Robotic Process Automation (RPA) leverages software robots to streamline repetitive rules-based tasks, enhancing efficiency and reducing errors. Advances in Robotic Process Mining (RPM) and Task Mining (TM) enable the identification and segmentation of automatable routines from user interaction logs. However, despite these advances, significant gaps and challenges persist in various stages of the RPM pipeline. These challenges hinder the effective discovery and optimization of routines, limiting the efficiency and robustness of the resulting software robots. This work systematically identifies and organizes these challenges in a structured framework. Drawing on previous research, we define four key categories of routine optimization issues. This classification provides a foundation for analyzing and addressing existing gaps, offering a broad perspective of the complexities involved. By developing and applying these categories, we provide a flexible framework to guide further research on routine optimization for improving software robots.