NASHVILLE—Epicor executives opened this year’s Insights conference with a common theme: AI is arriving faster than anyone knows how to fully use it. CEO Steve Murphy told attendees that the requirements for succeeding in this new environment are still taking shape. However, company president Vaibhav Vohra argued that transformative technologies wait for the “right vehicle” before they can create value.
For manufacturers, Epicor believes ERP will be that “vehicle.” The conference stressed that ERP no longer serves as a system of record, but rather a system of action. At the same time, AI does not simply improve ERP; it exposes it.
Manufacturers want to use AI for planning, scheduling and operational decision-making, but many struggle with the fundamental problem of collecting reliable data. Mike Zahn, a software engineer at plastics manufacturer Federal Foam Technologies, explained that older machines can perform effectively but lack PLCs or other mechanisms for feeding operational data into ERP systems.
“Without good information, you can't leverage the great tools that are out there to make good decisions off of that data,” Zahn said. “It’s possible to use Internet of Things devices and attach them to pieces of equipment, but there are limits to that. We work with flexible products, so the cadence of manufacturing is not dictated necessarily by machine cycles.”
Epicor President Vaibhav Vohra speaks at Insights 2026.
Easton Parks, an IT and controls engineer at machinery maker Van-Am Tool & Engineering, described AI as most useful when it operates as an internal knowledge layer instead of an open-ended search tool. Parks said his company uses an isolated agent connected to internal systems and documentation, enabling workers to access company information rather than relying on web searches.
The most impactful early AI use cases are not shop floor robotics, but administrative, communication and decision-support layers inside ERP systems.
Van-Am Tool & Engineering’s goal for 2026 is to identify areas that do not require human interaction, like RFQs and purchase order reminders. Parks added that the company’s sales order entry process previously involved a person notifying engineering of a new part. Now, when a sales order employee enters an order, if the system recognizes a new part, it will copy the quote’s details, enter the information in the engineering workbench and notify the worker associated with the customer.
In manufacturing, the problem often lies in the lag between when a question is asked and when the answer returns to the system. Epicor Director of Product Management of AI Michael Atkisson labeled that gap as a clear area for AI to change how an ERP system is used.
“Some analyst has to figure out an answer to an inventory question,” Atkisson said. “They come back a week and a half later with an answer, but we’ve already made the decision. If I could make an inventory decision when the issue came up, then those expedited decisions are like real money for the business. Then I don't have to buy as much or have to buy more at a worse price.”
Epicor announced a new wave of Prism agents designed to reduce coordination work by executing tasks such as shipment tracking, freight and document processing and managing business rules within ERP workflows. Additionally, the new Prism Agent Foundry allows customers to build and customize their own agents.
This approach reflects a common tension in AI adoption. Customers want the freedom to create original agents, but ERP governance remains critical to preserving auditability, control and system integrity.
“What we're discovering is that there's kind of a gap in ERP,” Atkisson said. “You can’t just record the change in the database. You have to understand the context that you used to make that decision, and you have to build this matrix that could understand and audit, end-to-end, every decision that was made.”
Epicor Insights 2026
The need for structure also appears in how users engage with AI systems. Epicor’s internal testing found that users do not always interact with agents as designers anticipate, and they will accept default outputs or fail to challenge assumptions.
This challenge shows a broader challenge facing AI adoption. As the technology takes on more duties, manufacturers still expect humans to make the final call. Should an AI agent ever make a suggestion that contradicts a seasoned employee’s judgment, Zahn stressed the need for diverse opinions and a consensus that will build trust in the process.
Once trust is established, Epicor Principal Product Marketing Manager Andrew Robling argued that ERP can become the vehicle for AI.
“What AI needs is data,” Robling said. “It needs a lot of data to be able to come up with trends and analysis and things like that. So to me, ERP probably is the one system that every business has that has a lot of data in it that can be looked at. You've got things like procedures and workflows—all those kinds of things that customers are using today that AI can help with.”
