The same month can give you three different numbers, and all three are true.
I closed one August three times. Once on the 31st, once on the 1st, once on the 2nd. The impressions read higher every time. Nothing was wrong with the data and nobody made a mistake. Posts keep accruing. A piece published on the 12th is still picking up views on the 3rd of the following month, and every platform reports what it holds at the moment you ask.
That sounds like a technicality. It is not. It is the reason so many monthly marketing reports spend a paragraph explaining why this month's numbers disagree with numbers the client already has sitting in an email. That paragraph is the most expensive thing in the document. A client who watches a figure move twice stops believing the third one. You do not lose the account over it. You lose something quieter, which is the assumption that what you hand them is settled.
What actually causes it
Three things, and they compound.
Reporting lags. Platforms finish counting a day late, sometimes two. A number pulled on the last day of the month is a number for a month that is not over yet.
Old posts keep earning. Reach does not stop at the calendar boundary. On LinkedIn especially, a post that found its audience late can keep climbing for weeks after the month it belongs to.
Fresh pulls get compared to fresh pulls. This is the one that does the real damage. If you pull this month today and re-pull last month today, last month reads higher than it did when you reported it, and the growth rate you calculate is true of nothing. It is not last month against this month. It is two different afternoons against each other.
The rule I run now
A month closes on the 3rd of the following month. Not the 25th, not the 27th, not the last day. The month has to actually be over and the platform has to have finished counting, and the 3rd is the first honest day for both.
The pull taken on the 3rd is that month's figure of record. Permanently. It goes in the client's file and it does not get restated afterwards, however much the real number keeps climbing. A figure of record that can be revised is not a record. It is a draft in a confident font.
And prior months never get re-pulled. When this month is compared to last month, last month's figure comes out of the file, not off the platform.
Why the date matters more than it looks
The old version of this ran on the same schedule as the content batch, which fires on the 25th so the next month's work has an approval runway. Step one of that process said to close out the just ended month. On the 25th. Of that month.
Nobody was careless. The two jobs were stapled together because they were both monthly, and one of them was quietly asking for a number that could not exist yet. That is what most reporting problems turn out to be. Not sloppiness. A calendar nobody questioned.
So the jobs are separate now. Content runs on the 25th because it needs the runway. The close runs on the 3rd because that is when the month is real.
The build rule that comes with it
Reports get generated from a figures file, one per month, feeding one template. Numbers never get typed into a layout.
That sounds like a preference. It is the difference between a class of error being discouraged and being impossible. I once found a benchmark template still carrying a set of retired figures months after they had been withdrawn. They sat there looking finished, waiting to walk into a live client document. Nobody would have caught them by proofreading, because they looked exactly like numbers that belonged there.
Separate the data from the layout and a dead number has nowhere to hide.
What to do if you report for someone else
Pick your close date and write it down where the client can see it. Say plainly that the figure is taken on that day and does not move afterwards. Then hold to it, including in the month when holding to it makes your numbers look worse than a later pull would.
That last part is the whole thing. The value of a figure of record is not accuracy, because a later pull is arguably more accurate. The value is that it is fixed. Fixed numbers can be compared. Moving numbers can only be explained.
If a new pull genuinely disagrees with something already sent to a client, flag it as a correction, with its cause and its date. Never average two readings and never quietly replace one. A correction you volunteer costs a minute. A number that changes without explanation costs the relationship's assumption of care.
The wider point
Most of what looks like a marketing problem is a structure problem wearing marketing's clothes. The reports were not wrong. The people writing them were not careless. There was simply no rule about when a month was over, so every month it got decided again, slightly differently, by whoever happened to be closing it.
One sentence fixed it. That is usually the shape of it.
That is the gap I close.