Screen a whole list with Deep Runs — your AI intern

The work nobody has time for

Every operator has a version of it: a spreadsheet of 400 creators, or leads, or applicants, and a set of criteria to judge each one against. Done by hand it takes days, so it does not get done — or it gets done badly.

Deep Runs is Zami's engine for exactly this. Hand it a sheet and a task, and it works through every row like an intern who does not get tired — returning a verdict, a score, and a reason for each one.

Do it in five steps

1. Open Deep Runs → New deep run.
2. Upload your sheet (CSV or XLSX). Zami previews the columns and picks the one that names each row.
3. Write the task and list your criteria in plain language — for example, "screen these creators for an AI-literacy campaign across five East-African markets," then your must-haves as a numbered list. Press Optimize to expand a short task into a thorough brief, and attach a criteria document if you have one.
4. Turn on enrichment if you need it. With enrichment on, the intern researches each row (web and platform signal) to fill gaps before it judges — slower and more thorough. Off, it reasons over what is already in the sheet — fast.
5. Watch the grid fill, then export. Each row gets a verdict — Qualified, Maybe, or Rejected — a fit score, a pass or fail on each criterion, and a short rationale. Filter to the Qualified, open any row to see its reasoning and sources, and export the result back to a sheet.

Try this: upload a list of 200 creators and ask, "Qualify each for a skincare launch: must post in beauty, audience mostly Ethiopia, no risky content, real engagement."

What it is great at — and what to double-check

Great at: turning a day of judgment work into a coffee break, consistently, with a rationale you can audit row by row.

Double-check: the intern is bounded on purpose — it reasons over public signal and what you give it, and flags thin evidence. For high-stakes decisions like hiring or big spend, treat the Qualified list as a ranked starting point for a human, not a final verdict. Time and cost scale with enrichment and list size, so start with a sample.