The latest week in a performance report almost always looks weaker than the ones before it. In most accounts this is an artefact of when conversions are recorded rather than a change in performance.
Conversions arrive after the click
Very few purchases complete in the same session as the advertising exposure. People return hours or weeks later, and the conversion is recorded then.
When a platform dates that conversion back to the original exposure, the figure for a past day continues rising for as long as the attribution window remains open.
Any report of a recent period is therefore reading a partial total that will grow after it has been circulated.
The distortion is largest where it is watched most
Attention concentrates on the most recent data, because that is where a problem would first appear and where intervention feels possible.
That is exactly the period with the least complete data, so the most scrutinised numbers are the least reliable ones in the report.
The effect scales with the length of the consideration cycle, so it is mild for impulse purchases and severe for anything involving research.
Reacting to it makes things worse
A campaign paused because a recent week looked poor is often paused while its conversions are still arriving.
The pause then reduces the following week's exposures, which reduces the following period's conversions, producing evidence that appears to confirm the original decision.
This is how an account can be optimised steadily downward while every individual decision is supported by the data available at the time.
Measuring the lag makes it manageable
Most platforms report time to conversion, which shows what proportion of conversions arrive on each subsequent day.
That profile is stable enough per campaign to be used as a maturity curve, indicating how complete a given period's data currently is.
Once the curve is known, recent figures can be adjusted for expected completion rather than compared directly against fully matured periods.
Reporting conventions that avoid the trap
The simplest fix is to exclude the most recent days from any trend comparison, using a cut-off matched to the typical conversion lag.
Marking incomplete periods explicitly in the report prevents the number being read as final by anyone who did not build it.
Longer comparison windows help as well, because the proportion of unmatured data shrinks as the period lengthens.