Edsal Manufacturing Company has designed, manufactured, and distributed storage systems and industrial furniture from Chicago for more than 65 years, with customers across consumer and industrial markets. Their data landscape operates on a SQL Server environment managed by a small IT team which means any new capability has to be practical to build and easy to govern.
Arthur Courtney, Director of Information Systems at Edsal, set out to deploy AI-powered voice and chat agents through Retell.AI to handle customer support calls autonomously.
Building AI agents that handle customer calls on live data
AI agents deployed for live customer support operate under a constraint that other AI interfaces don't face: silence is failure. When a customer calls in, the agent has to answer immediately, not after a data sync catches up, and not from a snapshot that was current yesterday. Every question has to be answered from live data, on demand.
There's a second challenge that's less obvious: customer questions are inherently non-deterministic. No two calls follow the same path. They say "where's my stuff?" or "what was in that last shipment?" and the agent has to handle whatever comes in. That requires a data foundation structured in a way the agent can actually work with: the right data, exposed in the right shapes, so the agent can retrieve what it needs regardless of how the question is phrased.
Building that foundation, scoped, structured, and accessible via a standard endpoint was the problem CData Connect AI solved.
"The bar for AI customer support is higher than it looks. Customers have already been burned by bots that loop them in circles or can't answer a basic question. An agent that hesitates, guesses, or pulls stale data doesn't just fail. It damages trust. Getting it right meant the agent had to answer accurately, immediately, every time."
Arthur Courtney
Director of Information Systems, Edsal Manufacturing
Governed data access for the agent
Connecting to CData Connect AI was straightforward. Arthur then created a dedicated user account and workspace in Connect AI, added curated views and tables as read-only assets, and connected Retell.AI through the Toolkit MCP endpoint. With this setup, the agent only sees what IT assigned to that workspace. And critically, every request the agent fields, runs live against Edsal's production SQL Server with the actual current data at the moment of the call.
The MCP tool structure is what makes the non-deterministic nature of customer calls manageable. Rather than hard-coding query paths for every possible question, Connect AI exposes the data as a set of callable tools the agent can reason about. When a customer asks about an order, a shipment, or a replacement part, the agent determines which tool to call based on the intent of the question, and retrieves a live answer, dynamically, without requiring a separate integration for each scenario.
"With Connect AI, I can give the agent exactly what it needs to do its job. And nothing it doesn't. That's not something you can easily do when you're connecting an AI directly to a database."
Arthur Courtney
Director of Information Systems, Edsal Manufacturing
AI agents taking customer calls, grounded in live data
Within two weeks of starting the trial, Arthur was already demoing the integration to Edsal’s management. And four months later, Ava, Edsal’s AI customer support agent was live, handling customer support calls with real time access to Edsal's SQL Server data and the proper guardrails in place.
Ava looks up the customer’s data on the spot and responds accurately without being transferred, put on hold, or routed to a human for a data lookup. And because the agent's data access is governed entirely within Connect AI, IT controls what Ava knows without having to touch application code. In her first week live, Ava handled over 800 calls.
"Not only are we hoping that it's going to satisfy most of the requests that come in without involving a person—it's available 24/7. That means customers can get answers outside of staffed hours, not just when our team is at their desks."
Arthur Courtney
Director of Information Systems, Edsal Manufacturing
Ava is only the beginning. Arthur is already building out a second Connect AI implementation for internal staff. Edsal has licensed Claude, and the plan is to give employees conversational access to their ERP data: sales orders, inventory status, invoicing, vendor purchase orders, and preventative maintenance records. A group of super users is actively working with the data now, with a broader rollout planned once the asset collection is finalized.
"This will be a huge win for us," Arthur said. "The users will be able to ask complex questions for analysis, generate their own reports. All from a collection of assets that we control through Connect AI. We are really happy with this implementation.”
