Stop 7 of 9 · Daily
Quality control for juniors
A build checklist Claude runs before Nour looks, feedback the juniors can act on, and a defect count that should fall.
What it is
Sam and Priya build campaigns. Nour signs them off, which means every mistake they make is eventually hers. The old shape of this was a forty-minute review she kept postponing. The new shape is a written checklist Claude runs first, so she reads failures rather than everything, and the juniors get feedback the same day instead of the same month.
The checklist
The checklist is the agency setup standard turned into pass or fail items. Every line has to be checkable from the build export or a screenshot, because a line that needs judgment belongs to Nour, not to the check.
| Area | Items that are pass or fail |
|---|---|
| Tracking | Correct conversion action, primary flag set, tracking template present |
| Targeting | Location set to presence not interest, radius matches the brief, language set |
| Structure | Ad group themes match keywords, no single ad group over the agreed keyword count |
| Copy and assets | Required extensions present, headline count, no banned compliance phrases |
| Budget and bidding | Budget matches the brief, bid strategy matches the brief, no unapproved change |
The last line matters more than it looks. If a junior build arrives with a bid strategy nobody approved, that is the same class of problem the guardrails exist for: no bid, budget, status, negative, audience or placement setting changes without the person who owns the account agreeing to it.
[paste or upload the junior campaign build export and the campaign brief] Run every item in checklist.md in this Project against this build. Return a table: item, pass or fail, the evidence you used, and for each failure the exact correct value from the brief. Then list separately the two or three things the checklist cannot judge that I should look at myself. Do not rewrite the campaign, and do not propose any change to bids, budgets, status or targeting beyond restoring what the brief specifies.
The feedback note
Feedback that lands names the rule and shows the cost, in under 150 words. Nour never sends the checklist output raw, because a table of failures reads as a scorecard.
From the checklist results, write the note I send Sam. Lead with what he got right and be specific about it. Then each failure as: what the setup standard says, what the build does, and what it would have cost in the first two weeks if it had gone live. End with one thing to change in his own process, not mine. Under 150 words, direct, no scoring.
Here are the checklist results from the last eight builds by Sam and Priya. Rank the defect types by how often they recur and by their likely cost. Tell me which two are worth a thirty minute session, which one is a documentation fix rather than a training issue, and draft the checklist line that would catch the top defect earlier.
The measurable
Defects per build, counted from the checklist, tracked per junior per month. It should fall. If it is flat while the build volume is flat, the checklist has stopped covering what actually goes wrong, and the last prompt above is how you find out what to add.