Frequency caps are set per platform and enforced per identifier. A person watching across a phone, a laptop and a television is counted as several viewers, and receives the cap several times over.
A cap only sees what its own system sees
Each platform enforces its cap against the identity it holds, whether that is a logged-in account, a device identifier or a browser cookie.
Nothing is shared between platforms, so a cap of three set in four systems permits twelve exposures to the same person.
The plan says one number and the viewer experiences another, with no error anywhere in the execution.
Identity fragments within a single platform too
Logged-in environments track a person reliably across their devices, which is why platforms with accounts enforce caps more accurately.
Open web inventory relies on browser storage, which is cleared, blocked or partitioned, and each reset creates a new person as far as the cap is concerned.
Connected television adds another identifier again, and household-level delivery means several people share it while one device receives the cap.
Why over-delivery concentrates rather than spreads
Optimisation systems favour users they can identify and predict, which means the reachable, stable, heavily tracked portion of the audience absorbs disproportionate delivery.
The result is a small group seeing an ad constantly while a substantial share of the target sees it rarely or never.
Average frequency conceals this completely, because a distribution with a long tail can produce a perfectly reasonable mean.
A plan reporting an average of three may in practice be delivering once to most of the target and twenty times to a small, easily addressable minority.
The cost is not only wasted impressions
Excessive repetition produces irritation, and irritation attaches to the brand rather than to the media plan.
It also wastes budget on people already saturated, at the expense of the unreached portion the campaign was funded to cover.
Both effects are invisible in delivery reporting, which confirms that the cap was respected inside each system.
What can actually be controlled
Frequency distribution, rather than average frequency, is the number worth requesting, and it is available from most large platforms on request.
Consolidating spend into fewer platforms improves control directly, because each additional buying system adds an independent cap.
Where reach matters more than repetition, buying against a reach objective rather than capping an optimisation objective produces a flatter distribution and fewer saturated viewers.