This paper explores the development of the Elevator Pitch Intelligent Companion (EPIC), an Intelligent Tutoring System (ITS) designed to improve students’ oral communication skills. Three design variants of EPIC were evaluated on a group of 246 engineering students at a Latin American university, focusing on different task selection strategies: one with a single, iterative task, another with different tasks solved in one step, and the last involving tasks with dependent sub-tasks. Evaluations, considering usability, learning, and student engagement, revealed that no one design was superior in all aspects. Each model showcased its unique strengths and weaknesses, which emphasizes the complexity of designing ITS systems for ill-structured domains. The findings provide valuable insights into ITS design for such domains.