01
Client Profile & Background
A large-scale manufacturing company in Bekasi, Indonesia, with over 5,000 SKUs and daily order volumes exceeding 10,000 line items. The company operates three production facilities and a central warehouse, supplying automotive components to major OEMs. Their existing ordering process relied on manual data entry from paper-based purchase orders and invoices, leading to bottlenecks and errors. The IT environment consisted of legacy servers from HP and Dell, with storage from Synology. The company had already invested in VMware virtualization and Backup & Disaster Recovery solutions but lacked automation for document processing. Their goal was to digitize the entire ordering workflow to improve accuracy and speed.
02
Technical Challenge
The manual order processing resulted in an average of 12% data entry errors, causing shipment delays and inventory discrepancies. Each order took 15 minutes to process manually, leading to a backlog of 2,000 orders per day. The company faced a 24-hour lag in updating inventory levels, resulting in stockouts and overstock situations. Additionally, the existing OCR solution had only 70% accuracy for Bahasa Indonesia documents, requiring extensive manual verification. The IT team needed a solution that could handle mixed-language documents (Bahasa Indonesia and English) with complex table structures. The system had to integrate with their existing ERP (SAP) and warehouse management system (WMS) without disrupting operations. Security requirements mandated on-premise deployment due to data sensitivity, with Cyber Security compliance for supplier data.
03
Implemented Solution
We deployed an enterprise-grade OCR AI system using Microsoft Azure Cognitive Services for OCR, customized with custom models trained on Indonesian business documents. The solution runs on Lenovo ThinkSystem SR650 servers with Synology RS3617RPxs storage for high availability. For disaster recovery, we implemented Veeam Backup & Replication with replication to a secondary site. The system integrates with SAP via REST APIs, automatically creating purchase orders and updating inventory. We also deployed Enterprise WiFi for mobile scanning devices in the warehouse. The OCR pipeline includes pre-processing, layout analysis, and machine learning-based text extraction, achieving 99.5% accuracy. A human-in-the-loop validation step handles exceptions, reducing manual effort by 80%.
04
Results & ROI
Post-implementation, order processing time dropped from 15 minutes to 30 seconds per order, a 97% reduction. Data entry errors decreased from 12% to 0.5%, virtually eliminating shipment errors. Inventory accuracy improved to 99%, reducing stockouts by 60%. The system processes 12,000 orders daily with 99.5% accuracy, handling peak volumes without delays.
The ROI was realized within 6 months: annual savings of $500,000 from reduced labor costs and error-related penalties. The RPO for order data improved from 24 hours to near real-time, with backup windows reduced by 90%. The solution scaled to support new product lines without additional IT overhead. The company now plans to extend OCR AI to accounts payable and shipping documents.
