The number that changes
Why the number changed after the monthly close
The number changed for one of three reasons: the source reprocessed data after you read it, nobody saved the result of the closed period, or the calculation’s definition changed midway. You can tell the three apart in an afternoon. None of them gets fixed by arguing about the number in the meeting.
Updated August 5, 2026 · 2 min read
Reason 1: the source reprocesses data after the fact
Analytics platforms don’t publish a final number on the same day. They keep receiving late-arriving events, they recalculate attribution when the missing touchpoint shows up, and on large properties they apply sampling with thresholds that change with the date range you query. The result: the same report, with the same date filter, returns different values depending on the day you open it.
This isn’t a system error. It’s how the system works. The error is treating a live reading as if it were a close. If you presented the month on the 3rd, you presented a snapshot of data that was still moving.
How to check: take the same period and the same report, and read it on three different days, noting the value and the date you read it. If the number moves and then settles, you’ve found the day your source finishes closing. That day, not the 1st, is your real close date.
Reason 2: the period wasn’t frozen
Most teams don’t save the close. They save the link to the report. When someone opens it again, the tool recalculates everything against the current state of the data, and the number you see is no longer the one you presented. There’s no record of what was said, or what it was compared against.
Freezing a period means saving three things together: the result, the exact date range, and the rules used to calculate it. With that, any later difference gets explained instead of argued. And when a legitimate correction comes in, it goes in as a note on the closed period, not as a silent replacement of the number.
A frozen period can’t be rewritten. What was presented is what stands.
Reason 3: the definition changed and nobody wrote it down
Someone fixed the purchase event that was firing twice. Someone added an internal traffic filter. Someone changed what counts as a conversion. Each of those changes is correct on its own, and none of them was written down anywhere. March’s number calculated with April’s rules is not March’s number.
The telltale sign here is that the number changes all at once, in a jump, instead of drifting gradually. And it usually affects several earlier periods at the same time, not just the latest one.
For this to be diagnosable, the calculation rules have to be versioned: each rule with its number, its effective date, and the periods it calculated. Without that, the only way to investigate is the memory of whoever made the change.
How to tell which of the three happened to you
What to do this week, without buying anything
- Set a close date based on evidence. Measure when your source settles and close from that day on. Not the 1st, and not the 5th just because it’s always been the 5th.
- Save the close as a file, not a link. Export the result with the date range and the read date in the file name. It’s crude, but it already lets you argue with evidence.
- Start a log of definition changes. One line per change: what was touched, by whom, when, and from what date it applies. A spreadsheet is enough to start.
- Declare the numerator. Write down what counts as a sale and what doesn’t. If two teams use different definitions, that’s the problem, not the number.
What a system has to do so it doesn’t happen again
The four steps above hold up by hand for two or three months, and then someone forgets. What makes them permanent is a system that does three things without depending on anyone’s discipline:
That’s what NORQ Revenue does: it labels which period each number refers to, freezes what was closed, and alerts you when something moved. If your whole operation also runs on cross-referenced spreadsheets, the problem is bigger than the close, and it starts with an operations diagnostic.
What doesn’t work
Switching analytics tools doesn’t fix anything: the new one will reprocess too. Adding another dashboard doesn’t either: it multiplies the versions of the number instead of resolving them. And asking someone to check every close by hand works until the month that person is on leave.
Frequently asked questions
Does Google Analytics change data from previous months?
Yes. Late-arriving events come in, attribution is recalculated, and sampling thresholds are applied. A report read live on the 3rd is not the same report read on the 20th.
Can you freeze a report so it doesn’t change?
Yes, but not inside the analytics tool. You freeze it by saving the period’s result together with the rules used to calculate it. Any later correction goes in as a note, not as a replacement.
How long should you wait to close a month?
It depends on when your source settles, and that can be measured: read the same period on different days until it stops moving. That day is your close date.
Yesterday’s data looks different today. Is it broken?
The last few days of a series almost always move, and that’s expected. The problem starts when a period already presented as closed moves, or when nobody can explain why it moved.