AI is moving into the operating core of logistics networks, shaping how loads are priced, routed and matched to assets. As these systems begin to influence real decisions, companies are restructuring data, governance and partner models to keep control over cost, service and network stability.
In Brief
- Emerging pattern: Routing, pricing and matching decisions are shifting from human dispatch and static rules to optimisation systems embedded in execution workflows.
- Configuration: Companies are rebuilding data layers, decision rights and governance structures so systems can influence port-to-door flows, long-haul freight and asset-heavy networks.
- Constraint: The limiting factor is not AI capability but whether organisations can align incentives and partner economics without destabilising service or capacity.
AI Adoption Is Broad, Decision Control Is Not
Across Ryder System, Full Truck Alliance, Herc Holdings and ZTO Express, AI is no longer confined to pilots or visibility layers. It is being embedded into how loads are matched, routes are designed, assets are deployed and customers are prioritised. The shift is not about dashboards or incremental automation. It is about whether systems can influence the core economic decisions of the network.
What separates these companies is not whether they “have AI,” but how far they allow optimisation systems to shape real operating choices. In practice, that means deciding which loads move on which assets, at what price, under which service commitments, and with which partners. Those decisions sit at the intersection of cost, service and resilience, and they are where AI begins to challenge legacy operating logic.
The distinction between AI, optimisation and analytics is critical. Full Truck Alliance represents the clearest case of agentic systems making end-to-end decisions. AI agents operate across matching, dispatch, pricing, risk and customer service, with a large share of transactions already processed without manual intervention, from voice-based shipper requests through to carrier screening and price negotiation.
Ryder and ZTO Express are embedding optimisation into execution layers, where AI influences decisions but does not fully automate them. Ryder is integrating decision support into its RyderShare and RyderGyde platforms, shifting from visibility to proactive exception management and network orchestration. ZTO is applying route optimisation and demand forecasting across a dense, hub-and-linehaul network, reducing transport unit costs while maintaining service performance under regulatory pricing constraints.
Herc sits further upstream. Its ProControl platform provides telemetry on asset location, utilisation and access, feeding decisions on fleet allocation, disposals and capital timing. This is not autonomous routing, but disciplined asset control. The role of analytics here is to constrain decision-making, not automate it.
This spectrum matters because the constraint is not technical capability. It is how much decision authority organisations are prepared to hand over to systems.
The Real Constraint: Data, Incentives and Partner Economics
As AI and optimisation move deeper into operations, the limiting factor is shifting from technology to operating model design. Systems can recommend more efficient routing, pricing and capacity allocation, but acting on those recommendations creates second-order effects that organisations must absorb.
Four conditions are emerging as critical.
First is data integrity. Optimisation systems depend on clean, standardised event data across orders, assets, pricing and partner behaviour. Fragmented inputs or local workarounds undermine model outputs and push teams back toward manual overrides.
Second is decision rights. Organisations need clear rules on when systems act autonomously and when human intervention is required. Without this clarity, optimisation remains advisory, and its impact on cost and service is diluted.
Third is incentive alignment. Planners, dispatchers and commercial teams are often still measured on volume, utilisation or revenue targets that conflict with optimisation logic. If AI recommends rejecting low-quality loads or reallocating capacity away from legacy customers, those decisions must be reinforced, not penalised.
Fourth is partner economics. This is where optimisation becomes most fragile. Systems that maximise network efficiency can unintentionally erode the economics of carriers, franchise operators or last-mile partners. Full Truck Alliance’s use of credit scoring and behavioural penalties reflects an attempt to manage this balance at scale. ZTO’s focus on outlet profitability and fairness points to the same pressure. Ryder and Herc face it in different forms through dealer networks and customer relationships tied to asset availability.
These constraints are already shaping outcomes. Full Truck Alliance has prioritised quality and compliance over raw order growth, using AI-driven governance to remove low-quality transactions and improve fulfilment rates. ZTO is balancing efficiency gains in routing with a mix shift toward higher-cost, higher-service shipments under regulatory pressure. Ryder and Herc are using data-informed pricing and utilisation to protect margins while holding latent capacity for a potential demand recovery, even at the cost of near-term volume expansion.
In each case, optimisation is working but only within the limits set by governance, incentives and partner tolerance.
From Optimisation to Organisational Design
The structural shift underway is not simply the adoption of AI tools. It is the gradual transfer of decision influence from humans to systems that learn from execution data at scale. That transfer forces changes in how networks are designed and governed.
Fleet composition, contract structures and partner models are increasingly being shaped by what optimisation systems can exploit. Assets are positioned where demand is predictable and data density is high. Contracts are being written to allow dynamic pricing and flexible capacity allocation. Partner relationships are evolving toward models that reward behaviours systems can measure and reinforce.
Resilience is also being redefined. It is no longer just a function of redundant capacity or alternative routes. It depends on how quickly optimisation layers can detect disruption, re-learn and adapt decisions in real time.
The implication is not that AI will replace human control, but that control itself is being restructured. Organisations that treat optimisation as a tool will struggle to capture its full value. Those that redesign decision rights, incentives and partner frameworks around how these systems operate are more likely to convert AI from a capability into a durable operating advantage.
The next phase of competition will not be defined by who deploys AI in logistics, but by who can let it operate, without destabilising the network it is meant to optimise.
This article is based on recent earnings reports and public disclosures from the companies referenced.
