Guides

Find what worked

Explain which angles, personas, formats and hooks were associated with more results than their spend, with the evidence and its limits.

Use this when you want to know which creatives win, and what to learn from past ads. In Claude Code the plugin starts it with /winandscale:what-worked.

Evidence first

Call meta_get_evidence_health and quote its warnings before any number. Stale sync, coverage gaps and untagged spend all limit what the rest can say. See Evidence health.

Read the history

meta_get_account_history returns lifetime totals per ad. If it is empty, call meta_pull_account_history and wait for it. On Meta's hourly quota the pull is queued and runs by itself.

Get the verdicts

meta_get_history_winners takes a window: 30d, 90d, 180d or lifetime. It returns:

  • a verdict per creative, judged in its own funnel against same-market peers;
  • shared.winning and shared.losing Traits: Values that bought more than their share of spend, with lift, how many distinct creatives carry them, and in how many markets they held;
  • nextTests, each with its basis, limits, reverseIf and prior.

Pivot

meta_get_cross_analysis crosses two Dimensions inside one funnel, such as angle by persona. A missing cell is an untested combination, not a bad one. meta_get_performance_cube splits by ad set or country.

Say what a finding rests on

For each finding, state the spend, the number of ads and the days behind it. Low delivery is inconclusive. Report purchases and ROAS only where the purchase pixel fired.

Everything is observational. Meta chose who saw each ad, so write "was associated with", not "caused", and point to a Test to confirm the findings that matter.

Over REST

Verdicts, Traits and Next tests are MCP-only: meta_get_history_winners and meta_get_cross_analysis have no REST endpoint. REST gives you the inputs:

A prompt to start with

What has worked in our Meta ads over the last 90 days, and why? Start with the
evidence health and tell me what limits the answer.

On this page