Performance, Economics, and Managerial Decision-Making

Introduction and context

Omnichannel distribution creates a managerial paradox: customers expect a seamless experience across stores, websites, marketplaces, mobile applications, and delivery partners, while the underlying network becomes more fragmented and expensive to coordinate. A single customer order may involve several inventory locations, multiple transportation modes, split shipments, reverse logistics, customer-service contacts, and different tax or regulatory treatments. Consequently, performance cannot be judged by sales or delivery speed alone.

A sound managerial system connects four questions:

1. Did the network deliver the customer promise?
2. What resources and costs were consumed?
3. Did the activity create economic value?
4. Was the environmental and social effect acceptable?

This chapter develops an integrated approach to those questions. It treats metrics not as isolated numbers but as instruments for diagnosis, experimentation, investment appraisal, and governance.

Learning objectives

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

• define and calculate major omnichannel service, inventory, demand, cost, profitability, and carbon measures;
• distinguish related measures such as fill rate, OTIF, and perfect-order rate;
• construct a KPI hierarchy linking customer outcomes to operational drivers and financial consequences;
• evaluate trade-offs among service, inventory, cost, speed, resilience, and emissions;
• design experiments and dashboards that support managerial decisions rather than merely report activity;
• calculate contribution margin, cost-to-serve, customer lifetime value, and basic capital-investment measures;
• identify measurement risks, data-governance requirements, and ethical limitations.

1. From operational measures to economic decisions

A KPI is useful only when its definition, population, time window, owner, and decision use are explicit. For example, “delivery performance” may mean the percentage of orders dispatched on time, delivered on time, or delivered on time and in full. These are not interchangeable.

A practical KPI architecture has four levels:

The hierarchy prevents a common error: optimizing a local measure while damaging the total system. For example, same-day delivery may improve conversion but increase split shipments, premium transport, failed deliveries, and emissions. Conversely, reducing inventory may improve turns while increasing stockouts and lost customers.

1.1 Measurement design principles

Every KPI specification should state:

• Unit of analysis: order, order line, item, shipment, customer, or period;
• Numerator and denominator: including treatment of cancellations, substitutions, and partial shipments;
• Clock start and stop: for example, order confirmation to customer receipt;
• Service promise: the promised date and quantity, not an undocumented expectation;
• Data source and timestamp: transaction system, warehouse system, carrier scan, or survey;
• Owner and action threshold: who investigates and what decision follows;
• Segmentation: channel, product category, geography, customer segment, and fulfillment mode.

A metric without a denominator or population definition is an assertion, not a measurement.

2. Service and reliability metrics

2.1 Fill rate

Fill rate measures the proportion of demand supplied immediately from available inventory. At order-line level:

The numerator and denominator must use the same unit and period. A unit fill rate can instead be calculated as units supplied divided by units demanded. Line and unit fill rates can produce different conclusions when large orders are short on only one line.

Fill rate is an availability measure, not a complete delivery measure. An order can be filled but arrive late, damaged, or in several shipments.

2.2 On time in full (OTIF)

OTIF is the proportion of qualifying deliveries that meet both the promised time and promised quantity:

“On time” requires a defined tolerance, such as delivery on or before the promised date, or within a specified time window. “In full” requires a defined quantity rule. A delivery that is early but incomplete is not OTIF. The denominator should normally exclude only events explicitly outside the supplier’s or network’s control under a documented policy; broad exclusions can artificially inflate performance.

2.3 Perfect-order rate

A perfect order satisfies all specified conditions, commonly: complete, on time, damage-free, and correctly documented or invoiced.

If the criteria are treated as independent, the rate may be approximated by multiplying component probabilities. For example, with on-time rate 0.96, in-full rate 0.98, damage-free rate 0.995, and documentation accuracy 0.99, the independence approximation is 0.96 × 0.98 × 0.995 × 0.99 ≈ 0.926, or 92.6%. This is an approximation, not a substitute for order-level calculation, because failures may be correlated.

2.4 Lead time and variability

Lead time is elapsed time between defined events. Customer order lead time might be measured from order acceptance to delivery; replenishment lead time might run from purchase-order release to receipt and availability.

Average lead time is:

where L_i is the lead time for observation i, and n is the number of observations. The mean alone hides volatility. Managers should also examine median, percentiles, and the proportion exceeding the promise. A 95th percentile lead time is the value below which approximately 95% of observations fall; it is often more useful for service design than the average.

2.5 Return rate

Return rate must specify whether it is based on units, orders, or revenue:

Returns may indicate fit uncertainty, inaccurate product information, damage, late delivery, buyer remorse, or deliberate bracketing behavior. A high return rate is not automatically a logistics failure, but it has reverse-logistics, inspection, markdown, labor, and carbon implications.

3. Inventory, demand, and capacity metrics

3.1 Inventory turns

Inventory turns indicate how many times average inventory is consumed or sold during a period. For a merchandise business, cost of goods sold should generally be used rather than sales revenue:

A common approximation is Average\ inventory = (Beginning\ inventory + Ending\ inventory) ÷ 2. The ratio is dimensionless. Annual turns of 8 imply an approximate annual holding period of 45.6 days if usage is steady, because 365 ÷ 8 = 45.625. This conversion is only approximate when demand is seasonal or inventory is unevenly distributed.

High turns may signal efficiency, but can also reflect understocking, lost sales, or excessive dependence on expedited replenishment. Low turns may be justified for strategic assortment, long lead times, or resilience.

3.2 Forecast accuracy

Forecast accuracy must identify the forecast horizon and aggregation level. Mean absolute percentage error is:

where A_t is actual demand and F_t is forecast demand. MAPE is undefined when actual demand is zero and becomes unstable for very small denominators. Weighted absolute percentage error is often preferable for portfolios:

Forecast bias measures systematic over- or under-forecasting:

A forecast can have low absolute error but substantial bias in particular products or locations. Accuracy should therefore be segmented and linked to inventory and service outcomes.

4. Cost and profitability economics

4.1 Fulfillment cost

Fulfillment cost includes the costs required to process and deliver an order. Depending on managerial purpose, it may include picking, packing, packaging materials, payment processing, transportation, delivery failure, customer service, and allocated technology costs. The definition should distinguish variable, avoidable, and allocated costs.

A simple cost-per-order measure is:

This average is insufficient for channel decisions. Cost should be segmented by basket size, location, delivery promise, shipment count, and customer type.

4.2 Contribution margin

Contribution margin shows what remains after costs that vary with the decision volume:

Contribution margin percentage is:

Variable costs may include product cost, payment fees, pick-and-pack labor, delivery, expected returns, and promotional discounts. Fixed corporate overhead should not be deducted when evaluating a short-run incremental fulfillment choice unless it changes because of that choice.

Worked example. An order generates revenue of $120. Product cost is $60, payment cost $3, fulfillment labor and packaging $8, outbound delivery $10, and expected return cost $6. Contribution margin is $120 - ($60+$3+$8+$10+$6) = $33. The contribution margin percentage is $33 ÷ $120 = 27.5%. If a premium delivery option adds $5 of variable cost and increases revenue by $9, its incremental contribution is $4, before considering any effect on returns or repeat purchase.

4.3 Cost-to-serve

Cost-to-serve (CTS) assigns the resources consumed by a customer, order, channel, or segment. A driver-based model may include:

where for segment or order j, c_p is procurement or product-handling cost, c_h handling cost, c_s shipping, c_d delivery exception cost, c_r returns cost, and c_c customer-service cost. The terms must be defined to avoid double counting.

Profitability by segment can then be examined as:

CTS is especially valuable in omnichannel settings because identical revenue can conceal very different costs: a store pickup may avoid last-mile delivery but create store labor and inventory-transfer costs; a marketplace order may carry commissions and compliance fees.

4.4 Customer lifetime value

A simplified customer lifetime value model is:

where m_t is contribution margin per active customer in period t, r_t is the probability of retention into that period, d is the discount rate, T is the planning horizon, and CAC is customer-acquisition cost. The model assumes that margin, retention, and discount rate are estimable and that the time periods are consistently defined.

For constant periodic contribution m, retention probability r, and discount rate d, an infinite-horizon version is:

This special case is sensitive to assumptions and should not be interpreted as a precise individual prediction. It is a decision model for comparing segments or investments.

5. Carbon and broader sustainability measures

A useful carbon measure is emissions per fulfilled order or per delivered unit:

Transport emissions may be estimated from activity data:

For transport, activity data may be vehicle-kilometres, tonne-kilometres, fuel consumed, or electricity used. The emission factor must match the fuel, geography, accounting boundary, and reporting standard. Managers should state whether emissions are direct, energy-related, or value-chain emissions; otherwise comparisons are misleading.

Carbon should be reported with service and economic outcomes. Consolidating shipments may reduce emissions per order but lengthen lead time. Local inventory may reduce delivery distance but increase duplicated safety stock. Packaging reduction may lower material emissions while increasing damage. These are optimization problems with constraints, not single-metric contests.

6. Trade-off analysis and decision models

A managerial decision should compare alternatives using a common time horizon and relevant incremental consequences. A weighted score can support screening:

where z_jk is the normalized performance of alternative j on criterion k, and weights satisfy w_k ≥ 0 and ∑w_k = 1. Normalization and weights are judgmental; therefore sensitivity analysis is essential.

For investment choices, net present value is:

where I₀ is initial investment, CF_t is net cash flow in period t, r is the discount rate, and TV_T is terminal value when applicable. If cash flows are measured annually, the discount rate must be annual. A positive NPV means the investment exceeds the required return under the stated assumptions; it does not prove that the assumptions are reliable.

A business case should include a base case, optimistic and pessimistic cases, implementation costs, working-capital effects, service effects, labor implications, risks, and a post-implementation measurement plan. Payback period can be useful for liquidity discussion but ignores cash flows after payback and, in its simple form, ignores discounting.

7. Experiments, attribution, and dashboards

Operational changes should be tested when feasible. Examples include comparing delivery-promise wording, pickup incentives, packaging designs, replenishment rules, or delivery-slot prices. A randomized controlled experiment assigns comparable units—customers, stores, regions, or time blocks—to treatment and control groups. The treatment effect on a metric is commonly estimated as:

The design must control for contamination, seasonality, novelty effects, and operational interference. If randomization is impossible, a difference-in-differences design may compare changes over time between treated and untreated groups, but it requires a credible parallel-trends assumption.

Dashboards should be designed around decisions. A useful executive dashboard includes:

Dashboards should show target, actual, variance, trend, confidence or data-quality status, and accountable owner. “Red” without a prescribed decision encourages cosmetic reporting rather than control.

8. Illustrative case: Northstar Apparel

Background

Northstar Apparel, an illustrative omnichannel retailer, sells through stores and an online channel. Online growth has increased revenue, but the company reports rising delivery complaints and declining profitability. Management is considering whether to promise two-day delivery nationally or to retain a three- to five-day promise while improving reliability.

Problem and data

During one quarter, Northstar records 100,000 online orders. Of these, 96,000 are delivered by the promised date, 97,000 are complete, 94,000 are both on time and complete, and 91,000 meet all four perfect-order criteria, including damage-free delivery and accurate documentation. The online channel generates $12 million revenue. Product cost is $6.6 million; variable fulfillment cost is $2.1 million; expected return processing and markdown cost is $0.5 million; and marketing cost attributable to the channel is $0.8 million.

Current OTIF is 94,000 ÷ 100,000 = 94%. Perfect-order rate is 91,000 ÷ 100,000 = 91%. Contribution after product, fulfillment, and expected return costs is $12.0m - $6.6m - $2.1m - $0.5m = $2.8m, or 23.3% of revenue before attributable marketing. After that marketing cost, channel contribution is $2.0 million.

Analysis

The gap between OTIF and perfect order is three percentage points, indicating that damage and documentation failures are material. Management also finds that expedited orders have a fulfillment cost $4 higher per order, a 1.5 percentage-point higher return rate, and only a 0.4 percentage-point increase in repeat purchase after six months. The apparent speed benefit is therefore not necessarily economically attractive.

Cost-to-serve analysis shows that rural orders and small baskets produce negative contribution after delivery and returns, while store-pickup orders produce positive contribution despite modest store labor costs. This suggests that a universal two-day promise would cross-subsidize costly orders and consume capacity during peaks.

Approach and solution

Northstar adopts a segmented promise:

• two-day delivery for products stocked in regional fulfillment centers and customers within dense delivery zones;
• standard delivery elsewhere, with transparent date commitments;
• free or incentivized store pickup for eligible customers;
• a minimum basket threshold for free expedited delivery;
• packaging and documentation controls targeted at the perfect-order gap;
• a controlled pilot in two regions, with matched control regions.

The pilot dashboard tracks OTIF, perfect-order rate, return rate, incremental contribution per order, cost-to-serve, repeat purchase, and CO₂e per order. The company defines success before launch: no more than a one-point deterioration in perfect-order rate, positive incremental contribution, and no increase in emissions per order.

Outcome and learning points

The case illustrates that the strategic choice is not “fast versus slow.” It is a portfolio decision involving customer value, network capability, economics, and environmental consequences. A segmented promise can protect high-value service where it matters while preventing universal premium costs. It also demonstrates why outcome metrics must be connected to driver metrics and financial measures.

9. Governance, limitations, and contemporary practice

Metric gaming is a serious risk. If managers are rewarded for dispatch speed, they may dispatch incomplete orders; if rewarded for inventory turns, they may reduce safety stock below a rational service level. Governance should therefore use balanced measures, audit definitions, and review unintended consequences.

Customer-level CLV and return analytics raise privacy concerns. Data minimization, access controls, retention limits, explainable segmentation, and compliance with applicable privacy law are necessary. Algorithms should not quietly deny delivery options or impose discriminatory prices on protected groups. AI forecasting can improve granularity and detect patterns, but it inherits bias from historical promotions, stockouts, and missing data. Forecast models should distinguish observed demand from censored demand: sales of zero may reflect no demand or no inventory.

Contemporary networks increasingly use event-driven visibility, digital twins, machine-learning forecasts, dynamic delivery-slot pricing, and automated exception management. These technologies are valuable only when data definitions and decision rights are sound. More real-time data does not automatically create better decisions; it can increase noise, alert fatigue, and false precision.

Best practice is to combine: a metric dictionary; order-level event data; clear financial boundaries; segmentation; controlled experimentation; scenario analysis; independent data-quality checks; and periodic review of whether KPIs still represent strategy. The most mature organizations treat performance management as a learning system rather than a punishment system.

Chapter Summary

Omnichannel performance management requires an integrated view of customer service, operations, economics, and sustainability. Fill rate measures immediate availability; OTIF combines timeliness and completeness; perfect-order rate adds criteria such as damage-free delivery and documentation. Lead-time distributions are more informative than averages alone. Inventory turns and forecast measures must be interpreted in relation to stockouts, bias, seasonality, and resilience.

Fulfillment cost, contribution margin, cost-to-serve, and CLV translate operational choices into economic consequences. Carbon measures extend the analysis beyond financial outcomes, but their boundaries and emission factors must be explicit. Trade-off analysis, experiments, dashboards, and NPV-based business cases allow managers to evaluate alternatives systematically. The central managerial principle is to optimize the total customer and business system, not a locally attractive KPI.

Key Terms

KPI: A defined measure used to monitor performance and support a decision.

Fill rate: The proportion of demand supplied immediately from available inventory.

OTIF: The proportion of qualifying deliveries that are both on time and complete.

Perfect-order rate: The proportion of orders meeting every defined service criterion.

Lead time: Elapsed time between specified operational events.

Inventory turns: Cost of goods sold divided by average inventory at cost.

Forecast bias: Systematic tendency to overforecast or underforecast demand.

MAPE: Mean absolute percentage error, a scale-relative forecast-error measure.

WAPE: Weighted absolute percentage error using aggregate demand as the denominator.

Fulfillment cost: Variable and assigned costs required to process and deliver an order.

Contribution margin: Revenue minus relevant variable costs.

Cost-to-serve: Resource cost attributable to a customer, order, segment, or channel.

Customer lifetime value: Discounted expected future customer contribution less acquisition cost.

Carbon intensity: Emissions associated with a defined unit, such as an order or shipment.

Trade-off: A gain in one objective accompanied by a loss in another.

Experiment: A structured comparison designed to estimate the effect of an intervention.

Net present value: Present value of future cash flows less initial investment.

Metric governance: Policies controlling definitions, data quality, ownership, and use of KPIs.

Review Questions

Short Answer Questions

1. How does OTIF differ from fill rate?
2. Why can inventory turns be misleading when interpreted alone?
3. State two limitations of MAPE.
4. What costs should be considered in cost-to-serve analysis?
5. Why should carbon emissions be reported with service and economic measures?

Long Answer Questions

1. Design a KPI hierarchy for an omnichannel retailer and explain how its levels are connected.
2. Compare perfect-order rate and OTIF as measures of customer service quality.
3. Explain how contribution margin and cost-to-serve support channel decisions.
4. Discuss the assumptions and limitations of a simplified CLV model.
5. Explain how a retailer should evaluate a proposal for universal two-day delivery.

Analytical / Application Questions

1. A retailer has 50,000 orders, of which 47,500 are on time and 48,000 are complete. If 46,000 are both on time and complete, calculate OTIF and explain why it cannot be inferred by simply averaging the two component rates.
2. Design a controlled pilot to test whether an incentive for store pickup improves contribution margin without reducing customer retention. Specify treatment, control, outcome metrics, and possible confounders.
3. A fulfillment investment requires $2 million initially and is expected to generate $900,000 in year 1, $850,000 in year 2, and $700,000 in year 3. Explain which additional assumptions are needed to calculate NPV and identify at least three nonfinancial effects that should be included in the decision.

Multiple Choice Questions with Answers

Question 1

Which measure requires an order to satisfy every defined service criterion, such as completeness, timeliness, damage-free delivery, and correct documentation?

A. Fill rate

B. Perfect-order rate

C. Inventory turns

D. Forecast bias

Correct Answer: B. Perfect-order rate

Question 2

A retailer has 20,000 order lines demanded and immediately fulfills 19,000. What is the line fill rate?

A. 90%

B. 92%

C. 95%

D. 98%

Correct Answer: C. 95%

Question 3

Which formula is most appropriate for inventory turns?

A. Revenue ÷ ending inventory at retail

B. Cost of goods sold ÷ average inventory at cost

C. Average inventory at cost ÷ revenue

D. Units returned ÷ units sold

Correct Answer: B. Cost of goods sold ÷ average inventory at cost

Question 4

Why is MAPE problematic when actual demand is zero?

A. It cannot measure lead time

B. Its denominator becomes zero

C. It always produces a negative value

D. It requires inventory cost data

Correct Answer: B. Its denominator becomes zero

Question 5

An order has revenue of $200 and variable costs of $150. What is its contribution margin percentage?

A. 20%

B. 25%

C. 33.3%

D. 75%

Correct Answer: B. 25%

Question 6

Which measure is specifically designed to assign resource consumption to a customer, order, segment, or channel?

A. Cost-to-serve

B. OTIF

C. Forecast accuracy

D. Lead-time percentile

Correct Answer: A. Cost-to-serve

Question 7

In the simplified CLV model, what does the retention probability represent?

A. The probability that a shipment is damage-free

B. The probability that a customer remains active into a future period

C. The probability that demand exceeds inventory

D. The probability that a forecast is unbiased

Correct Answer: B. The probability that a customer remains active into a future period

Question 8

What is the principal purpose of random assignment in an operational experiment?

A. To guarantee a positive financial result

B. To eliminate the need for a control group

C. To improve comparability between treatment and control groups

D. To increase the number of KPIs reported

Correct Answer: C. To improve comparability between treatment and control groups

Question 9

Which statement about NPV is correct?

A. It ignores the timing of cash flows

B. It discounts future cash flows and subtracts initial investment

C. It is identical to simple payback period

D. It measures only delivery reliability

Correct Answer: B. It discounts future cash flows and subtracts initial investment

Question 10

Which practice most directly reduces the risk of KPI gaming?

A. Rewarding one operational metric exclusively

B. Removing all denominator definitions

C. Using balanced measures and auditing metric definitions

D. Excluding all failed orders from the denominator

Correct Answer: C. Using balanced measures and auditing metric definitions