Claude-Powered Email Order Automation at Meesenburg
Meesenburg is a leading distributor of building materials and construction supplies, serving professional customers across Germany. Operating in a high-volume environment where speed and accuracy are critical, they continue to invest in digitalization to strengthen scalability and service quality.
Meesenburg was processing a large volume of customer orders via email, but these emails varied wildly in structure and quality. Many lacked standardized product codes, used free-form descriptions, or included scanned documents, photos, and handwritten notes as attachments. Employees had to manually interpret each order, match products against internal catalogs, correct ambiguities, and prepare structured outputs for ERP entry. The process was reliable but time-consuming, and increasingly difficult to sustain as order volume and complexity grew. The core challenge wasn’t volume alone — it was variability.
Claude-Powered Email Order Automation
We built an AI-powered order automation system on Claude, Anthropic’s AI model, integrated through the Claude API. The system monitors a designated inbox, interprets incoming orders, and sends validated orders directly into Meesenburg’s ERP. Claude reads both the email body and its attachments, so orders can arrive as free-text emails, scanned documents, photos, PDFs, handwritten notes, Word files, or Excel files, with no standardized format required. It extracts products and quantities, matches free-text descriptions to Meesenburg’s internal product catalog, and routes unclear cases to an operator for human-in-the-loop validation. The solution runs continuously in production and improves over time by learning from historical order data and user corrections.
The results: according to Meesenburg, approximately 50% of orders received by email are currently processed automatically. In a performance evaluation focused on actual order emails, 98% required no manual modification once processed. Mapping accuracy improved steadily as the system learned, with fully and partially mapped orders consistently reaching 75 to 80% and manual intervention dropping close to zero in later stages.
Smart email order automation
The features
