According to C.H. Robinson, a United States logistics enterprise, employee productivity has increased by 45 per cent since 2022, when the company promoted AI applications. Against the backdrop of global shipping demand falling from the epidemic, the system has also helped companies to double-digitize their earnings per share since 2023, despite a decline of about 34 per cent over the same period.

Quote process compressed to 31 seconds

This company is the main freight forwarder, especially in case-packing operations. According to Dave Bozeman, CEO of the company, the company now deploys hundreds of AI agents across multiple business lines to handle high frequency, repetitive tasks.

Of these, the client-oriented quotation process is the most typical scenario. Used to take 20 minutes for a single offer to be made manually, the AI agent can now do it in 31 seconds and can run all year round. Bozeman believes that a faster response and more complete presentation of information enhances the willingness of clients to submit requests for quotations and the chances of companies taking orders.

Staff restructuring rather than mass substitution

Bozeman states that the company did not use AI as a tool for direct layoffs, but instead shifted the staff originally responsible for offering offers to higher value-added jobs, such as assisting clients to cope with the changing tariff environment.

However, AI has brought significant human savings. According to the company, the annual rate of natural loss is approximately 11 to 14 per cent, and after AI agents took over part of the process, the enterprise was not required to re-engage its employees on a continuous basis. For some operations, the relationship between increased volume of processing and the size of personnel has been significantly weakened.

Bozeman also wishes to use AI to drive business upgrading, away from acting as a freight forwarder, but to extend to the supply chain by consulting and even taking on more complete supply chain functions from clients. At the same time, the company plans to re-engage its SME clients who have lost them in recent years and claims that additional staff are still being recruited in two lines of supply chain consulting and small and medium-sized customer service, although these positions will be staffed by AI assistants.

Self-study models have low usage costs

In terms of cost control, Bozeman states that most AI agents in the company are developed by internal teams and are based mainly on own or open source models rather than on external suppliers. C.H. Robinson currently has approximately 450 engineers, most of whom have a shipping background.

He claimed that this approach had allowed the company to earn hundreds of millions of dollars in business-level gains at a cost of less than $2 million token. According to them, if external agencies want to replicate the system, they may need to coordinate between 15 and 20 different partners.

Bozeman also mentioned that when developing AI agents, companies organize cross-sectoral teams such as engineering, operations, finance and legal services to work together, combe the process and decide which links should be cancelled and which are suitable for automation.

Organizational culture is also included in the reform of AI

In addition to the technical inputs, Bozeman attributed AI advancements to changes in organizational management. In developing the AI system, companies use failure models and impact analysis methods to pre-assess the problems that may arise in the system and to advance their response.

He also asked the team to be more directly exposed to project risks. The internal progress report of the company retains only “green” and “red” status, with no “yellow”. According to him, many “yellow” projects were inherently off-target, although managers were reluctant to explicitly acknowledge them. In this way, companies want to identify problems earlier and to pool resources for correction.

In the case of this logistics enterprise, AI's landings do not depend solely on modelling capacity, but are also closely related to process re-engineering, post adjustment and management mechanisms.