As artificial intelligence continues to advance, organizations may feel pressure to implement enterprise AI as quickly as possible. However, leadership should first consider the direction they want enterprise AI to take. For instance, the organization should determine how the technology will find reliable information, connect with existing systems, follow approved processes, and protect sensitive data.
Within medium to large organizations, these decisions become especially important. For instance, a manufacturer may want AI to help employees locate maintenance procedures, anticipate equipment failures, or review inventory shortages. Meanwhile, a life sciences organization may want to search controlled documents, summarize quality information, or identify patterns across a large product portfolio.
In all of these cases, the organization should build a workflow through methodically constructed AI connections.
Build the AI Workflow One Layer at a Time
Consider a medium-sized manufacturer with thousands of maintenance documents spread across different repositories. When a machine develops a problem, an employee may know how to describe the symptoms, but may not know the exact terminology used in the maintenance procedure.
Semantic search could help bridge that gap. Rather than searching only for exact keywords, it considers the meaning and intent behind the employee’s request. Once the system locates the right procedure, retrieval-augmented generation, commonly known as RAG, could provide that information to a large language model. The model could then prepare a short explanation and direct the employee to the original source.
At this point, the AI can find and interpret information, but it cannot necessarily interact with the organization’s other systems. Model Context Protocol, or MCP, can provide a standardized connection between enterprise AI applications and external data sources, tools, and workflows. If properly authorized, an MCP-connected tool might retrieve the related maintenance history or parts information from an operational system.
An AI skill could then guide the technology through the organization’s preferred process. For example, the skill might instruct the system to review the approved procedure, check the maintenance history, identify missing information, cite its sources, and request employee approval before creating a service ticket.
Instead of competing against one another, these technologies can build upon each other. However, adding each layer also adds new risks. For instance, an RAG system may retrieve an outdated procedure or prepare an incorrect summary. Additionally, an AI connection may expose information to an employee who should not be able to access it. If the system can take action, a small error may also spread into later portions of the workflow. As a result, organizations must incorporate access controls, monitoring, source validation, activity logs, and human approval throughout the architecture.
Bring Artificial Intelligence into the ERP Environment
Next, consider a large life sciences organization that needs to identify patterns across a significant product portfolio. To ensure that all items are considered, an employee may need to spend a significant amount of time retrieving, processing, and inputting data.
Fortunately, for many organizations, an ERP system already contains the foundation for enterprise AI. SAP and similar platforms may hold years of information related to finance, procurement, manufacturing, inventory, supply chain, human resources, and customer activity.
The next question is how the organization can make better use of that information without disrupting the reliability of its central system. According to IBM’s overview of AI in ERP, organizations can integrate technologies such as natural language processing, machine learning, and predictive analytics with ERP systems to automate tasks, analyze data, improve forecasting, and support decision-making.
Once again, the appropriate technology depends on the process. For instance, if the employee is struggling to navigate a complicated ERP interface, they may benefit from natural language processing. This way, the employee can ask about what products should be included in the data analysis, in ordinary language. The system could then translate that request into a structured search across authorized data sources.
Additionally, instead of repeatedly transferring data manually, the employee may gain more value from robotic process automation, especially if the steps taken are stable and rules-based. Finally, as the employee works to find patterns and predictions through the data, they may benefit from machine learning.
Generative AI could provide an additional layer by explaining those findings. Rather than presenting an employee with a series of disconnected data points, the system might prepare an initial summary of the data and patterns that were found. However, the team should still validate the underlying information before making a consequential decision.
Within the SAP environment, SAP Signavio can help an organization understand where a process is slowing down or producing unnecessary work. Joule can provide conversational access to supported information and tasks, while SAP Business AI can incorporate AI into a wider range of business functions. Finally, SAP Business Technology Platform can support the data, integrations, automations, and extensions surrounding those capabilities.
Next Steps
Ultimately, successful enterprise AI adoption is not determined by the number of technologies an organization implements. Rather, business leaders should consider whether those technologies work together to improve a reliable business process.
In the next two articles, we will explore these connections in greater detail. First, we will examine how RAG, semantic search, MCP, and skills build upon one another. Then, we will consider how natural language processing, automation, machine learning, and generative AI can enhance SAP and other ERP environments.
Stay tuned! In the meantime, feel free to take a look at our previous series to gain a deeper understanding of the AI solutions available:
How to Build an AI Foundation That Can Scale
Which AI Platform Is Your Best Conductor?
Leave A Comment