Risk, Resilience and Sustainable Logistics Analytics

Introduction

Logistics systems operate in conditions of uncertainty. Demand may change abruptly, transport corridors may close, suppliers may fail, cyberattacks may interrupt information flows, and extreme weather may damage infrastructure. At the same time, logistics organizations are expected to reduce greenhouse-gas emissions, energy use, waste and resource consumption. These objectives are interdependent: a decision that improves efficiency under normal conditions may increase vulnerability during disruption, while a resilient alternative may have higher short-run cost or emissions.

Risk, resilience and sustainability analytics provides a structured basis for making such trade-offs visible. It combines supply-network mapping, probability and impact assessment, scenario modelling, optimization, recovery analysis and environmental accounting. The purpose is not to eliminate all risk, which is usually impossible and uneconomic, but to understand exposure, prioritize interventions and design systems that can absorb, adapt to and recover from disruption while meeting environmental and social responsibilities.

Learning Objectives

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

1. Map logistics vulnerabilities and distinguish hazards, exposure, vulnerability and consequence.
2. Quantify disruption likelihood, expected loss, service exposure and recovery performance.
3. Construct and evaluate disruption scenarios, including recovery alternatives.
4. Apply resilience metrics such as time to recovery and time to survive.
5. Calculate and interpret carbon, energy, waste, reverse-logistics and circularity indicators.
6. Evaluate trade-offs among cost, service, resilience and environmental performance.
7. Design an analytics governance approach that is auditable, ethical and operationally useful.

1. Risk in Logistics Networks

A logistics network consists of nodes—suppliers, plants, warehouses, terminals, distribution centres and customers—and links such as roads, railways, shipping lanes, information systems and contractual relationships. A vulnerability exists when the failure or degradation of a component can materially affect network objectives.

Important terms must be distinguished:

• Hazard is a potentially damaging event, such as a flood, strike, fire or cyberattack.
• Exposure is the presence of assets, flows or activities that may be affected.
• Vulnerability is the susceptibility of an exposed element to harm.
• Consequence is the resulting loss of service, revenue, safety, quality, environmental performance or reputation.
• Risk is a structured representation of uncertain adverse outcomes, commonly combining likelihood and consequence.
• Resilience is the capacity to prepare for, absorb, adapt to and recover from disruption while preserving critical functions.

A useful initial representation of a network is an adjacency matrix A = [a_ij], where a_ij = 1 if a direct link exists from node i to node j and a_ij = 0 otherwise. In weighted networks, a_ij may represent capacity, transit time, cost or emissions. Mapping should also include single-source dependencies, shared carriers, common geographic corridors, substitutable facilities, inventory buffers and information-system dependencies.

Vulnerability mapping

A practical vulnerability register records the asset or process, hazard, affected flow, existing controls, probable consequence, data confidence and accountable owner. Network analysis can identify highly connected nodes, bottlenecks and bridges. However, graph centrality alone is insufficient: a low-connectivity facility may handle a critical product, hazardous material or legally mandated service.

A disruption exposure measure for a flow can be expressed as:

where E_f is expected exposure for flow f, V_f is the value or critical volume affected, p_f is the probability of disruption during the analysis period, and d_f is the expected fractional degradation of service. If V_f is monetary, exposure is monetary; if it is units, exposure is in units. The expression assumes that the factors are defined on compatible time horizons and that a simple multiplicative approximation is appropriate. It is not a causal law and should not be used when probabilities and impacts are strongly dependent without adjustment.

For several mutually exclusive scenarios s, expected loss is:

where p_s is scenario probability, L_s is loss under scenario s, and ∑ p_s = 1. If scenarios are not mutually exclusive, joint probabilities or simulation are required to avoid double counting.

A risk matrix is useful for screening, but ordinal labels such as “high” and “medium” should not be mistaken for precise probabilities. Quantitative estimates should report assumptions, ranges and confidence levels. Historical frequency may be unreliable for emerging threats such as novel cyber vulnerabilities or climate-regime changes; expert elicitation, stress testing and external hazard data can supplement historical records.

2. Resilience Measurement and Recovery

Resilience is dynamic. A system may operate normally, degrade during an event, and recover through rerouting, substitution, inventory release, repair or demand prioritization. Let P(t) denote performance at time t and P₀ normal performance. A normalized performance index is:

where r(t) = 1 represents normal performance. Performance may be measured as fulfilled demand, throughput, on-time delivery or service availability. A simple cumulative resilience measure over a period [0,T] is:

This is dimensionless when r(t) is dimensionless. It assumes that performance can be meaningfully aggregated over time and that all periods have equal decision importance. If shortages during a critical medical-delivery window are more serious than equal shortages later, a weighted measure is preferable.

Two operational metrics are especially important:

• Time to survive (TTS): how long the system can maintain an acceptable service level without replenishment or restoration.
• Time to recovery (TTR): time required to restore performance to a defined target after disruption.

For a facility with usable inventory I units, normal demand rate d units per day and minimum operating stock S_min, a simplified inventory-based TTS is:

provided I > S_min, demand is approximately constant, no replenishment occurs, and inventory is usable for the relevant demand. If demand is stochastic, lead times vary or inventory is multi-echelon, simulation is more suitable.

A recovery curve should specify the disruption start, minimum performance, restoration milestones and recovery target. TTR is not merely repair time; it includes detection, decision approval, procurement, transport, testing and ramp-up. Consequently, analytics should separate technical restoration from managerial and logistical restoration.

Resilience investments include dual sourcing, geographical diversification, capacity flexibility, postponement, safety stock, cross-training, alternate transport modes, backup power, cyber redundancy and collaborative visibility. Each has cost and possible environmental consequences. For example, duplicated facilities may improve robustness but increase fixed energy use, while local sourcing may reduce transport distance but create dependence on a smaller supplier base.

3. Scenario Modelling and Decision Analysis

Scenario analysis asks “what if?” rather than predicting a single future. Scenarios should vary drivers that decision-makers can understand: supplier loss, port closure, fuel-price increase, flood duration, demand surge, cyber outage or simultaneous disruptions. A scenario specification should identify trigger, duration, affected assets, capacity reduction, demand response, recovery actions and termination condition.

Three levels are useful:

1. Sensitivity analysis: changes one parameter at a time to identify influential variables.
2. Scenario analysis: evaluates coherent combinations of changes.
3. Monte Carlo simulation: repeatedly samples uncertain inputs from specified distributions to estimate outcome distributions.

Simulation outputs should include service-level percentiles, expected cost, worst-case ranges and probability of breaching a threshold. A mean alone can hide severe but infrequent outcomes.

A decision alternative j can be assessed using expected total cost:

where C_j is preventive or implementation cost of alternative j, C_js is scenario-dependent operating, disruption and recovery cost, and p_s is scenario probability. The formula assumes costs are expressed in comparable present-value terms. If costs occur at different times, discounting is required; if risk aversion matters, expected cost should be supplemented by measures such as variance, conditional value at risk or service constraints.

A multi-objective formulation may be written as:

where C is cost, S is a shortage or service penalty, E is environmental impact, R is residual risk, and w_c,w_s,w_e,w_r are normalized weights. Because these measures have different units, normalization is essential. A more transparent alternative is constrained optimization, such as minimizing cost subject to emissions, service and risk limits. Weight selection should be documented and tested through sensitivity analysis.

Worked example: supplier disruption

An organization has annual product flow worth $10 million through a single supplier. A major disruption has probability 0.08 per year and would interrupt 60% of the flow for an average of 0.25 of the year. The approximate expected exposure is:

E = 10,000,000 × 0.08 × 0.60 × 0.25 = $120,000

Suppose dual sourcing costs $70,000 annually and reduces disruption probability to 0.03, while the affected fraction falls to 0.35 and duration to 0.20. Residual expected exposure is:

E_d = 10,000,000 × 0.03 × 0.35 × 0.20 = $21,000

The expected reduction is $99,000. Before considering transition cost, quality effects and emissions, the annual net expected benefit is $29,000. This is not proof that dual sourcing is optimal: probability estimates may be uncertain, disruptions may be correlated, and the alternative supplier may fail during the same regional event.

4. Sustainable Logistics Analytics

Sustainable logistics analytics measures environmental performance alongside economic and service outcomes. Boundaries must be defined before calculation. A carbon result should state whether it covers transport only, transport and warehousing, upstream production, packaging, returns or end-of-life treatment. Double counting can occur when supplier-reported embedded emissions are added to emissions already included in a product life-cycle factor.

Carbon emissions

For activity-based estimation:

where A_i is activity data—such as litres of fuel, vehicle-kilometres, tonne-kilometres or kilowatt-hours—and EF_i is the corresponding emissions factor in compatible units. CO₂e means carbon-dioxide equivalent, incorporating the chosen global-warming-potential methodology. The analyst must specify whether factors are well-to-tank, tank-to-wheel or well-to-wheel and whether refrigerant leakage is included.

For freight measured in tonne-kilometres:

where m is transported mass in tonnes, d is distance in kilometres and EF is emissions per tonne-kilometre. This approximation assumes a valid average factor. Actual emissions may depend on load factor, empty running, vehicle type, terrain, speed, fuel and temperature control.

Carbon intensity can be reported as:

where Q is a defined functional output, such as tonnes delivered, cases delivered or revenue. Comparisons are meaningful only when the denominator, boundary, product mix and service level are comparable.

Energy, waste and packaging

Energy intensity is:

Energy should be reported with units such as kWh, megajoules or fuel litres, and electricity factors should reflect the selected location-based or market-based accounting approach where applicable. Warehouse analytics should distinguish base load, material-handling equipment, refrigeration, lighting and heating.

Waste diversion rate is:

where W_recovered is material sent to reuse, recycling or approved recovery and W_total is total waste generated. Recovery is not equivalent to recycling: energy recovery, downcycling and reuse have different environmental outcomes. Weighbridge records, supplier certificates and mass-balance checks improve reliability.

5. Reverse Logistics and Circularity

Reverse logistics manages flows from customers or downstream users back to collection, inspection, repair, refurbishment, remanufacture, recycling or disposal. Analytics must account for uncertain return timing, variable condition, inspection capacity, disposition decisions and the environmental impact of additional transport and processing.

A return rate is:

A recovery yield is:

The denominator must be stated because “recovery” may mean resale, component harvesting, material recycling or any non-landfill route.

Circularity is multidimensional. A material circularity indicator may combine recycled input, recovery and product utility, but organizations should avoid presenting a single score as an objective fact when boundary choices and weighting are contestable. Useful operational measures include percentage recycled input, percentage reusable packaging cycles, product-life extension, repair success rate, landfill diversion, secondary-material yield and virgin-material avoided.

A closed-loop network model should optimize forward and reverse flows together. For example, locating a return centre near customers may reduce collection distance, while locating it near repair suppliers may reduce processing lead time. The best location depends on return density, product value, inspection technology, labour skills, transport emissions and residual-value uncertainty.

6. Integrated Trade-offs and Governance

Risk and sustainability metrics can conflict. Emergency air freight may protect customer service but create high emissions. Safety stock may improve TTS but increase obsolescence. Recycling may be preferable to disposal but can require energy-intensive transport or processing. Therefore, decisions should use a balanced dashboard rather than a single “green” or “resilient” score.

A mature dashboard includes:

Governance is essential because analytics can create false confidence. Data lineage should document source, timestamp, transformation, unit, factor version and responsible owner. Models should be validated against known events and subjected to stress tests. Access controls protect commercially sensitive supplier and customer data. Personal data from returns, delivery records or driver systems must be collected lawfully, minimized and protected.

AI can assist with anomaly detection, demand sensing, route re-planning, satellite-based hazard monitoring and predictive maintenance. However, automated recommendations require human oversight. Bias may arise when historical data underrepresents informal suppliers, rural customers or rare hazards. Explainability, override rules, monitoring for model drift and documented accountability are particularly important for decisions affecting safety, employment or essential service access.

7. Illustrative Case: Resilient and Lower-Carbon Cold-Chain Distribution

Background

The following is a fictional illustrative case. A pharmaceutical distributor serves three regional hospitals from one refrigerated distribution centre. The centre receives products from a domestic manufacturer and imports a temperature-sensitive product through a coastal port. Management wants to improve resilience without increasing emissions disproportionately.

Problem

A port closure would interrupt imported supply for up to ten days. The centre has five days of usable inventory above its emergency minimum. Refrigerated road transport is available from an inland alternative supplier, but the route is 30% longer. Emergency air freight is possible but expensive and carbon intensive. A power outage at the distribution centre could also cause product loss.

Analysis

The team maps suppliers, port, roads, cold storage, backup power, hospital demand and information dependencies. It estimates TTS from usable inventory and daily demand, then simulates port closures of 3, 7 and 14 days. Scenarios include normal operations, port closure alone, closure plus demand surge, and closure plus power outage. For each scenario the model records fulfilled doses, spoilage, cost, CO₂e and TTR.

The analysis identifies three vulnerabilities: dependence on one port, inadequate temperature-monitoring redundancy and a recovery delay caused by manual approval of alternative transport. The port has high network importance, but the power system has the greatest expected product-loss consequence.

Approach and solution

Management evaluates four alternatives: increase inventory, qualify an inland supplier, install battery-backed refrigeration and establish pre-approved emergency transport contracts. The preferred portfolio is not the cheapest individual measure. It combines moderate inventory, dual sourcing for critical products, backup power and digital temperature alerts. Air freight is retained only as a last-resort action under a service-critical threshold.

The dashboard uses service constraints—no hospital may fall below its critical stock level—rather than allowing emissions savings to justify unacceptable patient risk. Carbon is calculated from fuel and electricity activity data, while emergency-mode emissions are reported separately so that routine improvement is not hidden by crisis response.

Outcome

In the model, the portfolio reduces expected shortage exposure and TTR while producing a small increase in routine energy use. The increased energy is partly offset by improved refrigeration controls and reduced spoilage. The organization also gains auditable decision rules: when inventory falls below the trigger, the inland supplier is activated; when temperature data become unreliable, shipments are quarantined pending inspection.

Learning points

1. Resilience is a portfolio property, not the result of one buffer.
2. TTS and TTR reveal different weaknesses.
3. Environmental analysis must include emergency actions and avoided spoilage.
4. Service constraints are essential when failures affect health or safety.
5. Pre-approved decisions reduce recovery delay.

8. Limitations and Contemporary Developments

Risk estimates are limited by sparse data, changing climate hazards, correlated failures and strategic behaviour by suppliers. Environmental factors differ by geography, technology and accounting method. A model may be mathematically precise but operationally wrong if its boundaries or assumptions are poor. Analysts should use ranges, scenario narratives and sensitivity analysis rather than presenting uncertain outputs as exact forecasts.

Contemporary practice is moving toward digital twins, real-time control towers, satellite and weather data, supplier-risk platforms, product passports and interoperable emissions data. These technologies improve visibility but do not remove the need for governance. Digital twins require synchronization and validation; real-time systems may generate alert fatigue; product passports depend on data standards and supplier participation. The strategic capability is therefore not technology alone, but the combination of credible data, robust models, cross-functional authority and disciplined action protocols.

Best-Practice Checklist

• Define the decision, system boundary and critical service before selecting metrics.
• Map physical, informational, contractual and geographic dependencies.
• Use scenario ranges and test correlated disruptions.
• Report TTS, TTR, performance loss and recovery target together.
• Separate activity data, emissions factors and calculation assumptions.
• Compare absolute emissions with intensity metrics.
• Include reverse logistics, spoilage and end-of-life flows where material.
• Use constraints and sensitivity analysis for multi-objective decisions.
• Validate models against events and involve operational experts.
• Maintain data lineage, version control, human oversight and privacy protection.

Chapter Summary

Risk, resilience and sustainable logistics analytics integrates network vulnerability mapping, probabilistic exposure assessment, disruption scenarios, recovery modelling and environmental accounting. Risk analysis distinguishes hazard, exposure, vulnerability and consequence; resilience analysis examines degradation, time to survive, time to recovery and cumulative performance. Scenario and simulation methods reveal how alternative sourcing, inventory, capacity, transport and recovery policies perform under uncertainty. Sustainable logistics analytics quantifies carbon, energy, waste, reverse logistics and circularity, but results depend on clear boundaries, compatible units and reliable factors. Because cost, service, resilience and environmental outcomes may conflict, decisions should use transparent multi-objective or constrained analysis. Effective implementation also requires data governance, model validation, ethical AI oversight and accountable operational action.

Key Terms

Risk: Representation of uncertain adverse outcomes and their consequences.

Hazard: Potentially damaging event or condition.

Exposure: Assets, flows or activities subject to a hazard.

Vulnerability: Susceptibility of an exposed element to harm.

Consequence: Resulting loss or effect of a disruption.

Resilience: Capacity to absorb, adapt to and recover from disruption.

Time to survive: Duration for which acceptable service can be maintained without restoration.

Time to recovery: Time needed to restore performance to a specified target.

Scenario analysis: Evaluation of coherent possible future conditions.

Monte Carlo simulation: Repeated random sampling used to estimate outcome distributions.

Expected loss: Probability-weighted average loss across defined scenarios.

Carbon dioxide equivalent: Common unit for expressing greenhouse-gas impact.

Carbon intensity: Emissions relative to a defined operational output.

Energy intensity: Energy consumption relative to output.

Reverse logistics: Movement and management of products or materials back upstream.

Circularity: Retention or restoration of product and material value through loops such as reuse, repair and recycling.

Recovery yield: Proportion of returned material or products directed to a specified recovery route.

Data lineage: Record of data origin, transformation, ownership and version.

Review Questions

Short Answer Questions

1. Distinguish hazard, exposure and vulnerability in a logistics network.
2. What is the difference between time to survive and time to recovery?
3. State two assumptions required when using a tonne-kilometre emissions factor.
4. Why must environmental-system boundaries be defined before calculation?
5. Identify two governance controls needed for logistics analytics.

Long Answer Questions

1. Explain how a logistics organization can map physical, informational and contractual vulnerabilities.
2. Compare sensitivity analysis, scenario analysis and Monte Carlo simulation for disruption planning.
3. Discuss how inventory, dual sourcing and backup capacity affect resilience and sustainability.
4. Explain the calculation and interpretation of carbon, energy and waste indicators.
5. Critically evaluate the use of a single composite score for resilience and circularity decisions.

Analytical / Application Questions

1. A network carries annual goods worth $8 million. A disruption probability is 0.05, with 50% of flow affected for 0.4 of a year. Calculate approximate expected exposure and explain two limitations of the estimate.
2. Design three disruption scenarios for a cold-chain network and specify the performance, recovery and environmental metrics required to compare them.
3. A firm can choose emergency air freight, dual sourcing or additional inventory. Develop a decision framework that accounts for service, expected cost, TTR, CO₂e and uncertainty.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes exposure from vulnerability?

A. Exposure is the probability of failure; vulnerability is the financial cost.

B. Exposure is what may be affected; vulnerability is susceptibility to harm.

C. Exposure is recovery time; vulnerability is normal performance.

D. Exposure is emissions; vulnerability is energy consumption.

Correct Answer: B. Exposure is what may be affected; vulnerability is susceptibility to harm.

Question 2

A facility has 900 usable units, an emergency minimum of 300 units, and demand of 150 units per day. Under the stated constant-demand assumptions, its TTS is:

A. 2 days

B. 4 days

C. 6 days

D. 8 days

Correct Answer: B. 4 days

Question 3

If scenario probabilities are 0.2 and 0.8, with losses of $50,000 and $10,000 respectively, expected loss is:

A. $12,000

B. $18,000

C. $20,000

D. $40,000

Correct Answer: B. $18,000

Question 4

Which metric specifically measures the time needed to restore performance to a defined target?

A. Time to survive

B. Carbon intensity

C. Time to recovery

D. Recovery yield

Correct Answer: C. Time to recovery

Question 5

In the activity-based emissions formula, the emissions factor must be:

A. Independent of the activity-data unit

B. Expressed in units compatible with the activity data

C. Measured only in kilograms of product

D. Based exclusively on financial expenditure

Correct Answer: B. Expressed in units compatible with the activity data

Question 6

Which practice most directly reduces double counting in carbon accounting?

A. Using more than one emissions factor for the same activity

B. Combining all supplier and logistics estimates without checking boundaries

C. Defining system boundaries and checking whether embedded emissions are already included

D. Reporting only carbon intensity and omitting absolute emissions

Correct Answer: C. Defining system boundaries and checking whether embedded emissions are already included

Question 7

A recovery yield for returned products should be interpreted primarily as:

A. Returned units divided by units sold

B. Recovered units or mass divided by returned units

C. Recycling cost divided by transport cost

D. New product sales divided by returned units

Correct Answer: B. Recovered units or mass divided by returned units

Question 8

Which approach is most appropriate when disruptions are correlated and historical data are sparse?

A. Use a risk matrix alone

B. Assume all risks are independent

C. Combine expert-informed scenarios, stress tests and sensitivity analysis

D. Ignore low-frequency events

Correct Answer: C. Combine expert-informed scenarios, stress tests and sensitivity analysis

Question 9

Why are service constraints important in a pharmaceutical cold-chain model?

A. They ensure emissions always remain zero

B. They prevent cost minimization from permitting unacceptable critical-stock shortages

C. They eliminate the need for scenario analysis

D. They guarantee that all suppliers are local

Correct Answer: B. They prevent cost minimization from permitting unacceptable critical-stock shortages

Question 10

Which is an appropriate governance practice for AI-supported logistics decisions?

A. Remove human override to ensure consistency

B. Avoid documenting model versions

C. Monitor model drift and retain accountable human oversight

D. Use historical data without testing for bias

Correct Answer: C. Monitor model drift and retain accountable human oversight