2 min read
In brief
- Every conversion report is incomplete at the recent end, because attribution happens when the conversion occurs, not when the click did.
- Comparing a finished week to an unfinished one produces a false decline that reruns every Monday.
- Bid and budget decisions should read windows older than the attribution window, or use the platform's own modelled figures knowingly.
The mechanism
A click on Monday can convert on Wednesday or the following week. Google credits the conversion to the click date. So Monday's column keeps growing after Monday has passed, and today's report understates every recent day.
This is not a bug and not unique to Google. It means the shape of the last few days is always a partial curve, and that any comparison between a completed period and the current one is comparing a finished number to an unfinished one.
The rule of thumb
If your attribution window is seven days, treat the most recent seven days as provisional. If it is thirty, treat the most recent thirty as provisional. There is no setting that makes this disappear; there is only a convention for reading it.
Reading it without being fooled
- Compare complete weeks, never a complete week against this week.
- When you must look at recent days, compare dates that are all the same distance from today — day 1 to day 7 of last month against day 1 to day 7 of this month, both still settling.
- Track a cohort view: conversions from clicks in a given week, measured after the window has closed. It is the only stable series.
- Do not pause a campaign because yesterday looks weak. Yesterday is the least reliable day in the report.
| Decision | Read from |
|---|---|
| Pause or scale a campaign | Complete weeks, older than the attribution window |
| Compare two creatives | Equal-age cohorts, same number of days settling |
| Check whether tracking broke | Today's clicks and yesterday's events, not conversions |
| Report to a client or owner | Last complete period, with the provisional days labelled |
When it is not lag
Lag explains a growing recent column. It does not explain a series that falls in absolute terms for older dates, an event that stops entirely, or a step change that persists beyond the attribution window. Those are tracking failures and should be investigated with the event log rather than waited out.
The last few days of any conversion report are a curve in progress. Read completed windows for decisions, label provisional days honestly, and treat any change that survives the full attribution window as a real problem rather than lag.
