Skip to main content
Use Intelligence → Creatives to inspect creative insights and Creative analytics to examine observed asset performance. Use Learning when you have a settled control-versus-variant comparison to record. These views do not launch a new ad experiment for you.

​
Inspect creative analytics

  1. Confirm the selected account and open Creative analytics.
  2. Select the available reporting window, from 7d through All, and inspect the returned analysis.
  3. Refresh to recompute from the latest synced data when needed. Inspect individual creative evidence before applying an account-wide conclusion.
Creative duration, repeated attributes, or an observed metric difference do not prove profitability or causation. Check exposure, spend, result definitions, and sample size. Empty or failed analysis needs data recovery, not an invented winner.

​
Create a settled experiment record

  1. Open Intelligence → Learning and select New experiment if another record is open.
  2. Enter Control reference, Variant reference, Objective, and Audience. Choose Metric, Window starts, and Window ends, then enter Source, Provider, and Attribution. Select Create experiment.
  3. Under Add factual result, choose Side and enter Sample size, Spend, Events, and Freshness from the same settled reporting window. Select Record result and repeat for the other side. The metric is calculated from the raw counts.
  4. Inspect Evidence and the evidence check. Resolve missing or inconsistent inputs before treating the comparison as ready.
  5. Use Interpret the result and Record interpretation to describe what was observed. Under Propose learning memory, write a bounded statement and its rationale, then select Propose memory.

​
Review proposed memory

A proposal remains pending until a different eligible reviewer uses Approve or Reject in Learning memory. The proposer and workspace owner cannot review that item. If the page says Needs another reviewer, ask an eligible teammate to review the evidence in their own account access. Record the conditions under which a learning applies; an approved statement is not a universal guarantee. If saving fails, preserve the error and inputs, refresh the intended experiment, and check whether the record already exists before retrying.