At first glance, order confirmations seem like a minor process. However, over the course of a year, they tie up many hours of operational capacity in procurement that are then missing for negotiations and supplier development. This webinar shows how a manufacturing company handed this verification process over to an AI agent and now books four out of five order confirmations without manual intervention.

Oliver Kosalla (Head of Procurement at Anton Debatin) and Denise Nöske (Operational Procurement at Anton Debatin) join Moritz Flick (Customer Development at Tacto) to discuss how they automated order confirmation verification with the Tacto AI agent, which steps made the difference, and which projects are coming next.

Initial situation: When 2,700 order confirmations a year tie up 225 hours

‍

Anton Debatin is a family-owned company based in Bruchsal that manufactures flexible logistical packaging solutions made of polyethylene. Before the project, around 2,700 order confirmations per year from approximately 330 active suppliers were handled completely manually in procurement. Each confirmation arrived via email, was compared with the purchase order for item quantity, price, and delivery date, marked in the ERP system, and filed in the document management system.

With a verification time of three to eight minutes per transaction, this added up to more than 200 hours per year. Added to this were volatile prices and availability, increasing compliance requirements such as the PPWR, and the realization that strategic work in procurement was being neglected. The speakers noted that a reliable, central database is a prerequisite, as automation only works with structured and available data.

‍

Document understanding instead of masks: How the AI agent reads and reconciles order confirmations

‍

Existing solutions quickly reach their limits. PDF extraction with OCR only assigns items based on text, masks and templates must be created for each supplier and break with every format change, and image recognition requires large amounts of training data and, even with the best providers, achieves a maximum of 40 percent straight-through processing.

In contrast, the Tacto agent reads every format from every supplier on the first attempt, reconciles item numbers, quantities, prices, delivery dates, payment terms, and Incoterms with the order data from the ERP at the line-item level, and even recognizes handwritten changes on the confirmation. Discrepancies appear highlighted in color in the app, including an AI-generated suggestion for the email to the supplier. The practical examples from Debatin showed a price discrepancy from 32.13 euros to 31.13 euros per kilogram, where a discount was missing from the order, as well as a delivery date delayed by one day. Tolerances can be set down to the supplier level, and the system learns from every adjustment for all future confirmations.

‍

From specifications to straight-through processing: Project workflow and results at Debatin

‍

Today, around 80 percent of order confirmation verification runs without manual intervention; only the approximately 20 percent with discrepancies go back to the procurement team. This frees up around 210 hours annually and, according to the business case, results in savings of approximately 12,600 euros per year for this process alone. More important than the numbers for the speakers is the capacity gained for strategic topics in supplier management. In addition, there is greater resilience during vacations and illness, as well as less waste because deviating confirmations are noticed early.

‍

Outlook: After implementation is before implementation

‍

The order confirmation verification was the starting point for Debatin. Next, an automatic supplier evaluation is being developed that combines data from supplier self-disclosures in the Tacto SRM with delivery reliability and quality from the ERP. The goal is to use it for upcoming annual reviews by the end of the year. In the fourth quarter, the team will also test an RFQ agent for daily inquiries in the merchandise sector, which currently cause a great deal of administrative effort. Invoice verification, which is currently handled entirely by hand, is also on the list.

‍

Conclusion

‍

The practical example of Anton Debatin illustrates that operational routines in procurement can be gradually handed over to AI agents, even in medium-sized companies. The keys to success are a clean database, a clearly defined start with a few suppliers, and a team that helps shape the process from the beginning. The time saved flows into strategic tasks, and the first agent gradually becomes a team of AI colleagues.

‍

Oliver Kosalla and Denise Nöske from Anton Debatin, together with Moritz Flick from Tacto, show how the verification of around 2,700 order confirmations per year now runs about 80 percent without manual intervention. The webinar explains how the AI agent detects discrepancies in price, quantity, and delivery date, how Debatin started the project with five suppliers, and which agents are coming next.