What to check before you touch anything
- Google's own rule of thumb is that the learning cycle after a meaningful bid/target change takes about 1–2 weeks (7–14 days), during which performance can swing and won't reflect the strategy's real, settled behavior. But that's a time-based rule of thumb — the real deciding factor is how many conversions you've accumulated since the change (see below).
- How many conversions does the campaign rack up in that window? At low volume, 14 days may not be nearly enough data; you should be judging by accumulated conversions since the change (a common target is 30+), not the calendar.
- The right waiting period depends heavily on how often you get conversions. If your campaign reliably delivers 30+ conversions a week, a two-week window is usually enough for a meaningful read. But if volume is much lower — say, 5–10 conversions a week — 14 calendar days simply isn't enough: hitting that same statistically meaningful number of conversions could take 4–6 weeks or more. Waiting the "standard two weeks" and drawing conclusions from it is a real methodology error in that case — you just won't have enough data yet to separate a real effect from random noise. The right approach: first estimate how many calendar days it will take, at your current conversion pace, to hit a meaningful sample (say, that same ~30 conversions), and use that estimate as your minimum wait — not a flat 14 days.
- Is anything else overlapping this window — seasonal swings, other account changes, external events (sales, holidays)? These can distort your read on the tCPA change specifically. The longer the observation window has to be (which is unavoidable at low conversion frequency), the higher the odds something else sneaks in — worth factoring into how you interpret the result.
- Is there a lag between the click and the recorded conversion? If your sales cycle is long, the data visible at day 7–14 may be incomplete, and the real effect will only show up later. At low conversion frequency, this compounds with the small-sample problem and stretches the reliable evaluation window even further.
- Look at day-by-day or week-by-week trends, not just the average CPA for the period — has it actually settled by the end of the window, or is it still bouncing around with no clear direction?
Possible approaches
- Standard practice is to judge results no sooner than 2 weeks after a change, comparing against a matched-length period before the change, not the whole prior period. This is a reasonable default for campaigns with enough conversion volume.
- Some agencies shorten this window to 7 days for high-volume accounts, since the deciding factor is having enough data, not hitting a calendar minimum.
- For low-frequency campaigns (under roughly 30 conversions a week), many agencies deliberately widen the evaluation window to a month or more, and instead of a hard "check results after X days" rule, they track toward a target number of accumulated conversions since the change — however long that takes. Testing changes on low-volume campaigns more often than the data can support is a common reason targets get "jerked around" without ever really being evaluated.
- If the campaign still hasn't left "Learning" after 2–3 weeks, it's worth digging into why learning is taking so long (see the entry on that above) rather than continuing to judge an unstable period as if it were the final result.
- You can set "checkpoints" in advance — say, day 7 and day 14, or, at low volume, at the 10/20/30 accumulated-conversions mark — and compare the trend between them. If the trend is clearly worsening or clearly stabilizing early, that's informative on its own and doesn't always require waiting out the full planned window.
- Google generally advises against reacting to mid-cycle swings and against making more changes before the learning period wraps up — otherwise the learning clock resets. This matters even more at low conversion frequency, since every premature tweak throws away an already-long data accumulation process.