Delivering the Future - Built with OpenAI on Azure

Our customer processes and fabricates steel for manufacturers of cars and trucks, heavy equipment, and consumer appliances among others. In an industry hard-hit by the closing of steel plants during the pandemic, end-customers shop for aggressive pricing, confidence in product availability and assurance that vendors have a resilient supply chain.

"First to market - first to win" is critical as major manufacturers struggle with short supply, rising prices and consumer demand for cars and major appliances. Ryerson's 600 sales professionals deal with more than 100,000 SKUs and receive an email request for a quote every 4.5 minutes. Buyers are often impatient, moving on to another supplier if a quote isn't immediately available.

Historically, Ryerson sales representatives prepared quotes by hand, drawing from several systems to determine price and availability. Quote requests came largely by email with or without attachments, and often by phone. Time-to-Quote was typically hours and often required multiple clarifying communications between the customer and Ryerson's sales team.

Unprecedented competitive pressures ignited a need for the transformation of high-volume sales quoting and ordering technology. Automating and accelerating time-to-quote was key to growing market share by immediately capitalizing on every sales opportunity.

Using the Microsoft Cloud Adoption Framework as a best practice, Metal Detector is built on Azure OpenAI services, Azure Kubernetes Services for running custom models, Azure App Services, Azure Bot Services, Azure Storage, and integration with Dynamics 365. Concurrency incorporated Microsoft Responsible AI Principles, Microsoft Cloud Adoption Framework and Microsoft Well Architected Framework throughout the design and build process. Metal Detector also integrates with SAP and Oracle ERP systems, keeping inventory status, business and financial data up to date. Company leaders can use Power BI Reporting to assess sales in real-time.

Developing familiarity with the steel industry language and business context was a starting point. Metal Detector scans email in real-time and uses entity recognition to look for relevant information, such as SKUs and product information. As an example, the model is trained to recognize "steel," "stainless steel," and "SS" as the same description. The model can also interpret missing information, such as understanding that "W1" refers to the width of a steel beam; and that if no product is specified, it means carbon.

• Concurrency exceeded expectations by decreasing Time-to-Quote from hours to minutes. Customers can submit their requests, evaluate the reliability of Metal Detector's response, and place their order in a few minutes - entirely without the human engagement of a salesperson.
• Metal Detector now handles over 50% of quote requests without human intervention. Self-learning features continuously increase that number.
• More than half of Ryerson's 600 sales representatives are trained in the system and the number continues to grow daily.
• With a goal of driving impact and ROI for our customers, the Concurrency team addressed Ryerson's business urgency accelerating time to market with a rapid build to MVP in six months.
• By challenging the status quo, Ryerson credits Metal Detector with as much as a 3% increase in top-line revenue.
• Ryerson's strategic investment in Concurrency's AI solution and Microsoft's platform has demonstrated the power of technology to become a revenue center, compared to its traditional role as a cost center.
• Ryerson sales teams are now available to engage more deeply with end customers, building brand loyalty and developing familiarity with their business. This enables Ryerson to deliver on the brand promise to "Provide More than Metal, Streamline Your Supply Chain, and Becoming an Extension of Your Business."