Intelligent Document Routing System for Efficient Office Automation

Thesis Abstract

In modern office environments, information workflows are frequently hindered by manual routing, inconsistent document handling, and silos between departments, leading to delays, misclassification, and reduced process transparency. This study addresses the persistent inefficiencies in paper-based and semi-structured document routing by developing an intelligent document routing system (IDRS) that combines natural language processing, machine learning-driven classification, and rule-based workflow orchestration to optimize assignment, prioritization, and delivery of documents across organizational units. The aim is to design, implement, and evaluate an ICT-driven solution that enhances accuracy, reduces cycle times, and improves user satisfaction in office automation contexts. Specific objectives include (1) to articulate a formalized routing taxonomy for office documents; (2) to develop a hybrid IDRS integrating ontological document representation, supervised text classification, and decision-rule engines; (3) to evaluate system performance against baseline manual routing in terms of accuracy, throughput, and user-perceived latency; (4) to assess impacts on cross-functional collaboration and process transparency; and (5) to provide a scalable deployment framework with governance and security considerations. The methodology adopts a mixed-methods research design combining engineering development with empirical evaluation in a real-world organizational setting. The population comprises 12 departments within a mid-to-large enterprise and a data corpus of 8,000 anonymized documents generated over six months, including invoices, purchase orders, HR forms, and internal memos. A stratified sample of 1,200 documents is annotated by domain specialists to establish ground truth labels for routing categories. The IDRS prototype is implemented using Python-based NLP pipelines (spaCy, transformer-based embeddings) for document encoding, a supervised multilabel classifier (BERT fine-tuned on domain-specific corpora) for topic and urgency labeling, a knowledge graph for entity and route reasoning, and a rule-based engine (Drools) for workflow orchestration. For evaluation, three data collection instruments are employed system logs capturing routing decisions and latency, a standardized user satisfaction survey, and semi-structured interviews with process owners. Validity and reliability are ensured through inter-annotator agreement (Cohen’s kappa > 0.8 on labeling), cross-validation (5-fold) for classifier performance, and triangulation of quantitative metrics with qualitative insights. Data analysis follows a multi-layered approach. Quantitative analysis includes descriptive statistics, precision, recall, F1-scores for multilabel routing accuracy, and time-to-route metrics analyzed via paired t-tests and ANOVA to compare the IDRS against baseline manual routing. Regression analysis investigates determinants of routing efficiency, while survival analysis assesses document handling time reductions. Qualitative analysis employs thematic analysis of interview transcripts to extract perceived barriers, facilitators, and acceptance factors, aligned with the Technology Acceptance Model (TAM) and the Diffusion of Innovations theory. A conceptual model maps the interaction between document features, classification outputs, routing rules, and workflow outcomes, validated through system usability testing and expert reviews. Expected findings indicate that the IDRS achieves a significant improvement in routing accuracy (F1-score above 0.92 on core document classes) and reduces average routing latency by 38% compared to manual processes. The integration of ontological representation with a supervised classifier is anticipated to outperform purely rule-based or purely statistical approaches. Enhanced traceability, auditability, and role-based access controls are expected to bolster compliance and governance. The study anticipates positive effects on interdepartmental collaboration and perceived process transparency, moderated by user trust and perceived usefulness, as suggested by TAM. The study contributes to knowledge by demonstrating a scalable, cross-domain Intelligent Document Routing framework that blends NLP, knowledge graphs, and rule-based orchestration to transform office automation. It provides empirical evidence on the performance and governance implications of AI-powered routing in organizational workflows, including a deployment blueprint, data governance practices, and a risk mitigation plan for sensitive information handling. The principal conclusion posits that AI-enabled document routing substantially improves efficiency and governance in office automation, given adequate alignment with organizational policies and user-centric design. Practical recommendations include developing domain-specific ontologies, adopting iterative deployment with pilot testing across departments, investing in change management to foster user adoption, and implementing continuous monitoring to ensure model fairness and accuracy over time.