Why an average CPA is the wrong question

A campaign with a CPA of 40 euros can describe two very different accounts: one that sits stably around 40 euros everywhere, and one that hits 15 euros on desktop during the day and 90 on mobile in the evening. The average is the same number in both cases, and the wrong question to ask in both cases.

Why the average survives the diagnosis

The question that does work is a breakdown: by time of day, by device, and where relevant by audience. Not because more detail is always better, but because a bid strategy at account level reacts to the average, while costs are generated by time of day and device. A tROAS target that fits the average therefore does not really fit either segment -- it is calibrated to a number that no single hour of the day actually produces.

A worked example

Illustrative case: an account reports a stable 40-euro CPA for a full month, nothing in the headline number suggesting a problem. Split by device and hour, the picture splits in two: desktop traffic between 9am and 6pm converts at roughly 15 euros, evening mobile traffic between 8pm and midnight converts at roughly 90 euros. The blended number was never wrong, it was just answering a question nobody asked -- how does this account perform on average, when the useful question was where does it perform differently.

The common mistake

The mistake that follows from trusting the average is adjusting the bid strategy for the whole campaign in response to a problem that lives in one segment. Lowering the account-wide tROAS target to fix an evening mobile spike also throttles the desktop traffic that was never the problem, trading a fix in one segment for a new one in a segment that used to work fine.

What to check first

In practice this means: before adjusting a bid strategy, first split the period where the problem occurred by time of day and device. A spike confined to a few evening hours on mobile calls for a different fix -- a bid adjustment scoped to that window, or a look at what changes on mobile specifically -- than a structurally too-high CPA across the whole day. Ctrl PPC's 6-step Decision Framework builds on this: the hypothesis that follows from a signal points to the segment where the problem actually sits, not to the account as a whole.