XPO

In less-than-truckload (LTL) freight, improper trailer loading is a persistent source of damage claims, safety incidents, and costly re-handling. XPO's dockworkers had for years used company-issued handheld devices to photograph every loaded trailer before closing the door — a standard procedural step across its service center network. But those photos served only as a reactive record after the fact. There was no mechanism to analyze images in real time and alert workers to loading deficiencies while the door was still open. At XPO's scale as a major LTL carrier operating dozens of service centers nationwide, even a small rate of improperly secured freight translated to meaningful damage claims and operational disruption across the network.

XPO built a computer vision model trained on its own accumulated library of trailer photos — an asset generated organically through years of routine dock operations. The model applies established freight-securing criteria to each new image, evaluating whether the trailer meets all required steps, checks, and procedures before the door can be closed. When the system detects a deficiency, it surfaces real-time guidance directly to the dockworker's handheld device, explaining what the issue is and why the door cannot be shut. Because the AI integrates into the existing photo-capture workflow rather than replacing it, deployment friction is minimal — workers use the same devices and the same process they already follow. The initiative launched as a pilot across all XPO service centers, with VP of Technology Erin Goheen presenting the effort at the SMC³ Jump Start conference in January 2026.

The headline outcome was development velocity: a project that would have required six months to complete two years prior was delivered in weeks, reflecting how XPO's technology teams operate under one-to-two week output targets. The system moved directly into active pilot across the full service center network.

• Development time: Weeks (vs. ~6 months previously)
• Financial leverage: XPO noted that even 1–2% efficiency gains at its operational scale translate to tens of millions to hundreds of millions of dollars in cost savings
• Deployment scope: Active pilot across all XPO service centers

Qualitatively, the system augments existing dock procedures rather than disrupting them, supporting worker adoption by working within the photo-taking workflow already embedded in daily operations.

• Existing operational data assets — even routine process photos — can serve as the foundation for a production-grade AI model, eliminating the cold-start problem.
• Real-time guidance delivered at the point of action (the dock door) is more effective than post-hoc review; intervening before the door closes is far cheaper than resolving a damage claim later.
• Integrating AI into an existing worker workflow, rather than introducing new steps, reduces adoption friction and speeds deployment.
• At enterprise LTL scale, seemingly small efficiency improvements carry outsized financial impact — making incremental AI gains worth pursuing aggressively.
• Short internal delivery timelines (one to two weeks) force teams to prioritize problem clarity over technical ambition, accelerating time-to-value.