Danish wholesaler Lemvigh‑Müller has deployed artificial intelligence to automate the process of supplier order confirmations, one of the most time-consuming tasks in procurement. The solution consists of multiple AI agents, each responsible for well-defined tasks, integrated into a single automated workflow built on SAP Business AI. The result is faster processing, better data quality, and more accurate shipping information for your customers.
When suppliers send order confirmations as PDF files, even small discrepancies in price, quantity, or delivery time can result in significant manual effort in procurement. For Lemvigh-Müller, one of Denmark’s largest wholesalers of steel, plumbing, heating and electrical products, this is a challenge that has long been familiar to it and one that has consumed significant time and resources.
The company is now tackling the exact point where early automation efforts tended to stall. A new solution based on multiple specialized AI agents developed on SAP technology and implemented in close collaboration with NTT Data Business Solutions can now automatically read, interpret, compare, and process supplier PDF order confirmations directly into SAP systems.
“In the past, we tried both RPA and traditional automation approaches, but without the desired effect. The main difference this time is that we split the task into multiple independent AI agents, each responsible for a specific part of the process. These agents can now work together to handle things that previously required manual review,” says Frederik Aakerlund, IT Director at Lemvigh‑Müller. says Mr.
10 weeks from idea to live AI agent
The project began with an email from Jess Frederiksen, an AI-savvy project manager in Lemvigh-Müller’s Markets and Procurement division. After successfully experimenting with matching order confirmations and purchase orders using ChatGPT, he approached his IT director to see if this could become a fully integrated system solution.
From initial testing to production deployment, the entire project took just 10 weeks. According to Lemvigh‑Müller, this short implementation timeline was important in allowing the solution to quickly demonstrate tangible business value and build internal support.
“This was not a long-term project. We went from idea to operational AI agent in 10 weeks and already delivered measurable value to procurement professionals,” says Aakerlund.
Over time, Lemvi Müller expects the solution to free up resources equivalent to three to four full-time employees. These resources are reallocated to higher value activities, such as processing the most complex and exception-driven orders.
“The goal is not to reduce headcount, but to leverage our expertise more effectively. AI agents handle the day-to-day tasks, allowing procurement professionals to focus on the cases where their experience really matters,” Aakerlund added.
Over 100,000 order confirmations automated
Every year, Lemvigh‑Müller sends approximately 175,000 purchase orders to more than 2,000 suppliers. While some of this volume is processed in a structured manner via EDI, approximately 60% of supplier order confirmations are still received as unstructured documents.
By deploying a tuned AI agent, the company can now automatically identify delays, quantity changes, and price discrepancies and respond much faster.
“Previously, if order confirmations were processed manually, it could take hours or even days for changes to propagate across the organization. Now, our AI agents update data almost instantly, so customers get a more accurate delivery status faster,” said Klaus Heinemann, head of SAP ERP at Lemvigh-Müller, who led the development with the project team. “Additionally, we can now identify price discrepancies before the final invoice is issued, saving time for both us and our suppliers.”
Multiple AI agents coordinated in a single workflow
The solution is built around three collaborating AI agents with clearly defined roles within the process. One agent processes incoming emails and attachments, a second extracts and structures data from PDF documents, and a third compares the extracted information to SAP purchase orders to determine if there are matches or discrepancies.
As a result, complex, unstructured supplier data can be processed in integrated, automated workflows without requiring procurement professionals to open and manually review long PDF files.
“What makes this solution robust is the interaction between the agents. Each agent is highly specialized, but tailored to ensure that the process flows seamlessly from start to finish,” explains Heineman.
Three AI agents working together in Lemvigh-Müller
Lemvigh‑Müller’s solution is built around three specialized AI agents, each responsible for a well-defined task within the procurement process. Together they form a single end-to-end automated workflow.
1. Email agent Receive and categorize emails from suppliers. The agent identifies relevant order confirmations and attachments and sends them to the next step in the process.
2. Data extraction agent Extract important information such as price, quantity, and delivery date from PDF documents and structure the data for direct comparison with SAP purchase orders.
3. Matching agent Compare the extracted data with existing purchase orders in SAP to determine if there are matches or discrepancies. If there is a match, the process automatically continues, but deviations are flagged for further processing.
During the project, the importance of master data quality also became increasingly apparent.
“We identified improvements that needed to be addressed in areas such as Incoterms and other master data, which were important learnings not only for this effort but for our broader AI efforts,” he says.
Although it is still too early to measure the full impact on customer experience, error rates, and claims, the hope is that processing supplier verification faster and more accurately will reduce surprises and significantly improve delivery transparency over time. Internally, this solution has generated strong interest and curiosity among employees.
“Procurement professionals have clearly recognized the value of being freed from the most tedious of day-to-day tasks. This has led to a constructive dialogue about how technology can best support their day-to-day responsibilities,” says Heineman.
Business AI that delivers tangible business outcomes
According to Lemvi Müller, this investment is expected to yield a profit in a relatively short period of time.
“When it comes to ROI, we’re talking about quarters, not years, so it was important for us to focus on processes that quickly put the solution into production and had a clear and measurable impact,” says Aakerlund.
For SAP, this project serves as a concrete example of how artificial intelligence can be built directly into core business processes, rather than remaining a disconnected experiment.
“Many companies talk about AI agents primarily in terms of automation. Lemvi Müller shows that the real challenge and the real opportunity lies in coordination,” said David Pontoppidan, Head of Nordic and Baltic AI at SAP. “What makes this solution robust is the orchestration of three specialized agents directly within the core process. This is also where many multi-agent efforts fail, due to lack of coordination rather than individual agent limitations. Lemvigh-Müller was successful by anchoring the solution in an SAP landscape where the data, business rules, and governance frameworks are already well-established.”
“Innovation is not about enterprise scale. Lemvi Müller shows how Danish organizations with short decision paths and a pragmatic approach to technology can advance faster than many large global companies still in the planning stages. Ten weeks from idea to product is far from the norm, but perhaps it should be.”
Designed for operation and scalability
This solution was implemented in close collaboration with NTT Data Business Solutions. NTT Data Business Solutions was responsible for operationalizing the solution and fully integrating it into Lemvigh-Müller’s SAP environment.
“By distributing responsibility across multiple AI agents, Lemvigh‑Müller was able to automate complex processes without losing transparency or control. This enables a quick and safe transition from pilot to production and ensures a more robust solution that can be easily scaled as new requirements emerge,” said Christian Dahl, SAP UX Manager at NTT Data Business Solutions.
According to Dahl, the modular, agent-based architecture was a key element in moving efficiently from proof of concept to production.
The first step in a broader AI agent strategy
Initially, AI agents were deployed in selected supplier inboxes and business areas. However, Lemvigh‑Müller already sees great potential in applying the same agent-based approach across additional management processes.
“This is the first AI agent solution we have deployed in production, and that experience has given us the confidence to consider similar approaches in other areas, such as invoice processing and order management,” concludes Aakerlund.
Ellen Vig Nelausen is a Nordic unified communications expert at SAP.
