The world of AI has been increasingly relevant, especially the matter of Large Language Models (LLM), reaching across multiple aspects of work, including business and data analysis. This study explores the capability of LLMs to extract meaningful insights from textual survey data using three established methods: Few-Shot Prompting, Chat Completion, and Fine-tuning. Each approach was systematically assessed for accuracy and cost-effectiveness. Our experiments on a synthetic corpus feedback of students revealed that costs are closely tied to the performance of each method, offering valuable insights into their practical applications and trade-offs.