Behind the Scenes: Developing a Cutting-Edge OCR System (for Banks)
Behind the Scenes 🎥: Developing a Cutting-Edge OCR System (for Banks) More than a year ago, we embarked on developing an OCR system aimed at processing payment orders in bank branches across Bosnia and Herzegovina. Given our deep understanding of banking processes, we identified several pain points: processing times between 45 and 70 seconds, long waiting lines that lower customer satisfaction and hinder cross-selling opportunities, bank tellers working under pressure leading to significant input errors and wasted time on corrections, etc. Our targeted solution's characteristics 💡 were clear: 1. Live-feed: Real-time document processing 2. Speed: Processing time under 1.5 seconds 3. Accuracy: Over 95% 4. Cost Efficiency: Comparing favorably to full-time employee costs 5. Security: No cloud technology involved 6. Scalability: ID handling (and other sorts of documents and inputs) However, creating the OCR software alone wasn't enough to achieve our targets 🎯, except for characteristics 5 and 6. To meet the first four objectives, we needed to develop specialized hardware/MF scanner. Testing and fine-tuning the hardware parameters proved to be a challenging yet fascinating journey 🎉, ultimately enabling us to achieve our goals. This approach to product (nominally software) development took us into uncharted territory, providing us with invaluable insights from multiple perspectives. In my next post, I will discuss why we continue to use template matching rather than Intelligent Document Processing (IDP), at least for now. PS: The photo features our WizoDoc hardware, ready for delivery today. WizoDoc - Data Perfection in the Blink of an Eye Graip.AI Nanonets ABBYY Rossum Klippa Agencija za bankarstvo Republike Srpske Agencija za bankarstvo Federacije BiH #banking #automation #OCR #data #processimprovement #processoptimization