Recommendations

What the review checks

The agent uses fixed rules. The same data always gives the same result, and a rule says nothing when there isn’t enough data.

Finding	When it’s raised
Spend spike	The last 3 days’ daily spend is at least 1.5 times the 14 days before, and cost per conversion rose by 30% or more
Low ROAS	Return over the last 14 days is below the minimum you set
Missing costs	Variants have no unit cost
Out-of-stock products	Active products have no stock while campaigns run
Low-margin products	Products are below your minimum margin while campaigns run
Stale catalog	Your catalog hasn’t synced for 48 hours
Stale ad data	Ad data is more than 3 days old

A rise in spend with proportionally more conversions isn’t flagged.

Reading a recommendation

• Issue and risk: what was found and how risky acting on it is.
• Evidence: the numbers behind it, with the date of the data.
• Proposed action: what the agent suggests, and how to undo it.
• Policy check: how the action fits your limits.

Policy check results

Result	What it means
Advice only	Nothing is sent. You act on it and mark it done.
Waits for approval	Nothing happens until you approve.
Automatic	It fits limited automatic mode and is applied within your limits.
Blocked	It breaks one of your limits. The reason is shown.
Can’t make this change	The channel doesn’t support it from the app. You can make it in Ads Manager and mark it done.

What you can do

Approve, reject or dismiss, or mark an item done. Everything goes into the audit log. When you approve, the app checks your policies again with fresh data.

A recommendation you reject or finish won’t be raised again for a week. While one waits for you, its evidence keeps up with each new review. The proposed action stays as it was.