Designing and evaluating AI-assisted secretarial workflows in corporate offices

Thesis Abstract

The study addresses the persistent inefficiencies in traditional secretarial workflows within corporate offices and examines how AI-assisted processes can redefine task automation, scheduling accuracy, and information governance to improve organizational responsiveness and executive support. Despite advances in AI, there is limited empirical evidence on how integrated AI tools—from natural language processing assistants to predictive analytics for scheduling and document management—affect productivity, accuracy, and knowledge management in real-world corporate settings. The aim is to design, implement, and evaluate an AI-assisted secretarial workflow model that augments decision support, reduces cycle times, and enhances data quality across administrative operations. Specific objectives include (1) mapping current secretarial processes and identifying bottlenecks amenable to AI augmentation; (2) developing an integrated AI-enabled workflow prototype incorporating task automation, intelligent routing, and document management; (3) implementing the prototype in three corporate offices with diverse industry backgrounds; (4) evaluating impacts on task cycle time, error rates, and user satisfaction; and (5) generating a scalable implementation framework and governance model for AI-assisted secretarial work. The methodology adopts a mixed-methods design, combining a quasi-experimental intervention with qualitative inquiry. The study will be conducted in three mid-to-large-sized corporations in finance, manufacturing, and professional services, selected via purposive sampling to ensure variation in organizational maturity and digital infrastructure. A total of 90 secretarial staff and 15 senior administrative managers will participate (30 staff per site; 10 managers per site). The AI-assisted workflow prototype will be designed in collaboration with an industry partner and deployed over a six-month period. Data collection instruments include (i) system usage logs and process metrics captured by the prototype (ordering turnaround time, document latency, error frequency, and approval queue lengths); (ii) time-and-motion observations to triangulate workflow efficiency; (iii) pre- and post-implementation surveys measuring perceived usefulness, ease of use, and job satisfaction, based on the Technology Acceptance Model (TAM); (iv) semi-structured interviews with secretaries and managers to explore experiential dimensions and governance concerns; and (v) a formal error audit of critical communications and document outputs. Validity and reliability will be established through pilot testing of instruments, triangulation across data sources, and inter-rater reliability checks for observational data. Analytical techniques will include quantitative analyses using regression methods to assess the impact of AI-assisted workflow on cycle time, error rates, and task throughput, controlling for role seniority and prior digital literacy; repeated-measures ANOVA to examine changes in perceived usefulness and satisfaction over time; and multilevel modeling to account for nested data structures (individuals within teams within sites). Qualitative data will be analyzed thematically following Braun and Clarke’s approach, with coding performed by two independent researchers and adjudicated through consensus to ensure credibility. A conceptual model anchored in the Diffusion of Innovations theory and the Technology-Organization-Environment (TOE) framework will guide interpretation of adoption determinants and organizational context, while Institutional Theory will frame governance and compliance implications. Key expected findings include reductions in average document processing time by 25–40%, a 15–25% decrease in error rates in routine correspondence, and measurable improvements in schedule responsiveness and meeting coordination. The study anticipates that user acceptance will be moderated by perceived interoperability, data governance clarity, and perceived impact on professional autonomy. The contribution to knowledge lies in offering a rigorous, context-sensitive evaluation of AI-enabled secretarial workflows, an operational blueprint for implementing AI in administrative domains, and an empirical test of theoretical models in a corporate administrative setting. The study will conclude with practical recommendations for governance structures, data privacy safeguards, change management strategies, and a scalable deployment roadmap that accommodates varying levels of digital maturity. Policy implications for human–AI collaboration in executive support are discussed, along with limitations and avenues for future research focused on cross-industry generalizability and long-term sustainability of AI-assisted secretarial functions.