Machine Learning-Based System for Document Management Optimization: A Case Study of a Small and Medium-Sized Enterprise

Mansilla O.; Lopez-Haya R. · CEUR Workshop Proceedings (2026)

Resumen

This study addresses inefficiencies in administrative document management at Kazoku No Mizu SAC (Peru), where manual transcription processes create operational bottlenecks, resulting in an average processing time of 70.64 seconds per document. Despite advances in Optical Character Recognition (OCR), conventional approaches exhibit limitations in handling noisy handwritten data, particularly in resource-constrained environments such as small and medium-sized enterprises (SMEs). To address this gap, the study proposes a hybrid Intelligent Document Processing (IDP) system based on Machine Learning that integrates computer vision through YOLOv8 for region-of-interest detection and a Large Language Model (Gemini 2.5 Flash API) for semantic validation and intelligent data recovery.The research adopts an applied, quantitative, non-experimental, and longitudinal design, analyzing a sample of 119 documents (n = 119). System performance was evaluated by comparing manual and automated processing outcomes using a paired-samples Student’s t-test. The results demonstrate a statistically significant reduction in average processing time, from 70.00 seconds to 3.86 seconds per document (t(118) = 90.21, p < 0.05), achieving a temporal efficiency factor of 18.1 relative to the manual approach. The Automation Rate (AR) reached 98.66%, while the Intelligent Recovery Rate (IRR) achieved 89.77%, enabling the recovery of ambiguous information beyond the capabilities of conventional OCR systems. Furthermore, Data Accuracy (DA) reached 66.39%, representing a substantial improvement over the baseline scenario. However, inferential analysis revealed that human validation achieved significantly higher accuracy levels (89.83%) than the automated system (66.39%) (t(118) = −10.45, p < 0.001).The findings indicate that hybrid AI architectures significantly enhance operational efficiency while providing effective semantic support for document interpretation and recovery. The proposed solution aligns with the quality characteristics established in ISO/IEC 25010 and ISO/IEC 25012 standards, highlighting the potential of cost-effective IDP systems to optimize high-volume document processing workflows in SMEs operating under resource constraints.

Instituciones: Universidad Tecnológica del Perú