CHEQUE READING PROJECT WITH ARTIFICIAL INTELLIGENCE

Ziraat Bank significantly improved customer satisfaction and operational efficiency by introducing an AI-powered cheque processing system that dramatically accelerated transaction speed and reduced error rates. Previously, processing a single cheque took 80 seconds; with the new system, this was cut down to just 19 seconds. Handling an annual average of 2.3 million cheques, the system—capable of processing 15 cheques per second—led to a 48% reduction in staffing needs for this task, resulting in a 23 man/day operational savings. This innovation has positioned Ziraat Bank as a pioneer in the digital transformation of the banking sector, optimizing resource use while delivering faster, more accurate service.

The new system integrates directly with branch cheque scanners, which send images to an AI engine for real-time processing. The AI not only verifies and corrects image orientation but also extracts key information from each cheque. If the DataMatrix code is readable, its information is prioritized; otherwise, the AI generates the missing data, such as issue date, amount, and location. To ensure data integrity, any mismatch between AI-generated and DataMatrix data is flagged in yellow, allowing human review. This intelligent architecture is driven by a deep learning model made up of seven CNN layers and two LSTM layers, trained on 3.2 million data points. With nearly 4 million parameters, the model delivers high-speed, accurate predictions, streamlining one of the bank’s most resource-intensive operations.