In online learning settings interactive and meaningful feedback is becoming increasingly important. However, feedback from teachers is oftentimes missing in online learning settings. To overcome challenges that arise from the missing representation of teachers, our study analyzes the relevance of avatar designs in learning settings. We therefore rely on avatars as game design elements and analyze how their design can influence emotional attachment, learning process satisfaction, and extraneous cognitive load in learning. To achieve our goal, we conduct a qualitative comparative analysis with 998 datasets that were collected in a 2x2x2 pre-post online experiment that was developed to train participants in learning functions in Excel. Our results indicate that interaction, familiarity, motivation, and aesthetic experiences are important configurations for avatars that are used in learning. We contribute to different streams of theory such as self-expansion and guide practitioners by providing implications about how to create meaningful avatar designs for learning applications.