The usual suspects
Branded search, broad retargeting, and Performance Max routinely cannibalize organic, email, and SMS. The platforms have no mechanical reason to disclose this: their reporting counts every assisted touch.
How to measure it
A holdout experiment measures it directly. The exposed audience receives ads; the control does not. Cannibalization is the share of attributed orders the control produces anyway.
What to do about it
Cut, cap, or kill the cannibalizing campaign, depending on whether the holdout shows zero, partial, or fully redundant incrementality.