Making Email-Based Order Processing Reliable with LLMs
Industry: Bathroom & Sanitary Equipment
Stable and scalable LLM-based order extraction from customer communication for an international bathroom equipment manufacturer
About the project
As a manufacturer of bathroom equipment serving customers across multiple markets, Radaway processes a high volume of orders that arrive through everyday customer communication, often via email. To modernize this process, the company introduced an LLM-based system designed to automatically extract order data from incoming messages. While the initial implementation demonstrated clear potential, limitations became visible in daily operations. To address these challenges, Radaway partnered with theBlue.ai to refine and strengthen the existing solution. Rather than starting from scratch, the focus was on improving reliability and operational stability so the system could be used confidently at scale.
Challenge
When exposed to real customer communications, the system struggled with inconsistent results. Order details were sometimes misinterpreted and product references didn’t always match the database. Additionally, email attachments were not part of the automated extraction process, which meant that orders sent in files were handled manually at a later stage. Minor discrepancies in phrasing and missing structure often caused downstream failures. The absence of strict output formats or comprehensive validation logic meant the system wasn’t operationally ready for enterprise use, requiring frequent manual intervention. As the solution moved closer to production use, the complexity of operating LLMs reliably introduced challenges that required additional specialized experience to address.
Solution
theBlue.ai began with a detailed technical review to identify and address the core issues affecting system performance. Rather than starting from scratch, the goal was to refine and strengthen the current system to meet real-world operational requirements.
The first step was refining prompt engineering to guide the LLM toward precise order extraction, reducing ambiguity and improving the accuracy of the model’s responses. Structured output schemas were introduced, ensuring that each LLM response adhered to a predictable format, which is essential for a reliable automated process.
Product matching was another critical area of improvement. Semantic alignment based on transformer models was integrated to identify product names accurately, even when the phrasing differed from the database. A final LLM-based validation step was added to evaluate the context before confirming the correct product, significantly reducing mismatch errors and improving data integrity.
To enable processing of attachments, additional capabilities were integrated into the system to extract order data from both the body of the email and its attachments. Separately, an email classification step was introduced to determine whether the content related to placing an order, a cancellation, or another request. This ensured that only relevant content was passed to the extraction and matching stages, reducing the risk of misinterpretation.
Continuous testing across diverse real-world scenarios ensured that the system remained reliable and scalable in production. The improved system was rigorously validated to ensure it could handle a wide range of inputs without failure, meeting the robustness and precision required for enterprise environments.
Results
With support from theBlue.ai, Radaway transformed their initial automation prototype into a reliable, production-ready solution. The system’s order extraction accuracy was vastly improved, and product matching errors were significantly reduced. Manual intervention was minimized as the automation was able to handle a wider variety of input scenarios more consistently.
By improving key components and incorporating new capabilities, the solution became a dependable and scalable workflow that could be trusted to operate across the business with minimal oversight. The collaboration demonstrated how incremental improvements, combined with the right expertise in LLMs, could elevate an existing system to meet enterprise-grade standards.
Contact us
How can we assist you? Please give us a brief description of your project so that we can provide you with the best possible support.