A study on optimizing large language models for enterprise workflows reveals that selective context retention and summarization significantly enhance efficiency and reliability in automated tasks like expense itemization. The best configuration achieved a 91.6% completion rate while reducing token usage and processing time compared to full context retention methods.