Smarter SAP Procurement with AI Agents

Free-text purchase requests in SAP often require an additional manual review. Does the selected material group match what someone wants to order? Should the product have been ordered through a catalogue instead? And can the request be approved, or does it need to be adjusted? These checks require context and interpretation, which is why they often end up with an operational buyer.

For this showcase, we developed an AI Agent for SAP Procurement. The agent reviews Purchase Requisitions from SAP S/4HANA, assesses the request based on the available information and independently provides an approval or rejection. The reasoning behind the decision is automatically written back to SAP.

From manual checks to intelligent decision-making

Traditional automation works well when the rules a system needs to follow can be defined upfront. Free-text requests are different. An operational buyer needs to understand what someone is trying to purchase and determine whether the information entered is logical and correct.

The AI Agent takes over this type of review. As soon as a Purchase Requisition is ready for assessment, the relevant information is retrieved from SAP and passed to the agent.

In our showcase, the agent performs two checks:

• Material group check: Does the selected material group logically match the requested product?

• Catalogue item check: Is a suitable catalogue item available, and should the request have been made through the catalogue?

Based on these checks, the agent determines whether the request can be approved or should be rejected. Both the decision and its reasoning are then automatically written back to SAP S/4HANA.

An AI Agent that reasons within the SAP process

Imagine an employee wants to order promotional materials. The agent reviews the description, assesses whether the selected material group is appropriate and checks whether a suitable catalogue item is available. If everything is correct, the request can be approved.

But what happens if someone tries to request a Tesla Model 3 under the material group ‘Office Supplies’?

The agent recognises that this combination does not make sense. In the showcase, it concludes that a material group for vehicles would be more appropriate, rejects the request in its current form and records the reasoning directly in SAP.

This means AI is doing more than automating a technical process step. The agent interprets information, performs multiple checks, reaches a decision and makes the reasoning behind that decision transparent to the user.

Less repetitive checking, more focus on exceptions

This approach demonstrates how AI Agents can enable further automation of operational procurement processes. Standard checks no longer need to be performed manually for every individual request, while requests that do not meet the defined criteria can be identified immediately and accompanied by a clear explanation.

For procurement teams, this creates the potential to reduce repetitive review work, process requests faster and perform checks more consistently. Employees can focus their attention on exceptions and procurement challenges where human expertise adds the most value.

From Procurement Agent to broader SAP automation

The current showcase focuses on two specific checks, but the same approach can be expanded further. An AI Agent could, for example, incorporate additional SAP data, supplier information, contract data, purchasing policies or historical procurement data into its assessment.

The principle also extends beyond procurement. Many SAP processes still contain manual steps in which an employee gathers information, interprets it and makes a decision. These are precisely the areas where AI Agents can enable a new level of automation.

The key takeaway? AI Agents make it possible to automate not only actions within SAP, but also parts of the decision-making process. This showcase demonstrates what that can look like in Procurement – and provides a starting point for exploring which other manual decision and control steps within SAP could be made smarter.