Why your ad platform keeps rewriting last week's numbers

You screenshot Tuesday’s performance on Wednesday morning. You look again on Friday and Tuesday is better. Nobody changed anything. Nobody fixed a tag.

This is not a bug, and it is not your tracking. It is how ad platforms are designed to report, and once you can see the mechanism the whole class of “our numbers don’t match” arguments becomes much shorter.

The platform is writing conversions backward

When someone clicks an ad on Tuesday and buys the following Sunday, there are two defensible days to record that sale against. Your analytics tool and your accounting system both say Sunday, because that is when it happened. The ad platform says Tuesday, because that is when the click it is taking credit for occurred.

Google states it plainly: the primary conversion columns are calculated based on the time of the click, not the time of the conversion. Their own example — an ad clicked last week producing a conversion this week gets reported back to last week. Meta does the same thing against the impression date.

So Tuesday is not a finished number. Tuesday is a number that will keep going up for as long as the attribution window stays open.

What this actually does to you

Recent days always look worse than older ones. Yesterday has had one day to accumulate conversions; last month has had a month. Comparing this week to last week while this week is still filling in makes performance look like it is declining when nothing has changed. This is the single most common false alarm in paid media reporting.

Your dashboard disagrees with the platform, permanently. Not because one is wrong — because they are answering different questions on different days. Reconciling them day-by-day is not a solvable problem, and teams lose a lot of hours to trying.

The window sets how long the past stays liquid. A 1-day window means yesterday is basically settled. A 90-day window means a day from three months ago can still move. Most people never look at this setting.

Pausing a campaign hides its tail. Turn something off on Friday and conversions from clicks it already bought keep landing for days, getting written back onto days when the campaign was still running. The campaign looks better after you kill it.

The defaults nobody set

Every account is running these numbers, and almost nobody chose them. Read from each platform’s own documentation:

Platform	Click window	Range	View window	Reports against
Google Ads	30 days	1–90 days	1 day	Time of the click
Meta	7 days	fixed options only	1 day	Impression date
Microsoft	30 days	1 min – 90 days	1 day	—

Three things fall out of that table.

The click defaults differ by more than four times. A conversion eight days after the click is credited by Google and invisible to Meta, from the same customer doing the same thing. Any cross-platform comparison of “conversions” is partly a comparison of these settings.

View windows are a thirtieth of click windows. Both major platforms default view-through to one day against a click window of thirty. That is a deliberate and sensible gap — a view is much weaker evidence — but it means the two numbers in your report were assembled under wildly different standards.

Meta’s options are a short list, not a range. You pick from four combinations. Google and Microsoft let you set nearly any duration. So “make the windows match across platforms” is not fully achievable, only approachable.

The one that surprises people

Inside Meta, two products disagree with each other. Ads Manager uses whatever attribution setting you chose and reports on impression date. Commerce Manager’s tagged-content reporting uses a fixed 28-day click and 1-day view window, cannot be changed, and ignores your Ads Manager setting entirely — and it reports on conversion date instead.

Same company, same conversion, two different windows and two different days. If you have ever had a Commerce Manager figure refuse to reconcile with Ads Manager, that is why, and no amount of tag debugging will close it.

What to do about it

Stop comparing incomplete days. Give recent days time to fill before judging them — at minimum the length of your click window. A weekly view is more honest than a daily one for exactly this reason.

Write the window on the report. A conversion figure without its window and its reporting basis is not comparable to anything. This costs one line and prevents most of the arguments.

Use conversion-time columns when reconciling. Google provides columns reported by conversion time rather than click time. When you are trying to match the platform against your own analytics or your finance numbers, those are the ones that can line up.

Set the window deliberately, once. Look at how long your customers actually take to buy, set the window to match, and write down why. Then leave it alone — changing it does not restate history, so every change puts a step in your trend line that is not a result.

None of this makes the numbers agree. It makes the disagreement predictable, which is the achievable version. The remaining gap between what a platform credits itself with and what your advertising actually caused is a different question entirely — that one is incrementality, and no reporting setting answers it.