CSX boosts supply chain agility using Microsoft Copilot Studio and Azure AI Foundry

CSX Corporation is one of the largest freight railroad transportation companies in the United States. Over 3.4 million carloads are transported annually across a rail network of more than 35,000 track miles that extends across the eastern United States and into Canada.

The company’s rail and intermodal businesses provide rail-based transportation services including traditional rail service and the transport of intermodal containers and trailers. CSX partners with a wide variety of industries, including agriculture, automotive, chemicals, coal, food products, metals, minerals, and paper, and collaborates with over 240 short-line and regional railroads.

Every day, more than 7,000 customers visit ShipCSX—the company’s online portal. With ShipCSX, customers can track freight, manage shipments, and access rail logistics services. To provide even better customer service, CSX recently launched “Chessie for ShipCSX”—an AI-driven assistant embedded in the site.

Built using Microsoft Copilot Studio and Azure AI Foundry, this new assistant is designed to make it easier to access information and services using natural language. Customers can ask to track a specific shipment, for example, and Chessie will instantly bring up the details. Better yet, it can provide fast answers to frequently asked questions—something that previously might require a longer search of resources on the site.

“With Chessie, we’ve reimagined how we support our customers—delivering fast, intelligent assistance through natural language interactions. By leveraging Microsoft Copilot Studio and Azure AI Foundry, we accelerated delivery and brought real value to our customers in record time,” says Dave Rich, Vice President of Technology Strategy, Architecture, and Governance at CSX.

In just the first 45 days of operation, over 1,000 customers consulted Chessie across more than 4,000 conversations. “Customers are coming back again and again to use our agent," says Ryan Rogan, Senior E-Business Project Analyst at CSX. “It’s a good indicator that they like the experience and that they are getting exactly the information they need.”

Flexible development tools

With Copilot Studio and direct integration with Azure AI Foundry, the makers at CSX had just the tools they needed to build an advanced, multi-agent experience. For starters, Copilot Studio saved time on basic design. “Microsoft Copilot Studio makes it extremely easy to build an effective agent interface, which gave us more time to focus on advanced functionality,” says Praveen Rajaram, Software Architect, Enterprise Architecture at CSX.

Copilot Studio also streamlined another key development process: integration of knowledge sources. CSX initially planned on consuming FAQ and training documents using traditional Retrieval-Augmented Generation (RAG) methods with Azure services like Azure Service Bus (ASB), Blob Storage, and AI Search. This involved creating a program for chunking and embedding data and building custom UI screens where business users could upload documents. The implementation would take several weeks or more to complete.

Using built-in RAG capabilities such as file upload in Copilot Studio, the team was able to integrate all necessary documents in just a few hours. “Being able to connect knowledge sources so easily to an agent highlighted the true power of Copilot Studio and enabled us to deliver value quickly without a lengthy development cycle,” said Rajaram.

Enabling employees with no development experience to upload documents for use as a knowledge source is also helping to rapidly evolve and improve the agent. Working off telemetry gathered through Copilot Studio and into Power BI dashboards, the team can quickly identify trending topics raised in Chessie by customers—and then just as quickly upload new knowledge articles to address these topics.

Along with serving as a powerful FAQ tool, Chessie also delivers automated assistance for more complex tasks—from tracking shipments and checking schedules, to helping open and manage cases in Salesforce. This is where prompt flow in Azure AI Foundry came in. Prompt flow was used by CSX to create additional agents that drive real-time data retrieval through custom logic across multiple systems.

One of the agents—the supervisor agent—performs many other tasks such as a multi-stage user validation process. Because the same rail car is assigned to different customers over time, the agent must validate that a customer submitting a tracking request is assigned to that particular rail car at the time of a status request. The supervisor agent also manages the routing of customer questions to the other agents and can even reformat a question so that it’s more meaningful to the agent.

AI Foundry—and other Azure tools such as Azure AI Search—will continue to play a key part in development of Chessie, However, CSX is looking forward to doing more of this work using the latest low-code features in Copilot Studio. As Rajaram explains, “We are planning to add more agents over time to the agents driving our AI assistant—and when we do, we want to explore how we can build and connect these agents using the new multi-agent orchestration in Microsoft Copilot Studio.”

Jeff Moore, Director of e-Solutions at CSX is equally excited about the value of Copilot Studio. "We have been very excited that Microsoft Copilot Studio provides a solution where my team can make timely changes to address customer needs with limited technical support. This really allows the business to try new things quickly based on customer interactions. We think this release gives us a strong foundation on which to expand our offerings for our customers."

A broader Power Platform environment

Chessie for ShipCSX is just one of many solutions built by CSX using Copilot Studio. It all started with a basic, rules-based agent for the company’s property division. The agent manages frequently asked questions from utility companies about access and building procedures on CSX rail properties.

This agent caught the attention of an employee who asked if something similar could be created for frequently asked questions related to medical benefits. By then, makers at CSX were actively working with the AI features in Copilot Studio and jumped at the opportunity to build “Chessie for Medical," the company’s first, AI-driven agent—which then led to the more advanced “Chessie for ShipCSX."

All of this work—including development with Power Apps and Power Automate—is governed by the company’s Power Platform Center of Excellence (CoE). The CoE was built using the Microsoft Power Platform Center of Excellence Starter Kit, including DLP policies and tools for monitoring applications and agents across the organization.

To help manage a growing community of makers, the CoE team also leverages Managed platform—a suite of enhanced security, governance, and operations features for Power Platform such as environment routing and pipelines for Power Platform. “The managed platform capabilities in Power Platform make it easier to maintain control of our development environments while also giving our makers the freedom to innovate,” says Scott Welfel, Senior Product Manager, Power Platform at CSX.

Building on Chessie's success, other CSX agents are in development, starting with an internal-facing agent that supports thousands of rail workers by answering questions about operating rules and terminal timetables. As Rich says, “Power Platform and Azure are empowering teams across CSX to innovate faster and smarter. Chessie is just one example of how low-code and AI can come together to solve real business challenges and elevate customer engagement.”