Cleaning an 850-row vendor spreadsheet is exactly the kind of messy ops work AI handles well.

Data cleaning pipelines—fuzzy deduplication, phone normalization, email validation—are structurally identical each run. The same logic applies whether the file has 850 rows or 8,500, making this highly automatable.

Phone formatting and email validation have crisp rules, but deduplication thresholds are inherently fuzzy—'ABC Manufacturing' vs 'ABC Mfg' is obvious, but borderline cases require a judgment call. The merge_notes column mitigates this by surfacing decisions for human review.

The agent needs only the Excel file and standard libraries (pandas, fuzzywuzzy/rapidfuzz, email-validator). No external APIs or live credentials are required, and the file is self-contained.

Incorrectly merging two distinct vendors could corrupt spend tracking or route communications to the wrong contact, which is a real operational problem. However, the output is a new file—the original is preserved—and the merge_notes column makes errors auditable and reversible.

Most of the work is mechanical, but ambiguous deduplication cases (same name, different state or product category) genuinely need a human to confirm. The agent should flag these rather than silently merge them.