Purpose of this document:
To explain how to detect week-over-week CPA drift at both the campaign and ad-group level, and when to investigate.
Why this procedure exists
CPA drift is one of the fastest signals that something has broken or shifted in the account — a search-term cluster going wrong, a competitor entering the auction, conversion tracking misfiring, or a landing page regression. Catching it weekly means investigating while the cause is recent and reversible. Waiting until the monthly review means three weeks of compounded waste.
There are two drift signals:
- Campaign-level CPA — the headline metric; jumps usually mean something account-wide shifted
- Ad-group-level CPA — local drift; usually means one ad group started attracting bad searches or its ad copy went stale
When this step doesn’t apply
Skip in the first 30 days. Brand-new campaigns don’t have a stable baseline; weekly CPA will swing wildly while the campaign learns. Wait until at least 4 weeks of consistent conversion data before treating WoW jumps as drift signals.
How it works
The check is a week-over-week comparison of Cost / Conv. at two levels: campaign-wide and per ad group. A jump past the threshold is the trigger to investigate — not to act. The action (pausing keywords, adding negatives, refreshing ad copy) lives in the corresponding daily or weekly optimization once the cause is identified.
The ad-group check carries a “wasn’t there last week” qualifier. Persistent 2× ad groups are structural problems for the monthly review; what the weekly check catches is the NEW spike.
The Process
-
Open the campaign → Overview
-
Set date range to Last 7 days, with comparison set to Previous 7 days
-
Note the campaign-level Cost / Conv. for both windows
-
Drill into Ad Groups (left sidebar) and sort by Cost / Conv. (descending)
-
Check both levels against the decision rule:
Level Condition Action Campaign 20%+ WoW increase, no obvious explanation Investigate Campaign 20%+ WoW increase with a known cause (budget change, ad-copy rotation, seasonal shift) Note the cause; no investigation Ad group At 2× campaign avg AND wasn’t there last week Investigate (drift signal) Ad group At 2× campaign avg, persistent multi-week Note for monthly review (structural, not drift) Either Below threshold No action -
For each flagged item, drill into its search terms, keywords, ad copy, and landing page to identify the cause
-
If a cause is identified, the fix lives in the corresponding daily or weekly optimization. If no cause is identifiable, re-check next week
Edge cases
Edge case 1: campaign had a major change last week (budget, bid strategy, ad copy rotation)
Treat the first 1–2 weeks after a major change as a recalibration period. CPA will move; that’s expected. Don’t flag it unless the magnitude is unusual (50%+ rather than 20%).
Edge case 2: ad group has very few conversions (1–3)
A 2× campaign-average CPA on a 2-conversion ad group is statistical noise, not drift. Don’t flag ad groups with fewer than 5 conversions in the comparison window — the variance is too high to call.
Edge case 3: seasonal niche in a known low-conversion window
HVAC keywords in October. Lawn care in January. The drift will look real but is seasonal. Note the seasonality in the weekly log and skip the investigation.
Edge case 4: campaign-level CPA improved 20%+ WoW
Note it but don’t investigate the same way. Track the improvement direction in the log so future-you knows what’s working. Usually one of: a winning ad-copy rotation kicking in, a newly-promoted converting keyword, a competitor pausing, or seasonal tailwind.