The first step to automating the whole order process was processing the emails and collecting and understanding the context. Automating the email processing workflow came with several challenges. We had to understand the client's complex workflow and build the application from scratch in a user-friendly way. This meant working closely with them to refine requirements and expectations. The next hurdle was handling multiple email formats from various dropshippers, each with their unique structure and labels for order confirmations, invoices, and shipment details. On top of that, we were working against a tight deadline before the peak season, which meant developing multiple modules in parallel. Finally, we had to optimize the AI solution to keep costs under control without compromising accuracy or efficiency. To address these challenges, we focused on
- Categorizing Emails Effectively: Since there were dozens of dropshippers, emails arrived in different formats. The system had to classify order confirmations, invoices, and shipment details while filtering out irrelevant or duplicate emails.
- Improving AI Consistency: Implemented an error-handling mechanism to refine OpenAI-generated data and ensure reliability.
- Handling Multiple File Formats: We implemented PDF-to-image conversion before extracting data to ensure accurate text recognition. Using OCR (Optical Character Recognition) technology, we then extracted structured data from PDF, CSV, and XLSX attachments efficiently.
- PlentyMarkets Integration: Developed API-driven automation to update order details without manual intervention. Additionally, we provided an interface that allows the client to view data from PlentyMarkets and emails side by side, significantly reducing processing time.
- Optimizing Performance: Configured the system to fetch and process emails instantaneously, ensuring seamless operation without delays. The system is also designed to handle high data loads efficiently, maintaining performance even during peak usage.
