Evaluating the suitability of handwritten text recognition for industrial operations: a practical case study on acceptance sampling inspection workflow

Abstract Handwritten Text Recognition (HTR) is an Artificial Intelligence (AI) technology designed to interpret and digitize handwritten text, enabling automated data extraction and reducing the reliance on manual processing. While HTR has been extensively developed in theory, leveraging advanced machine learning and pattern recognition models, its practical adoption remains uneven. Documented applications exist in fields such as historical document digitization, banking, and healthcare; however, its use in industrial and manufacturing contexts is still largely unexplored. This practical case study explores the use of HTR in an industrial setting by redesigning the operational execution of acceptance sampling activities in a pharmaceutical manufacturing company, in which handwritten records are traditionally used. The performance of the HTR software, which employs a patented hybrid stroke-based recognition method combining neural and statistical classifiers with structural matching, was tested and evaluated in terms of recognition accuracy and process efficiency to assess its potential to meet industrial standards. Preliminary findings indicate that while HTR can improve efficiency, limitations in recognition accuracy, and the need for human validation restrict full automation. Although promising, further advancements in recognition models and system adaptability are needed to ensure reliable and seamless integration into industrial workflows. Graphical Abstract