have identical meanings

B. Service level can refer to different measures such as fill rate or on-time delivery

C. Service levels affect only accounting depreciation

D. Service data cannot be measured

Correct Answer: B. Service level can refer to different measures such as fill rate or on-time delivery

Question 8

Which control most directly detects whether inbound and outbound flows are consistent with inventory movements?

A. Material reconciliation

B. Changing the discount rate

C. Increasing the tariff rate

D. Removing duplicate customer names without review

Correct Answer: A. Material reconciliation

Question 9

Two alternatives have fixed costs of $400,000 and $700,000, and variable costs of $8 and $5 per unit, respectively. At what volume are their costs equal?

A. 100,000 units

B. 300,000 units

C. 600,000 units

D. 1,100,000 units

Correct Answer: B. 300,000 units

Question 10

What is the main purpose of sensitivity analysis in network design?

A. To guarantee that forecasts are correct

B. To identify how changes in assumptions may alter the decision

C. To eliminate the need for data governance

D. To convert all OPEX into CAPEX

Correct Answer: B. To identify how changes in assumptions may alter the decision