US logistics company C.H. Robinson stated that since implementing AI applications in 2022, employee productivity has increased by 45%. Against the backdrop of a global shipping demand decline following the pandemic, the system has also helped the company achieve double-digit earnings per share growth since 2023, despite a revenue decline of approximately 34% during the same period.
The quotation process has been reduced to 31 seconds.
This company primarily engages in freight forwarding, especially LCL (Less than Container Load) shipping. CEO Dave Bozeman stated that the company has already deployed hundreds of AI agents across multiple business segments to handle high-frequency, repetitive tasks.
The customer-facing quotation process is the most typical scenario. Previously, a manual quotation typically took 20 minutes; now, an AI agent can complete it in 31 seconds and can operate 24/7. Bozeman believes that faster response times and more complete information display increase customers' willingness to submit inquiries and also increase the company's chances of winning orders.
Personnel restructuring rather than large-scale replacement
Bozeman stated that the company did not use AI as a tool for direct layoffs, but instead reassigned employees who were originally responsible for quoting to higher value-added jobs, such as assisting clients in dealing with the ever-changing tariff environment.
However, AI has still brought significant savings in manpower. The company states that its annual natural attrition rate is approximately 11% to 14%, and after AI agents took over some processes, the company no longer needs to continuously recruit departing employees. For some business lines, the relationship between processing volume growth and staff size has weakened significantly.
Bozeman also hopes to leverage AI to upgrade its business, moving beyond just freight forwarding to extend into supply chain consulting and even taking on more comprehensive supply chain functions for its clients. At the same time, the company plans to win back small and medium-sized enterprise (SME) clients it has lost in recent years, and says it is still hiring more staff in both supply chain consulting and SME client services, but these positions will be staffed with AI assistants.
Self-developed models reduce usage costs.
Regarding cost control, Bozeman stated that most of the company's AI agents are developed internally, primarily based on proprietary or open-source models, rather than relying on external vendors. C.H. Robinson currently employs approximately 450 engineers, most of whom have backgrounds in the aviation industry.
He claimed that this approach allowed the company to generate hundreds of millions of dollars in business revenue with a token cost of less than $2 million. According to him, if external organizations wanted to replicate this system, they might need to coordinate 15 to 20 different partners simultaneously.
Bozeman also mentioned that when developing AI agents, the company organizes cross-departmental teams, including engineering, operations, finance, and legal, to work together to first streamline the processes and then decide which steps should be eliminated and which steps are suitable for automation.
Organizational culture is also being incorporated into AI transformation.
In addition to technological investment, Bozeman attributes the positive impact of AI adoption to changes in organizational management. The company employs Failure Mode and Effects Analysis (FMEA) when developing AI systems to proactively assess potential problems and develop contingency plans.
He also demanded that the team expose project risks more directly. Internal progress reports only use "green" and "red" statuses, omitting the "yellow" designation. According to him, many "yellow" projects have essentially deviated from their goals, but management is unwilling to explicitly acknowledge this. The company hopes to identify problems earlier and concentrate resources to correct them through this method.
This logistics company's case demonstrates that the effectiveness of AI implementation depends not only on model capabilities but also on process restructuring, job adjustments, and management mechanisms.
