IN Brief:
- CJ Logistics America is deploying OneTrack's AiOn across more than 40 warehouses after seven years of working with the supplier.
- The agents connect multiple WMS platforms with enterprise systems, Snowflake data, vision sensors, and warehouse automation.
- CJ and OneTrack report an 18% network-wide increase in units per hour and a 19.7% reduction in aggregate lost time; the figures are supplier and customer reported.
CJ Logistics America is extending OneTrack’s AiOn agentic AI platform across more than 40 warehouses, connecting warehouse-management systems, enterprise data, vision sensors, and automation equipment so software agents can carry out operational work rather than only generate reports. The deployment expands a seven-year relationship between the companies and moves the technology into day-to-day network operations.
CJ Logistics does not run one warehouse-management system across every customer account. Its estate includes several Tier-1 WMS platforms alongside customer-specific systems, with AiOn also connecting to labour and enterprise applications, the company’s Snowflake data warehouse, OneTrack floor sensors, and robotics equipment.
That mixed environment is important for a third-party logistics provider. Individual customers can arrive with their own systems, processes, data definitions, and operating requirements, making network-wide optimisation more difficult than in a warehouse estate built around one standard technology stack.
AiOn is currently being used in four main areas: gap-time tracking, labour-performance management, safety and compliance workflows, and travel, zoning, and slotting optimisation. Agents can combine information from transactions and floor sensors, prepare operating analysis before shifts begin, and generate workflow or compliance documentation.
The companies have also published performance figures from the programme. They report that clock-in and clock-out gap time fell by 45% during the first weeks of one workflow, while units-per-hour performance increased by 18% across the network. CJ Logistics also reports a 19.7% reduction in aggregate network lost time.
Those figures come from CJ Logistics and OneTrack rather than an independent assessment, so they should be treated as company-reported operational results. They nevertheless give the deployment a more useful basis than an AI demonstration built around hypothetical warehouse tasks.
Gap time is a good example of the type of inefficiency conventional reporting can miss. A WMS may record when one putaway finishes and when the next pick begins without explaining why several minutes passed between the two. Combining those transaction records with other operating data gives supervisors a way to identify recurring periods of lost activity rather than relying solely on end-of-shift productivity totals.
Labour management follows a similar pattern. The agents can assemble performance information before supervisors begin coaching or intervention, reducing the manual analysis required to identify which process, employee, or area needs attention. The human decision remains important, particularly where the explanation involves training, equipment, congestion, or other factors that cannot be reduced to one performance number.
For CJ Logistics, a larger benefit could come from reusing those workflows across customer sites. OneTrack says an agent or application created at one facility can be deployed through the common platform across the wider network, avoiding the need to rebuild the same analysis independently for every warehouse.
That does not mean one configuration will fit every building. Customer processes, product dimensions, automation, labour models, and service requirements can differ substantially, and any shared workflow still has to be adapted to the data and operating rules at the relevant site.
The agentic model also creates a governance requirement that is less pressing with a passive dashboard. OneTrack says agents operate within granted permissions and actions are logged for audit. That becomes necessary when software can take a step in a live workflow rather than simply present a recommendation for a person to consider.
Deterministic calculations are another important safeguard. Units per hour, compliance thresholds, and other warehouse measures cannot sensibly be recalculated differently each time a language model receives a question. OneTrack says its approach separates governed operational calculations from the generative interface used to access or configure workflows.
Data remains the less glamorous constraint underneath the system. Different WMS platforms can define tasks and timestamps differently, while sensors create another stream that must be matched to the correct people, locations, and transactions. An AI agent acting quickly on inconsistent source data is capable of accelerating an error just as efficiently as a correct process.
The scale of CJ’s rollout raises the threshold for what can reasonably be described as operational warehouse AI. Connecting more than 40 facilities and allowing agents to participate in labour, slotting, compliance, and engineering workflows exposes the software to measurable consequences if the recommendations or actions are poor.
The next test is persistence. Early productivity improvements can fade after an initial operating change, particularly if staff behaviour changes temporarily under increased scrutiny. The stronger evidence will be whether the reported gains remain visible across different facilities, customers, peaks, and labour conditions.
If they do, the significant change will be less about warehouse staff conversing with AI and more about routine operational analysis no longer waiting for analysts or engineers to prepare it. CJ Logistics is attempting to turn that work into a continuously available operating layer across a multi-customer 3PL network, where the commercial value will ultimately be measured in throughput, labour efficiency, safety, and service performance.
