Descriptions:
VelvetShark demonstrates a complete, no-Python-required data analysis workflow using Claude Code on a real-world dataset: 541,909 transaction rows from a UK-based gift retailer sourced from the UCI Machine Learning Repository. The goal is a presentation-ready Excel workbook with KPIs, top products, top customers, sales trends, country breakdowns, and a customer cohort retention heat map—generated from a single prompt with no manual code writing or debugging by the user.
The workflow starts with guardrails: Claude Code is instructed to load the data and report findings rather than estimate, allowing the user to verify date ranges and row counts match expectations before any analysis begins. Returns and cancellations are deliberately kept in the dataset rather than dropped, enabling return rate analysis by product and seasonal cancellation patterns. Claude Code encounters and self-corrects several errors during script generation, ultimately producing a parameterized Python script (backed by pandas and matplotlib) that completes in 4 minutes and 24 seconds, outputting charts and a six-sheet Excel report including a QA reconciliation sheet that cross-validates net revenue using two independent calculation methods.
A CLAUDE.md file persists preferences—chart style, package manager (uv), output paths—so future sessions inherit the same configuration automatically. Parameters like top-N products, date range, and aggregation frequency (daily/weekly/monthly) can be passed at the command line, making the script reusable whenever fresh data arrives. The video is aimed at analysts and business users comfortable with Excel but unfamiliar with Python, arguing that knowing what questions to ask your data is the transferable skill.
📺 Source: VelvetShark · Published January 16, 2026
🏷️ Format: Workflow Case Study
