Epicor’s Next Step in Cognitive ERP: Prism GenAI Agent Network

Can you reimagine how people interact with enterprise software? Epicor is doing just that. Their announcement of the Prism agentic AI network and orchestrator takes the next step in moving from a system of record to a system of action.

Prism includes conversational AI for employees to interact with ERP, and Epicor plans to integrate automation and causal AI in the future.

Core Agent Capabilities

Kinetic, Epicor’s manufacturing-focused ERP, already has five types of Prism vertical AI agents. The graphic below illustrates and describes the specialized functions of these semiautonomous, pre-trained bots.

Prism's agents use LLM and RAG technologies. While LLMs and related generative AI excel at understanding unstructured data like documents, they struggle with structured data typically found in ERP systems. Prism's first innovation is its ability to navigate ERP structured data effectively.

These agents can answer questions and create filtered lists in multiple ways.

Most ERP users need only particular portions of the whole data set to perform their tasks, and Prism can streamline that.

Epicor has encoded the data structures of Kinetic ERP with typical fields and processes into Prism to pass context onto the LLM. In addition to the standard Epicor agents, they can include customer-specific questions and point to their customizations. So, it is both Kinetic-specific and customer-specific in function.

?

When crafting new approaches to support employees or automate processes, Prism will determine which agents are required and how many. Depending on the problem to solve or the process to improve, one or multiple agents may come into play.

How It Works

Epicor says it has an unfair advantage since it has broad enterprise functionality that many customers have had in place for years. So, historical data about the business is already stored in a structured data ontology that fits their industry. This forms a quick-start basis for RAG to create an Epicor- and customer-specific context for the large language model (LLM). Newer customers are encouraged to import a rich data set during their implementation.

Epicor talks about three elements of this agentic AI.

1.??????? Handshakes, which redefine how humans interact with enterprise-grade data

2.??????? Gears, which are the data models, integrations, and APIs that enable data to flow through the system

3.??????? Sparks, which are the insights the AI can distill to support decision-making

Security and Data Governance

One of the top issues or fears companies have with using AI is security, trust, and ensuring results are valid. For security, Prism agents act on behalf of each system user. That means the agents only have access to the portions of the system that the user has, based on existing security settings, through Kinetic.

On the LLM security front, Epicor follows standard guidance on OWASP’s 10 pillars for safety and security. They have separate data for each customer and do not pool data across customers. The LLM for Prism is always guided by the context of a specific customer. The business activity queries focus on the company’s data set, logic, and rules.

Coming up Next

As in the graphic above, Epicor is creating an agent for their Enterprise Content Management or ECM. While digitalization has come a long way, companies still tend to do business with documents such as contracts, requirements, etc. Today's many supply chain challenges suggest that this could be a beneficial agent. Questions around supply contracts will likely proliferate as companies restructure their supply chains in the face of tariffs, regulations, and natural disasters.

Another in-development agent will support automating mass modifications. For example, if you have 100 lines in an order, and they all need to be modified, Prism will be able to do that for the user at their request.

Continuing the Conversation

Epicor has continued to innovate and include leading-edge technologies in their ERP platforms for years. This is an excellent milestone for Kinetic in manufacturing. We look forward to hearing more as Epicor keeps moving forward. Thank you, @Steven Knuff, for keeping us updated, and to Arturo Buzzalino , Michael Atkisson, Ph.D. , and Michael Muckala for creating a concise and informative briefing!