Cleaning and flagging 2,800 product rows is exactly the kind of structured data work AI handles well.

The task is structurally identical every time: parse CSVs, apply formatting rules, join on SKU, compute markup, flag outliers. This is a textbook repeatable pipeline with no instance-specific judgment required.

Success criteria are crisp: fields are standardized, weights are filled from the lookup table where matched, and any product with markup below 35% is flagged. The only mild ambiguity is what 'standardize descriptions' means in edge cases, but that's resolvable with a short spec.

Both CSVs are explicitly available as inputs, and the output is a file for human review before any Shopify push — no live API access or credentials are needed to execute the core task.

The human explicitly plans to review flagged products and the full output before pushing to Shopify, which creates a meaningful checkpoint. Errors in the intermediate file are easily caught and corrected before any live impact.

Markup math, field normalization, and lookup joins are purely mechanical. The human's role is reserved for the post-processing review of flagged items, which is exactly the right division of labor.