Real-World AI Use Cases for OpenEdge Applications

Business questions often require information from several systems. A quality manager investigating a production issue may need maintenance records and supplier documentation alongside ERP data, while an accounts payable clerk may need both invoice details and the policies governing a correction.

That was the focus of our recent webinar, Real-World AI Use Cases for OpenEdge Applications, featuring Irfan Syed, Jay Sanderson and Cameron Wright. Together, they explored how Progress Agentic RAG, the OpenEdge MCP Server and Retrieval Agents Orchestrator can bring structured application data and unstructured information together to help users investigate problems and make informed decisions.

The session centered on three scenarios: manufacturing quality investigations and 8D reporting, invoice posting failures and capable-to-promise customer commitments. A demonstration showed how these workflows could operate within an ERP application, followed by Q&A on integration, governance and implementation.

Starting with the Business Problem

Irfan emphasized that successful AI initiatives begin with a specific business problem. Different applications of AI depend on different data and techniques, so organizations need to understand the task before deciding how to support it.

The three scenarios shared a common challenge. The information needed to answer a question already existed, but employees had to search disconnected systems, gather records and manually connect the findings. Because the employee has to search multiple systems it left less time for analysis and potentially could lead to decisions based on incomplete information.

The proposed approach brings that information together through natural-language questions. Users can retrieve relevant application records alongside supporting documents, then review the findings in the context of their existing workflow.

Investigating Manufacturing Quality Issues

The first scenario followed Linda, a quality manager at an automotive supplier. When first pass yield falls below a customer-defined threshold, she must submit a formal 8D report within four hours. First pass yield measures the percentage of units that pass through a process correctly on the first attempt, without rework, repair or rejection.

An 8D report follows the eight disciplines of a structured problem-solving methodology, including describing the problem, containing it, identifying the root cause and implementing corrective actions. Preparing that report requires evidence from quality records, production logs, maintenance information and material traceability data.

In the scenario, manually consolidating those records consumed much of the reporting window. Missing connections between materials, equipment and production events could lead to an incorrect root cause and recurring quality problems.

Jay demonstrated an alternative using a fictitious automotive manufacturer called Tallowa Plant. After a drop in first pass yield on line A2, a user asked what had happened. The retrieval agent gathered context from several systems and returned an explanation with supporting information, including relevant equipment media and a view of relationships among the contributing components.

A follow-up request generated a draft 8D report for approval. The demonstration illustrated how gathering evidence and preparing documentation could be compressed into a shorter workflow, while leaving review and approval with the user. The timings shown reflected the demonstration rather than a measured customer deployment.

Explaining Invoice Posting Failures

The second scenario focused on Maria, an accounts payable clerk handling invoices from suppliers across several countries. Just before period close, she faced 23 posting failures identified only by cryptic error codes. Resolving each one required moving between accounts payable, general ledger and vendor information.

The challenge extended beyond finding an incorrect field. Maria also needed to determine the correct value, understand the applicable policy and know whether the correction required approval. Under deadline pressure, clearing an exception with the wrong cost center could create an audit issue later.

In the demonstration, the user asked why the invoices could not be posted. The retrieval agent consulted the invoicing system and exception information, then explained that a cost center was closed. It also identified a proposed fix and the authority limit for the clerk who was logged in.

This connected the operational error with the policy governing its resolution. The demonstration showed information and guidance presented within the application, giving the user a basis for reviewing the correction rather than treating an error code as the complete answer.

Answering Customer Delivery Questions

The third scenario involved Hans, a customer representative asked whether a made-to-order part could arrive by a requested date. Answering required expanding the bill of materials, checking component availability, reviewing production capacity and examining purchase orders for materials in short supply.

This capable-to-promise assessment could take 30 to 90 minutes in the scenario. A date range from a planning system was insufficient because Hans still needed to understand the constraints and turn them into a customer commitment.

Jay demonstrated a request for 250 busbar sets for Moran Automotive by a specified date. The retrieval agent broke the question into smaller queries, gathered the necessary information and reported that a material shortage would cause delivery to miss the requested date by four days.

The user then asked what would happen if the missing component were expedited. Under that revised assumption, the scenario indicated delivery one day before the requested date. This showed how users could examine alternatives and understand the reasons behind a delivery estimate before making a commitment.

Connecting Application Data and Business Knowledge

The architecture combined three components. Progress Agentic RAG retrieved unstructured knowledge such as policies, manuals, contracts, images and videos. Its knowledge boxes held ingested content and the demonstration showed answers linked to specific passages in the source material.

The OpenEdge MCP Server provided access to structured data and application services through Model Context Protocol tools. It consumed OpenAPI specifications to expose REST APIs as tools that AI clients could use. Requests passed through services in Progress Application Server for OpenEdge, allowing those services to retrieve database information or execute application logic.

Retrieval Agents Orchestrator coordinates requests across those sources. It decomposes a question into smaller queries, routes them to the appropriate sources and combines the retrieved context into a unified response. The presenters noted that this approach could extend to other connected systems beyond the two sources used in the demonstration.

Jay showed how descriptions of MCP endpoints, prompts and model choices could be configured to guide retrieval. Responses could also follow a JSON schema, allowing an application to display the results in a format suited to its users. In the Q&A, Irfan explained that the demonstration application supplied the interface and APIs, while the platform handled orchestration and retrieval. You can learn more about how Progress Agentic RAG and OpenEdge work together here.

Integrating AI with Governance and Human Review

The Q&A addressed how users could evaluate the information returned. Irfan clarified that the demonstration used a simulated ERP environment with data in a real database and working PAS for OpenEdge and MCP services. The fictitious business scenario therefore ran through functioning infrastructure.

The presenters also discussed REMi quality metrics, which evaluate responses for groundedness, relevance and accuracy. Agentic RAG usage information can help teams examine answer quality and costs associated with a knowledge box. Citations give users another way to inspect the evidence behind an answer.

When asked what happens if application data and a policy document disagree, Irfan and Cameron emphasized human review. The workflow described in the webinar surfaces conflicting information for the user to resolve. It does not independently settle that conflict or take the resulting business action.

For organizations working with an existing ABL code base, Cameron recommended OpenEdge 12.8 or later as the starting point for connecting existing ABL application services to AI tools through the OpenEdge MCP Server. A REST interface is needed and he suggested a service interface that translates application data into meaningful information for AI tools. He also clarified that the MCP Server did not directly support the APSV protocol at the time of the webinar.

Routing depends on the context and tools available from configured sources. For example, an agent answering a delivery question may first retrieve a missing shipping address before passing the information to another step. The presenters explained that source descriptions and tool information help the Retrieval Agents Orchestrator determine where to retrieve the required context.

Organizations can also connect documentation from sources such as Google Drive, SharePoint and ShareFile. Irfan described sync agents that refresh content from connected locations, along with security guardrails and labels used to govern the information returned. Authentication and source availability remain necessary for successful retrieval.

The discussion also covered options for evaluation. The presenters pointed to an Agentic RAG trial and OpenEdge MCP Server download and setup documentation. Retrieval Agents Orchestrator is not yet publicly available. Organizations interested in the orchestration capabilities demonstrated in the webinar can contact Progress to discuss availability. The Agentic RAG trial tutorial walks through setting up a knowledge box, ingesting content and testing responses.

Final Takeaway

The webinar illustrated how AI can help employees answer questions that span application records and business documents. The practical value comes from connecting the evidence needed for a task, presenting it within the application and keeping users involved in decisions that require judgment or approval.

For OpenEdge organizations, these scenarios offer a way to identify workflows where searching across systems consumes time or leaves important context missing. Watch the full on-demand webinar to see the demonstration and hear the complete discussion and Q&A from our experts.