We have an agent that flags fatigue when frequency rises and CTR falls for three days. It is right about half the time. The other half it is a weekend or a budget change. Is anyone getting better than a dashboard threshold out of an agent here?
Give it the calendar and the budget log as inputs. Half the false alarms you describe are things the agent could have ruled out if it had been allowed to see them.
And feed it the creative scoring history. Fatigue on a low-scoring asset is not news; fatigue on a top-scoring asset is the signal.
Calendar, budget log, scoring history. Three inputs I was withholding from the thing I was asking to be smart.