AI-Driven Virtual Assistant for Enhancing Corporate Secretarial Compliance

Thesis Abstract

The rapid digitization of corporate governance processes has intensified the complexity of secretarial duties, yet many organizations struggle with ensuring timely regulatory filings, accurate record-keeping, and consistent compliance across jurisdictions. This study investigates the design, deployment, and evaluation of an AI-driven virtual assistant (VA) to enhance corporate secretarial compliance by automating routine tasks, improving document governance, and supporting decision-making in risk-prone areas. The aim is to develop a technically feasible VA prototype that integrates natural language processing, workflow orchestration, and policy-aware reasoning to reduce non-compliance risks and operational costs. Specific objectives are (1) to identify key compliance touchpoints within secretarial functions across multinational corporations; (2) to design an AI VA architecture incorporating NLP, knowledge graphs, rule-based engines, and secure data pipelines; (3) to evaluate the VA’s impact on accuracy, speed, and consistency of compliance tasks; (4) to examine user acceptance, trust, and perceived usefulness among corporate secretaries; and (5) to propose an implementation blueprint and governance framework for scalable deployment. The study adopts a mixed-methods, sequential explanatory design, underpinned by the Technology Acceptance Model (TAM) and the Information Asymmetry Theory to explain adoption and risk mitigation dynamics. The population comprises corporate secretarial teams in 20 multinational enterprises across finance, technology, and manufacturing sectors. A purposive sample of 60 secretarial professionals will be recruited for qualitative interviews, while a stratified random sample of 120 secretarial tasks will be analyzed quantitatively. The VA prototype will be implemented in a sandboxed environment using a corpus of 2,500 regulatory documents, 300 board resolutions, and 1,200 governance templates sourced from consenting organizations. Data collection instruments include task performance logs, system usage analytics, structured interview guides, and a standardized post-implementation survey measuring perceived usefulness, ease of use, trust, and intention to continue use. Validity and reliability will be ensured through triangulation of interview data with system metrics, pilot testing of instruments, and inter-rater reliability checks for qualitative coding. Quantitative analyses will employ descriptive statistics, paired t-tests, and regression analyses to compare pre- and post-implementation performance across timeliness, accuracy, and completeness of filings, with a difference-in-differences approach to control for confounding factors. Qualitative data will be analyzed using thematic analysis to identify patterns in user experiences, perceived barriers, and enablers for VA adoption. A knowledge graph-based reasoning module will be evaluated for its ability to map regulatory requirements to organizational processes, with performance assessed via precision, recall, and F1 scores against a gold-standard annotated set. The study will also test the integration of policy-aware constraint satisfaction to ensure generated recommendations comply with jurisdictional requirements, and a privacy-preserving architecture to safeguard confidential board information. Expected findings include (i) measurable improvements in task accuracy and turnaround times for key secretarial activities (e.g., filing deadlines, minutes transcription, governance compliance checks) and a reduction in accidental non-compliance instances; (ii) positive shifts in perceived usefulness and trust among secretaries, moderated by perceived organizational support and change readiness; (iii) evidence that the VA enhances decision quality by surfacing relevant regulatory obligations and historical precedents; and (iv) identification of critical success factors for scalable adoption, including data quality, governance of the knowledge graph, and clear escalation protocols. The study contributes to knowledge by integrating AI-driven assistance with corporate secretarial practice, extending theory by linking TAM with information governance and risk management in a secretarial context, and providing a reference architecture and implementation roadmap for multinational firms. It also offers practical implications for regulators and vendors regarding data handling, transparency of AI decisions, and auditability. The conclusion anticipates that a well-engineered AI VA can materially improve compliance outcomes while preserving confidentiality and human oversight. Recommendations include adopting a phased deployment strategy with continuous monitoring, establishing formal governance and audit trails for AI-generated outputs, investing in ongoing data curation and domain-specific ontologies, and conducting periodic user training to maintain trust and adoption.