Every fleet manager would agree that the primary goal of their fleet safety strategy is the prevention of motor vehicle crashes to protect drivers and the public from injury or death. But increasingly, fleet safety is tied directly to a company’s bottom-line health through compliance, legal, and insurance considerations.

Transportation companies that neglect to champion a proactive fleet safety culture not only put their drivers and vehicles at risk, but they leave themselves open to profit-crushing fines, legal fees, nuclear verdicts, and escalating insurance premiums.

Profit margins under siege

Without a robust fleet safety model, transportation companies risk violating Federal Motor Carrier Safety Administration (FMCSA) regulations, such as Hours of Service (HOS), vehicle technology standards, and rules regarding inspection, maintenance, and repair. These types of FMCSA violations can trigger costly operational disruptions and hefty fines.

In addition to non-compliance penalties, fleets must deal with the bottom-line impact of truck crashes. The FMCSA estimates that fatal large truck crashes cost $13.8 million per crash, with non-fatal crashes costing more than $400,000 per incident.

Motor vehicle accidents create a cost-amplifying ripple effect across the organization, driving costs up through reduced productivity, vehicle repairs, and the need to recruit temporary or new drivers. Plus, the blow to a company’s reputation can compromise new client acquisition and driver retention and recruitment.

The crippling impact of nuclear verdicts

The costliest ramifications of inadequate driver safety programs and increased crash rates stem from legal fees, jury awards, and expensive insurance premiums. A 2025 study of nuclear verdicts—awards greater than $100 million—found that the trucking and automotive industries are among the top targets of nuclear verdicts, mainly in wrongful death and negligence cases. These sectors faced 15 multimillion-dollar verdicts in 2024, totaling more $1.4 billion.

Similarly, an ATRI 2025 report concluded that while the number of awards in trucking litigation are increasing in general, the upper 50% of awards are increasing at a particularly alarming rate. Notably, from 2012 to 2020, the number of cases with verdicts over $1 million increased by 235% compared to the six years prior, with the average size of a crash-related verdict increasing by a staggering 967% between 2010 and 2018—and this upward trend shows no sign of abating.

Insurance premiums pounding profits

While nuclear verdicts can damage a transportation company’s financial health beyond recovery, the fallout extends to insurance premiums. Escalating claim severity, rising 93.5% between 2015 and 2024, is driving insurance premiums to new heights.

Recent ATRI research shows that liability insurance premium costs rose by 18.6% from 2021 to 2024, outpacing consumer inflation by 5.4 percentage points even while heavy-duty truck crash rates fell by 2.6% industry-wide.

Mitigating risk with predictive fleet safety

Facing the costly prospect of nuclear verdicts and climbing insurance rates, many transportation companies are re-evaluating their fleet safety strategies, with a focus on protecting both drivers and the bottom line through a more proactive, predictive approach.

On the legal front, plaintiff attorneys are increasingly evaluating whether a fleet can show a pattern of proactive risk identification, coaching, and intervention before an incident occurs. Similarly, juries are influenced by whether fleets can prove they took reasonable, documented steps to prevent foreseeable risk.

Consequently, it’s no surprise that fleets operating with reactive or inconsistent driver safety practices face significantly higher exposure. Preventing nuclear verdicts is no longer solely a legal strategy; it’s a safety strategy, and one that requires a proactive approach to risk mitigation.

Insurance premiums are influenced by similar considerations. Moving forward, the affordability and availability of insurance will be determined by data transparency and the ability to use data in insurance costing. Indeed, insurers are beginning to reward fleets that can show measurable reductions in risky driving behavior using predictive safety models.

Protecting the bottom line with data

In today’s transportation landscape, forward-thinking fleets recognize that reactive driver safety that typically rely on lagging indicators (e.g., compliance violations, incidents, claims) and in-cab cameras and video telematics is no longer sufficient for reducing risk and curtailing costs.

Profitable fleets are thinking beyond simple scorecards, arbitrary weighting, and reactive training. They’re building a proactive safety culture underpinned by a data-driven predictive driver safety program that integrates data from a wide range of sources: cameras, telematics, electronic logging devices (ELDs), FMCSA, customer feedback, training, dispatch, HR systems, and accidents and claims.

Using predictive analytics and machine learning to analyze billions of miles of driving data and hundreds of thousands of historical crashes across the industry, AI-enabled fleet safety programs can identify elevated risk earlier, prioritize interventions, and demonstrate continuous improvement. Fleet leaders can use these safety insights to proactively manage risk, intervening with at-risk drivers to provide meaningful coaching before a crash occurs.

In this era of nuclear verdicts, eye-watering insurance premiums, and razor-thin margins, safety has become a competitive cost advantage. By shifting from reactive safety models to investment in predictive analytics, driver development, and documented preventive practices, fleets can better safeguard drivers from crash risk while protecting themselves from costly litigation and strengthening their insurance position to keep premiums under control.

Hayden Cardiff, VP Safety Solutions at Descartes

The post Predictive Fleet Safety: Protecting Drivers and the Bottom Line appeared first on Logistics Viewpoints.

Calling logistics a network is accurate, but it is not sufficient. A network describes connections; it does not explain how independently managed transportation, warehousing, fulfillment, yard and dock operations, last-mile delivery, technology, partners, and people combine to produce one customer outcome. The more useful model is a system of systems.

The practical consequence is important. You cannot understand logistics performance by looking at any one component in isolation.

Every Function Is Rational. The System Still Can Be Wrong.

A warehouse is optimized around throughput, storage, labor, service requirements, and physical constraints. A transportation operation is optimized around modes, routes, carrier capacity, cost, service, and variability. A yard operation is optimized around appointments, dwell, doors, and trailer flow. None of those perspectives is wrong. The difficulty appears when they are connected.

None of those perspectives is wrong. The difficulty appears when they are connected.

Consider a seemingly simple promise to offer later customer order cutoffs. Commercially, it may be attractive. Operationally, it can change picking waves, labor schedules, carrier tender timing, dock congestion, linehaul departure, delivery commitments, and exception management. The decision spans multiple systems, and the value appears only if those systems can absorb the change together.

A system-of-systems view forces the organization to see that dependency before the change is made.

The Network Does Not Report to You

Traditional engineering often assumes a designer has meaningful control over the system being built. Logistics networks rarely offer that luxury.

Carriers have their own network economics. 3PLs optimize their facilities and labor. Parcel providers manage sort capacity and delivery density. Ports and terminals manage gates, berths, yards, and appointments. Customers change ordering behavior. Software providers determine product road maps. Regulatory agencies change requirements. Labor markets move independently of corporate plans.

These entities participate in the same operating system without being subordinate to a single designer. That is one of the defining characteristics of a system of systems. Its components can operate independently, evolve independently, and still affect the performance of the whole.

This helps explain why seemingly small external changes can create disproportionate logistics effects. A carrier-capacity shift can alter fulfillment strategy. A missed linehaul departure can invalidate an otherwise efficient warehouse wave. A new customer cutoff can create technology and process work across order release, picking, staging, tendering, and delivery.

Logistics is not simply complicated. It is adaptive.

Complexity Raises the Price of Weak Architecture

When systems are relatively simple, coordination can rely heavily on experience and informal processes. As complexity increases, that becomes less reliable.

Architecture provides structure.

Process architecture defines how work moves across the enterprise. Data architecture defines how information is created, governed, shared, and interpreted. Decision architecture defines who or what makes decisions, what inputs are used, how quickly decisions must be made, and when escalation is required. Technology architecture provides the applications, integration, analytics, and automation that support those processes and decisions.

These architectures overlap. They should.

The mistake is to allow one of them, usually technology architecture, to become the de facto operating model. When that happens, organizations begin designing work around system capabilities instead of designing systems around business requirements. The result may be technically integrated while remaining operationally fragmented.

Treat Every Handoff as a Design Decision

In a system of systems, interfaces are not plumbing. They are part of the product.

The interface between order management and fulfillment determines whether customer promises are executable. The interface between a shipper and a carrier determines whether capacity commitments are reliable. The interface between an optimization engine and a TMS or WMS determines whether a recommendation can actually be executed. The interface between an AI agent and a human determines whether automation accelerates decisions or simply creates another queue of suggestions.

These interfaces should have explicit requirements.

What information must cross the boundary? At what frequency? With what latency? Who owns data quality? What happens when the message is incomplete? Which decisions can be automated? What happens when two systems disagree?

Organizations frequently discover these questions during implementation. Systems engineering argues that they should be part of design.

Measure the Enterprise Outcome, Not the Local Win

A system-of-systems view also changes performance measurement.

Functional metrics remain necessary, but they are insufficient. A warehouse can hit its productivity target while order cycle time gets worse. Transportation can reduce cost per shipment while dock dwell or delivery variability increases. A parcel operation can lower rate per package while missed cutoffs rise. The larger question is whether the total logistics system is improving service, cost, flow, and responsiveness together.

The larger question is whether the total system is producing the outcomes the business requires.

That means metrics should connect across levels. Local operating measures should roll into end-to-end logistics measures such as on-time delivery, perfect-order performance, order cycle time, cost per shipment or order, dock dwell, capacity utilization, responsiveness, resilience, and decision speed. The organization needs to understand where local gains create system gains and where they simply move cost, delay, or risk somewhere else.

Change the Question, Change the Design

The system-of-systems idea may sound theoretical, but it is a useful operating model for logistics leaders. It explains why local optimization is dangerous, why interfaces matter, why technology integration alone does not create end-to-end performance, and why transformation requires architecture across organizational boundaries.

The practical shift is straightforward: stop asking whether each function is performing well in isolation and ask whether the combined system is producing the outcome the enterprise needs. In 1.3, that system view becomes operational: we move from understanding the whole to defining what the whole must actually do before technology enters the discussion.

Related Logistics Viewpoints research

Systems Engineering in Logistics
The New Architecture of Logistics
2026 Supply Chain Decision Intelligence Market Map
Guest Commentary: Control Tower 2.0 – Managing Logistics Costs in a Risky and Volatile World
Previous in this series: Why Logistics Needs Systems Engineering

The post Stop Managing Logistics as a Collection of Functions appeared first on Logistics Viewpoints.

Uber’s decision to reduce its workforce by about 10% will inevitably be discussed as another large technology-company layoff. For logistics executives, however, the more important question is not whether AI eliminated these jobs. It is whether AI is beginning to eliminate some of the organizational structures that made many of those jobs necessary.

Reuters reported that approximately 3,300 positions will be affected in Uber’s largest workforce reduction since the pandemic. CEO Dara Khosrowshahi said the company is removing management layers, simplifying team structures, clarifying ownership and redirecting investment toward its largest opportunities. Uber itself describes the objective as becoming a “simpler, faster” company. Importantly, Khosrowshahi did not attribute the workforce reduction directly to artificial intelligence.

That distinction matters because the deeper logistics story is not simply about automating individual jobs. It is about reducing the coordination required to make an increasingly complex enterprise operate.

The Coordination Tax

Large organizations accumulate complexity almost naturally. Products, geographies, customers and channels multiply, followed by planners, analysts, supervisors, project managers and functional organizations to manage them. Each addition can make sense individually while the cumulative result is an organization in which a surprising amount of work consists of coordinating other work.

Uber is explicitly attacking that problem. The company says growth brought more layers, more coordination and increasingly fragmented ownership. It is broadening management spans, reducing positions concentrated on coordination and eliminating many “micro-teams” with only one or two reports. Uber says the number of employees seven or more organizational layers below the CEO will decline by 20%, while the number of micro-teams will fall by nearly half.

Anyone who has worked around a large logistics organization should recognize the pattern. A routine transportation exception can generate an alert, an email, a carrier call, another information request, a supervisor escalation, a customer update and eventually a KPI entry explaining what happened. No single step is necessarily unreasonable. The inefficiency lies in the number of people and systems through which information must travel before somebody has enough context and authority to act.

AI agents begin to change that equation. An agent can monitor a transaction, identify an exception, gather contextual information, consult business rules, communicate with other systems, recommend an action and, within defined guardrails, execute it. The opportunity is therefore larger than making every planner or analyst incrementally faster. In some workflows, the larger gain comes from eliminating the handoffs themselves.

This Is a Systems Engineering Problem

This is why I believe much of the current discussion around AI in logistics remains too narrow. We tend to evaluate individual technologies when the more important issue is how those technologies interact with people, physical assets, information flows, decision rights and business processes.

That is the central argument in our recent white paper, Systems Engineering in Logistics. Logistics performance does not emerge from a TMS, WMS, control tower, robotics platform or AI model operating independently. It emerges from the behavior of the larger system.

Uber’s restructuring is a useful real-world example. The company is not simply deploying another AI application. It is reconsidering management spans, operating structures, accountability and capital allocation while introducing increasingly capable digital systems.

The logistics industry has spent decades digitizing individual functions. Transportation received a TMS, warehousing received a WMS, planning acquired specialized applications, customer service adopted CRM platforms, and visibility produced control towers. The technology architecture became more sophisticated, but the organizational architecture often remained substantially unchanged.

AI provides an opportunity to revisit that architecture. If systems can increasingly exchange information, interpret events and execute routine decisions without waiting for a human intermediary, logistics leaders should ask more than, “Which tasks can AI automate?”

A better question is: “Which organizational boundaries exist because humans historically had difficulty coordinating information and decisions across them?”

Uber Freight Is Already Showing Us the Model

Uber’s own freight business provides a concrete example of what that transition looks like.

In its second-quarter 2026 prepared remarks, Uber said Freight is investing in AI capabilities designed to optimize transportation decisions for customers. The company specifically identified earlier detection of shipment risks and automation of routine operational workflows, including responding to shipment inquiries, validating documents and correcting shipment data.

Those may appear to be incremental applications, but they target precisely the activities that generate administrative work throughout transportation organizations. Every automatically resolved shipment inquiry or corrected document can remove an email, a queue, a handoff or an escalation.

The progression matters. The first wave of generative AI in business was mainly about individual productivity: write an email faster, summarize a document, generate code, help an analyst find an answer. The next wave connects AI to workflows, enterprise data, APIs and business rules.

At that point, AI stops being merely a productivity application and becomes part of the operating model.

Logistics is particularly exposed to this transition because freight procurement, appointment scheduling, track-and-trace, carrier communication, invoice reconciliation, inventory exceptions and delivery management all depend on large volumes of structured information moving among organizations and systems.

That is fertile ground for agentic automation.

Uber Is Also Betting on Physical Autonomy

There is another dimension to the Uber story. While simplifying its human organization and expanding AI use, the company is simultaneously making a very large investment in autonomous transportation.

Uber said in its Q2 2026 prepared remarks that it expects to commit more than $10 billion over the coming years through equity investments, infrastructure and vehicle commitments intended to bring autonomous vehicles to market at scale. Autonomous vehicles were already live on Uber in seven cities, the company said, with as many as 15 expected by year-end. Its partners have committed approximately 120,000 vehicles to the Uber network over the coming years.

What is particularly interesting is how Uber defines its role. The company does not need to manufacture every vehicle or develop every autonomous-driving system. Instead, it can provide demand aggregation, dispatch intelligence, vehicle integration, fleet operations, charging infrastructure, financing, insurance and regulatory relationships around an ecosystem of partners.

Uber is increasingly positioning itself not simply as a transportation marketplace, but as an orchestration layer across digital and physical transportation.

That distinction should matter to logistics executives. Autonomous transportation does not end with removing the human driver. The network still requires demand forecasting, capacity allocation, dispatch, maintenance, charging or fueling, customer communication, exception management and financial settlement.

If physical automation develops alongside digital agents capable of coordinating those activities, the operating model changes much more profoundly.

Digital and Physical Autonomy Converge

We are therefore beginning to see two forms of autonomy develop at the same time. Physical autonomy moves vehicles and goods with less direct human operation. Digital autonomy makes and coordinates a growing number of the decisions surrounding those movements.

Consider an autonomous delivery network in which AI agents forecast demand, allocate capacity, reposition vehicles, schedule charging, monitor maintenance, communicate with customers and manage exceptions. Removing the driver is only one component of the automation. Much of the administrative infrastructure surrounding the vehicle can also become increasingly autonomous.

The important development is not any one technology. It is the interaction among them.

That is again a systems-engineering issue.

What Logistics Leaders Should Look For

This does not mean logistics companies should begin eliminating management layers simply because Uber is doing so. Nor does it suggest that human judgment becomes unimportant. The implication is that companies should begin identifying where coordination costs have become embedded in their operating models.

Where does information sit waiting for somebody to move it? Where does an exception pass through several employees before reaching someone with the authority to resolve it? Where are multiple groups maintaining slightly different versions of the same operational truth? Where do recurring meetings exist because underlying systems and decision rights remain poorly integrated?

These are no longer merely process-improvement questions. They are increasingly systems-architecture and AI questions.

The organizations that gain the most from AI may therefore not be those that deploy the largest number of copilots. They may be the organizations willing to redesign processes once the technological limitations that created those processes begin to disappear.

Human expertise remains essential, but its value shifts toward judgment, relationships, system design, accountability, risk management, strategic tradeoffs and genuinely novel exceptions. Routine information gathering, reconciliation and coordination become increasingly machine-assisted or machine-executed.

The likely result is a flatter logistics organization with clearer process ownership, broader spans of control, fewer administrative handoffs and more automated decision execution.

Uber’s restructuring is worth watching because several developments are occurring simultaneously. The company is reducing organizational layers, redesigning operating structures, applying AI to transportation workflows and investing billions of dollars in autonomous mobility.

Viewed independently, each initiative is interesting. Viewed as a system, they point toward something much larger.

The future logistics enterprise may not simply automate more tasks. It may require far fewer layers to coordinate them.

The post Uber’s Restructuring Shows Where AI in Logistics Is Really Going appeared first on Logistics Viewpoints.