EVALUATING GENERATIVE AI-ASSISTED CUSTOMER SERVICECOMMUNICATION THROUGH A HUMAN-IN-THE-LOOP APPROCH

In response to the challenge of analyzing categorical Property Management System (PMS) data where conventional K-Means clustering falls short, a study in the SKANIKA Journal introduces an unsupervised machine learning pipeline using K-Modes clustering and the Kneedle algorithm.

Analyzing 43,340 clean guest profiles from four-star hotels in Bali, the framework utilizes Cao initialization for reproducible clustering, the Kneedle algorithm to objectively identify four optimal guest personas ($k=4$), and Cramér’s V to measure attribute influence. The resulting personas—Domestic Agency-Reliant Guests (50.9%), East Asia Digital Short-Stayers (24.4%), Oceania High-Value Loyalists (14.4%), and Oceania Digital Explorers (10.3%)—provide actionable data driven primarily by gender ($V=0.489$) and length of stay ($V=0.425$), allowing hotel managers to enhance CRM strategies, automate targeted promotions, and improve overall guest satisfaction.