Transportation and Last-Mile Delivery

Introduction

Transportation is the physical execution layer of omnichannel distribution. It connects suppliers, fulfilment centres, stores, micro-fulfilment facilities, pickup points, lockers, and customers. In an omnichannel system, transportation is not merely a cost centre. It shapes assortment availability, delivery promises, customer experience, carbon emissions, working-capital requirements, and the economic viability of digital channels.

The last mile—the movement from a local fulfilment node to the customer or collection point—is usually the most complex and expensive segment. Deliveries are fragmented across many destinations, demand is time-sensitive, urban access is constrained, and failed delivery attempts can destroy margin. Effective managers therefore select transport modes, carriers, routes, delivery windows, and handover locations as an integrated service-design problem.

This chapter examines transport economics, carrier selection, routing, parcel strategies, alternative delivery models, urban logistics, performance measurement, and governance. The central managerial question is not “How can delivery be made fastest?” but “What combination of speed, reliability, convenience, cost, resilience, and sustainability creates value for the target customer and the firm?”

Learning Objectives

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

1. Select transport modes and carriers using service, cost, risk, and sustainability criteria.
2. Explain routing policies, delivery-window design, density, and failed-delivery economics.
3. Evaluate parcel, crowdsourced, locker, pickup-point, and urban logistics solutions.
4. Apply route and capacity metrics to operational decisions.
5. Assess carrier performance through service-level, cost, quality, and carbon indicators.
6. Critically evaluate technology, regulation, labour, privacy, and ethical issues in last-mile delivery.

1. Transportation as an Omnichannel Design Decision

A distribution network typically contains several transportation stages:

• Inbound transportation: movement from suppliers to distribution centres or stores.
• Line-haul transportation: high-volume movement between major nodes, often by road, rail, air, or sea.
• Middle mile: movement between fulfilment centres, stores, cross-docks, and regional depots.
• Last mile: movement from the final local node to the customer, locker, pickup point, or store.
• Reverse logistics: returns, exchanges, repairs, recycling, and product recovery.

The same order may follow different paths depending on inventory position, customer promise, basket size, product characteristics, and location. A ship-from-store order may reduce delivery distance but compete with store replenishment. A click-and-collect order may eliminate a home-delivery stop but transfer effort to the customer. A marketplace order may use a seller-managed carrier rather than the retailer’s network.

Transportation design must therefore align three decisions:

1. Service design: What promise is made—same day, next day, nominated window, economy, or pickup?
2. Network design: Which nodes hold inventory and which nodes serve each geographic area?
3. Execution design: Which carrier, vehicle, route, handover method, and technology will deliver the promise?

A premium service should not be offered universally if most customers do not value it enough to cover its incremental cost. Conversely, an apparently low-cost service may create hidden costs through failed delivery, complaints, refunds, churn, and additional packaging.

2. Transport Modes and Carrier Selection

2.1 Mode characteristics

Transport modes differ in speed, flexibility, shipment size, cost structure, environmental impact, and network reach.

Mode selection should consider total logistics cost rather than freight price alone. Total cost can include transport charges, inventory carrying cost caused by transit time, handling, damage, customs, insurance, stockout exposure, returns, and environmental compliance.

A useful conceptual expression is:

Here, TC is total logistics cost in currency units; F is fixed transport cost; V is variable cost per shipment or unit; Q is shipment quantity; H is handling cost; I is inventory-related cost; R is risk, damage, and returns cost; and E is environmental or regulatory cost. The expression is a managerial cost model, not a universal accounting identity. Its components must be defined consistently and should not double-count costs.

2.2 Carrier selection

Carrier selection should be based on a weighted scorecard and a contractual service model. Important criteria include:

• geographic coverage and depot density;
• delivery success rate and on-time performance;
• capacity during peaks;
• parcel size, weight, and special-handling capability;
• tracking quality and application-programming-interface integration;
• claims, damage, and loss performance;
• returns capability;
• labour and safety standards;
• carbon reporting and reduction plans;
• price structure, surcharges, minimum volumes, and peak fees;
• financial resilience and contingency capacity.

A weighted score can support comparison:

S_j is the score of carrier j; w_i is the importance weight of criterion i; and x_ij is the normalized performance of carrier j on that criterion. Scores must use a common direction: for example, a higher normalized value should consistently mean better performance. The method is useful for screening, but it does not replace capacity, legal, or resilience analysis.

A single-carrier strategy may simplify integration and improve volume leverage, but it creates concentration risk. A multi-carrier portfolio can improve resilience and geographic fit, although it increases operational complexity and may weaken bargaining power. Many mature retailers use a primary carrier, secondary carriers, specialist services, and an emergency playbook.

3. Last-Mile Economics and Density

Last-mile cost is strongly influenced by stop density. A vehicle serving many nearby customers can spread driver, vehicle, and dispatch costs across more deliveries. Sparse rural routes produce longer travel distances and fewer deliveries per hour. High density does not automatically guarantee profitability: congestion, parking scarcity, security restrictions, apartment access, and narrow delivery windows can reduce productivity.

Important drivers include:

• stops per route and parcels per stop;
• kilometres per delivery;
• driving and service time;
• vehicle capacity and load quality;
• customer availability;
• time-window tightness;
• reattempts and returns;
• package dimensions and handling effort;
• local wages, fuel, tolls, and parking costs.

Delivery productivity can be represented as:

where P is deliveries per route-hour, D is completed deliveries, and T is total route time in hours, including driving, service, waiting, and legally required breaks. Comparing productivity requires consistent definitions; excluding failed stops or waiting time can produce misleading results.

The average cost per successful delivery is:

where C_s is cost per successful delivery, C_r is total route cost, and D_s is successful deliveries. If a route costs 180 currency units and completes 30 deliveries, cost per successful delivery is 6 currency units. If five of those deliveries fail and the route cost remains 180, cost per successful delivery becomes 180 ÷ 25 = 7.20 currency units. The example illustrates why failed delivery affects economics even when planned route cost is unchanged.

Density can be measured in several ways, such as deliveries per square kilometre, stops per route-hour, or parcels per kilometre. No single measure is sufficient. A dense apartment district may have high stops per kilometre but long access and elevator times; a suburban route may have lower geographic density but rapid doorstep service.

4. Routing Policies and Optimization

4.1 The routing problem

The basic vehicle-routing problem assigns customers to vehicles and determines visit sequences while respecting capacity and operational constraints. Real last-mile routing commonly involves:

• vehicle capacity by weight and volume;
• driver shift limits and breaks;
• depot start and end requirements;
• customer time windows;
• product compatibility and temperature requirements;
• road restrictions, tolls, and vehicle access rules;
• delivery priorities and promised service levels;
• driver skills or security requirements;
• uncertain traffic and service times.

A simplified objective minimizes total route cost:

Z is total travel cost; c_ij is the cost of travelling from node i to node j; x_ijk equals 1 if vehicle k travels from i to j, and 0 otherwise. This is only a core objective. A practical model adds constraints for every customer being served once, vehicle capacity, route continuity, depot rules, and time windows.

A time-window constraint can be expressed conceptually as:

where t_i is the service-start time at customer i, and a_i and b_i are the earliest and latest permitted times. If travel from i to j takes τ_ij and service at i takes s_i, then a feasible sequence requires:

These relationships assume deterministic or planned travel and service times. In practice, buffers or stochastic optimization are required because traffic and customer interaction are uncertain.

4.2 Practical routing policies

Common policies include:

• Static routing: routes are planned before dispatch and changed rarely. It is simple and predictable but less responsive.
• Dynamic routing: routes are updated using new orders, traffic, cancellations, and driver status. It improves responsiveness but can increase driver workload and customer uncertainty.
• Zone routing: geographic areas are assigned to drivers or depots. It supports local knowledge but may ignore changing demand density.
• Priority routing: urgent, premium, perishable, or medically important orders receive preferential treatment.
• Milk-run routing: one vehicle visits multiple planned stops, useful for store replenishment and pickup-point servicing.
• Hybrid routing: scheduled routes handle predictable demand while flexible capacity handles overflow.

Optimization should not minimize distance alone. A route with slightly greater distance may produce lower total cost if it reduces parking delays, missed windows, overtime, or failed delivery.

5. Delivery Windows and Service-Level Design

Delivery windows are a form of customer segmentation and demand management. Narrow windows increase convenience but make route construction more difficult. Broad windows improve consolidation and utilization but reduce customer flexibility.

A retailer may offer:

• economy delivery with a broad window;
• nominated-day delivery;
• evening or weekend delivery;
• same-day delivery within a defined zone;
• premium one- or two-hour windows;
• pickup from a store, locker, or partner point.

The operational promise must be defined precisely. “On time” could mean arrival within the window, attempted delivery within the window, or delivery by the promised date. Performance reports should state the denominator and treatment of customer-caused exceptions.

A basic service-level measure is:

SL is the on-time service level; N_on-time is the number of eligible deliveries meeting the agreed promise; and N_eligible is the total number of eligible deliveries. Exclusions should be governed transparently. Excessive exclusions make a high service level meaningless.

Managers should also distinguish reliability from speed. A two-day service delivered consistently may create more trust than a nominal same-day service that frequently misses its promise. Service-level agreements should specify delivery, scan, tracking, claims, data, and escalation requirements.

6. Parcel Strategies and Delivery Alternatives

6.1 Parcel design

Parcel strategy links packaging, handling, carrier tariffs, damage risk, and customer experience. Excess packaging increases material use, dimensional weight, and disposal burden. Insufficient packaging increases damage and returns. Important decisions include carton-size rationalization, right-sizing, reusable packaging, consolidation, barcode quality, tamper evidence, and separation of hazardous or temperature-sensitive products.

Carriers may charge by actual weight or dimensional weight. A generic dimensional-weight rule is:

where L, W, and H are package dimensions in the carrier’s specified units, K is the carrier’s dimensional divisor, and W_d is dimensional weight in the corresponding weight unit. The billable weight is usually the greater of actual weight and dimensional weight, subject to the carrier’s tariff. Because divisors and units vary, the contract must be checked before calculation.

6.2 Lockers and pickup points

Lockers and pickup points consolidate deliveries and reduce the number of home stops. They can improve first-attempt success, customer control, and route density. They also require suitable location coverage, opening hours, security, accessible design, capacity planning, replenishment, and a process for uncollected parcels.

Pickup networks are particularly useful where homes are insecure, apartment access is difficult, customers work away from home, or curb access is restricted. However, convenience is heterogeneous: some customers value flexibility, while others perceive collection as an unwanted transfer of cost and effort.

6.3 Click-and-collect and ship-from-store

Click-and-collect uses stores as customer-service and fulfilment nodes. It can reduce last-mile cost, stimulate additional store purchases, and use existing real estate. Its risks include inaccurate inventory records, staff congestion, picking errors, limited storage, and inconsistent handover experiences.

Ship-from-store may shorten distance, but the apparent benefit can be offset by fragmented inventory, high store labour cost, and the opportunity cost of selling stock reserved for walk-in shoppers. The correct evaluation compares total incremental cost and service, not just kilometres saved.

6.4 Crowdsourced delivery

Crowdsourced delivery uses independent individuals or flexible contractors to perform deliveries, often through a digital platform. It can provide surge capacity and short-distance flexibility. Governance concerns include worker classification, pay transparency, insurance, safety, identity verification, data protection, and service consistency. The retailer remains responsible for ensuring that its customer promise and legal obligations are met, even when execution is outsourced.

7. Failed Delivery, Returns, and Reverse Logistics

A failed delivery may result from customer absence, incorrect address, access restrictions, unsafe location, damaged parcel, payment issue, or carrier error. The cost can include a reattempt, depot handling, customer-service contact, compensation, return transport, and lost customer trust.

A simple expected failed-delivery cost is:

where p_f is the probability of failure and C_f is the cost conditional on failure. If failure probability is 8% and conditional cost is 10 currency units, expected failure cost is 0.08 × 10 = 0.80 currency units per delivery. This is a simplified estimate; failure costs may vary by location, product, and cause.

Failure reduction measures include address validation, delivery instructions, customer-selected windows, real-time notifications, safe-place governance, signature rules, lockers, pickup points, and driver training. Firms should analyse failures by cause rather than treating all failures as carrier underperformance.

Returns are also a customer-experience process. Convenient returns may increase conversion but raise transport, inspection, repackaging, and disposition costs. Return design should distinguish resale, refurbishment, liquidation, recycling, and disposal pathways. Product categories with high return rates require special attention to reverse capacity and fraud controls.

8. Urban Logistics and Sustainability

Urban logistics operates within competing claims for road space, air quality, safety, accessibility, retail vitality, and delivery convenience. Solutions include:

• urban consolidation centres;
• micro-fulfilment centres;
• parcel lockers and neighbourhood pickup points;
• delivery time coordination;
• cargo bikes and electric vans;
• off-peak delivery where permitted;
• curb-management reservations;
• consolidated store replenishment;
• mobile depots or transshipment hubs.

Electrification can reduce tailpipe emissions and noise, but lifecycle impact depends on vehicle manufacture, electricity source, battery production, utilization, and replacement. Sustainability should therefore be measured across the relevant system boundary rather than inferred from vehicle type alone.

A basic emissions estimate is:

where E is estimated emissions, D is activity such as vehicle-kilometres or tonne-kilometres, and EF is the applicable emissions factor per activity unit. The formula is valid only when the emissions factor matches the vehicle, fuel, load, geography, and accounting boundary. For route comparison, managers should specify whether empty travel, warehouse energy, packaging, and customer travel to pickup points are included.

Consolidation often reduces emissions, but not always. A customer driving several kilometres to collect one parcel may offset some carrier savings. Analysis should consider the whole customer and delivery system.

9. Technology, Data, and Artificial Intelligence

Core technologies include transportation-management systems, route optimization, barcode and RFID scanning, telematics, geocoding, proof-of-delivery applications, electronic signatures, customer-notification platforms, and carrier APIs. These systems improve visibility only when data definitions, event timing, and exception ownership are consistent.

Artificial intelligence and machine learning can support demand forecasting, estimated-time-of-arrival prediction, delivery-density mapping, address correction, fraud detection, and dynamic dispatch. Their limitations include biased historical data, unstable traffic conditions, poor address quality, model drift, opaque recommendations, and over-optimization against narrow metrics.

Human oversight remains essential when decisions affect accessibility, worker safety, customer eligibility, or service exclusion. A route algorithm should not systematically deprioritize remote, low-income, elderly, or mobility-impaired customers merely because they are costly to serve. Governance should include model monitoring, explainability appropriate to the decision, audit trails, cybersecurity, and a process for correcting errors.

10. Carrier Performance Management

A balanced carrier scorecard should combine cost, service, quality, flexibility, sustainability, and compliance.

Metrics require careful definitions. For example, first-attempt success should distinguish customer-requested rescheduling from carrier failure. A low cost per parcel may conceal poor service, while excellent on-time performance may be purchased at unsustainable premium cost. Scorecards should therefore be linked to root-cause analysis and continuous-improvement plans, not used only for punishment or annual renegotiation.

Applied Case: Northstar Market’s Urban Delivery Redesign

Background

Northstar Market, an illustrative omnichannel grocery and general-merchandise retailer, operates three metropolitan stores and one regional fulfilment centre. It offers free next-day home delivery and paid two-hour delivery. Home delivery demand has grown rapidly, but customer complaints and delivery losses have increased.

Problem

The company’s average route cost is 240 currency units. A planned route contains 32 stops, but only 28 are completed on average. Four stops fail because of absence, inaccurate addresses, or building-access problems. Two-hour orders are frequently dispatched as separate routes, producing low vehicle utilization. Management is considering lockers, pickup points, electric cargo vehicles, and a second carrier.

Analysis

The current cost per successful delivery is:

The problem is not simply that routes are expensive. The firm has insufficient density in premium windows, weak address and access data, and no structured alternative for customers who are not home. It also measures carrier cost per route rather than cost per successful delivery.

Approach

Northstar introduces four service tiers:

1. broad-window economy delivery for low-urgency baskets;
2. nominated-window delivery with a delivery fee during low-density periods;
3. locker and pickup-point collection near transit stations and workplaces;
4. two-hour delivery limited to dense zones and minimum basket values.

It adds address validation, building-access instructions, customer reminders, and a second carrier for peak periods. A routing system groups orders by geographic density and time window. Electric cargo vehicles are piloted in the central district, while conventional vehicles serve lower-density suburbs. Carrier contracts include first-attempt success, tracking completeness, claims, and emissions reporting.

Outcome

After a pilot period, the retailer reports fewer reattempts, higher route productivity, and improved delivery-window reliability. The result is not attributed to one technology. It arises from aligning the promise with demand density, giving customers alternatives, consolidating suitable orders, and measuring successful delivery rather than planned stops.

Learning points

• Premium speed should be geographically and economically constrained.
• Pickup alternatives can improve both customer control and route economics.
• A carrier decision must include failure, claims, data, and resilience—not price alone.
• Measurement design can change managerial behaviour: cost per successful delivery is more informative than cost per route.
• Urban sustainability requires network redesign as well as vehicle substitution.

Challenges, Ethics, and Regulation

Transportation decisions have distributive consequences. Dynamic pricing may exclude customers who cannot pay for convenient windows. Automated delivery restrictions may disadvantage remote communities. Locker placement can improve access in one neighbourhood while creating surveillance or accessibility concerns in another. Firms must consider disability access, language, safe delivery, worker safety, and nondiscrimination.

Privacy governance is essential because delivery systems process addresses, geolocation, signatures, images, contact details, and behavioural data. Data should be collected for legitimate purposes, protected against unauthorized access, retained only as necessary, and shared with carriers under clear contractual controls.

Labour governance includes fair contracting, reasonable productivity expectations, breaks, safety training, injury protection, and transparent algorithmic management. Regulatory requirements vary by jurisdiction and may cover vehicle access, emissions, working time, consumer rights, packaging, dangerous goods, and data protection. Managers should obtain jurisdiction-specific legal advice rather than assuming that a platform or carrier transfers all responsibility.

Best-Practice Framework

A graduate-level manager can apply the following sequence:

1. Define customer segments and the service promise.
2. Map the network, order flows, inventory nodes, and delivery alternatives.
3. Measure density, route productivity, failure causes, and true cost.
4. Segment products by urgency, size, value, temperature, and handling risk.
5. Compare carriers and modes through weighted service-and-cost criteria.
6. Design delivery windows that support consolidation without undermining convenience.
7. Use routing optimization with realistic capacity, access, and time constraints.
8. Provide lockers, pickup points, or store collection where they create mutual value.
9. Pilot crowdsourcing and alternative vehicles with explicit governance controls.
10. Monitor service, cost, quality, emissions, equity, and resilience together.
11. Review exceptions and customer complaints as process data.
12. Recalibrate promises and network design as density, technology, and regulation change.

Chapter Summary

Transportation and last-mile delivery are strategic components of omnichannel retail rather than isolated dispatch activities. Mode and carrier selection must balance speed, flexibility, cost, risk, capacity, sustainability, and legal responsibility. Last-mile economics are shaped by density, route time, parcel characteristics, access conditions, delivery windows, and failed attempts.

Routing problems require assignment and sequencing under capacity, time-window, vehicle, and labour constraints. Optimization should pursue total service value rather than distance reduction alone. Lockers, pickup points, click-and-collect, ship-from-store, crowdsourced delivery, cargo bikes, electric vehicles, and urban consolidation can improve performance when matched to local demand and customer preferences.

Performance management should use transparent metrics such as on-time service, first-attempt success, cost per successful delivery, route productivity, damage, visibility, resilience, and emissions. Technology and AI can improve prediction and coordination, but they require data quality, human oversight, cybersecurity, privacy protection, and fairness controls. The most effective strategy integrates customer promise, network structure, execution capability, and responsible governance.

Key Terms

Transport mode: A method of moving goods, such as road, rail, air, sea, or cargo bicycle.

Line haul: High-volume movement between major logistics nodes.

Middle mile: Movement between fulfilment centres, stores, depots, or cross-docks.

Last mile: Final movement to the customer or collection location.

Carrier: A firm that transports goods for a shipper or retailer.

Delivery density: Concentration of deliveries within a geographic or time area.

Time window: The period within which delivery is promised or permitted.

Vehicle-routing problem: The assignment and sequencing of vehicle visits subject to operational constraints.

Dynamic routing: Route adjustment in response to new information or changing demand.

First-attempt success: Completion of delivery during the first attempted visit.

Dimensional weight: Freight weight calculated from package volume under a carrier tariff.

Pickup point: A staffed third-party or retail location where customers collect parcels.

Parcel locker: A secure self-service compartment for parcel collection or returns.

Crowdsourced delivery: Delivery performed through a platform using flexible independent capacity.

Urban consolidation centre: A facility that consolidates freight before coordinated urban delivery.

Micro-fulfilment centre: A small, strategically located facility designed for rapid local order fulfilment.

Proof of delivery: Evidence that a delivery was completed, such as a scan, signature, or image.

Service-level agreement: A contractual specification of required service performance.

Reverse logistics: Movement and processing of returns, repairs, recovery, recycling, or disposal.

Route productivity: Completed deliveries or other output per unit of route time or capacity.

Review Questions

Short Answer Questions

1. Why is last-mile delivery usually more operationally complex than line-haul transportation?
2. Identify four criteria that should be included in carrier selection.
3. What is the difference between a delivery window and a service-level measure?
4. Explain how lockers can affect both customer experience and delivery economics.
5. Why should managers measure cost per successful delivery rather than cost per route alone?

Long Answer Questions

1. Compare road, rail, air, sea, and cargo-bicycle transportation for omnichannel retail applications.
2. Explain the principal constraints in a vehicle-routing problem with time windows.
3. Evaluate the strategic advantages and limitations of crowdsourced delivery.
4. Discuss how delivery density, packaging, failed delivery, and returns influence last-mile profitability.
5. Develop a balanced carrier-performance scorecard for a large omnichannel retailer.

Analytical / Application Questions

1. A route costs 300 currency units and is planned for 40 stops. Only 34 deliveries succeed. Calculate the cost per successful delivery and explain two operational interventions that could improve it.
2. A retailer must choose between a broad four-hour delivery window and a narrow one-hour window. Analyse the likely effects on customer value, route density, failure risk, and carrier cost.
3. Design a pilot for lockers and pickup points in a congested city. Specify the target customer segment, network criteria, measures of success, risks, and governance controls.

Multiple Choice Questions with Answers

Question 1

Which factor most directly improves last-mile cost efficiency through consolidation?

A. Higher delivery density

B. More narrow delivery windows for every customer

C. More individual routes with fewer stops

D. Increasing empty vehicle kilometres

Correct Answer: A. Higher delivery density

Question 2

A route costs 180 currency units and completes 25 successful deliveries. What is the cost per successful delivery?

A. 5.20 currency units

B. 6.00 currency units

C. 7.20 currency units

D. 8.00 currency units

Correct Answer: C. 7.20 currency units

Question 3

In a weighted carrier score, what must be true of the weights?

A. They must all be identical

B. They must sum to 1

C. They must be measured in kilometres

D. They must exclude service criteria

Correct Answer: B. They must sum to 1

Question 4

Which is a principal advantage of parcel lockers?

A. They guarantee zero transportation cost

B. They eliminate all reverse logistics

C. They consolidate deliveries and can improve first-attempt success

D. They remove the need for customer communication

Correct Answer: C. They consolidate deliveries and can improve first-attempt success

Question 5

In a vehicle-routing model, a time-window constraint primarily specifies:

A. The maximum parcel weight

B. The permitted service period at a customer

C. The carrier’s annual revenue

D. The package’s dimensional divisor

Correct Answer: B. The permitted service period at a customer

Question 6

If a route cost is 240 currency units and 28 deliveries succeed, the approximate cost per successful delivery is:

A. 6.57 currency units

B. 7.86 currency units

C. 8.57 currency units

D. 9.60 currency units

Correct Answer: C. 8.57 currency units

Question 7

Which metric most directly measures whether a delivery succeeds without reattempt?

A. First-attempt success rate

B. Vehicle purchase price

C. Dimensional divisor

D. Warehouse square metres

Correct Answer: A. First-attempt success rate

Question 8

What is a major limitation of crowdsourced delivery?

A. It can never provide surge capacity

B. It raises governance concerns involving labour, safety, insurance, and consistency

C. It is restricted to rail transportation

D. It eliminates the need for delivery data

Correct Answer: B. It raises governance concerns involving labour, safety, insurance, and consistency

Question 9

Dimensional weight is primarily based on:

A. Package volume and a carrier-specific divisor

B. Customer income and delivery distance

C. Driver tenure and route duration

D. The number of product reviews

Correct Answer: A. Package volume and a carrier-specific divisor

Question 10

Why should AI-based routing systems retain human oversight?

A. Algorithms cannot process any operational data

B. Human oversight helps address bias, safety, explainability, and exceptional circumstances

C. Routing never involves constraints

D. AI systems are legally required to minimize distance only

Correct Answer: B. Human oversight helps address bias, safety, explainability, and exceptional circumstances