A guide explains how to use a Python script and Claude to classify support tickets, addressing the problem of manually cleaning messy CSV exports where similar complaints are worded differently. The process involves reading a real CSV of tickets, cleaning inconsistencies with plain code, and using Claude for complex categorization tasks, with uncertain items flagged for human review. The setup requires Python, a Claude API key, and about twenty minutes, and the author generates a test file to ensure script accuracy. The guide emphasizes that simple tasks like counting tickets per day or fixing whitespace should be done with pandas to save money, while Claude is reserved for understanding language nuances, such as recognizing that “can’t log in again, ugh” and “can’t authenticate after password reset” are the same complaint. Step 1 involves creating test data by installing three libraries—anthropic, pandas, and python-dotenv—via `pip install anthropic pandas python-dotenv`, then running a provided code snippet that builds a messy tickets CSV using random templates for login and billing issues, such as "cant login again ugh" and "charged $49 twice this month??".