The Date-Range Trap: Why Your Pinterest Numbers Change While You Watch

I once wrote somebody a two-page recovery plan for a crash that had not happened.

She had sent me a screenshot on the Tuesday — impressions down by about a third since Sunday, no warning, nothing changed — and I spent most of an evening on her boards, her descriptions and her publishing cadence before I thought to ask which date range she had been looking at.

Last thirty days. The same range she had checked every morning for a year.

Nothing had crashed. One very good day in the middle of July had reached its thirtieth birthday and dropped out the back of the window, taking about eighteen thousand impressions with it. The account was doing on the Tuesday exactly what it had been doing on the Monday. The number describing it had moved, because that number was never describing a fixed thing in the first place.

This is by a wide margin the most common false alarm in Pinterest analytics, and it took me an embarrassingly long time to start checking for it first rather than fourth.

A rolling thirty days is not last month

Almost every default in Pinterest analytics is a rolling window. The last thirty days means the thirty days ending today — so tomorrow it will mean a different thirty days, with one new day added at the front and one old day discarded off the back.

A calendar month is a fixed box. July is July forever. It does not change depending on when you look at it, and two people looking at it get the same answer.

Those are two different measurements wearing the same units, and nothing in the interface warns you when you cross from one to the other. That is the whole trap in a sentence: a rolling range answers “how have the last thirty days been”, and a calendar month answers “how was July”. People ask the second question and read the first number.

The monthly views figure on your profile is a rolling window as well, which is why it lurches for reasons that have nothing whatever to do with your work. If any of the words in your dashboard still feel slippery, all of those numbers translated into English is the companion to this one.

There is one more wrinkle here that catches people who are otherwise being careful: reach does not add up across ranges. Impressions do — a ninety-day impression total genuinely is three thirty-day totals stacked on each other. Reach counts separate accounts, so somebody who saw you in June and again in August is one reach across ninety days and one reach in each of two thirty-day windows. Add those windows together and you have counted her twice. The difference between impressions and reach matters more here than anywhere else in the product.

One enormous day is a landmine with a thirty-day fuse

Here is the mechanic with numbers on it, because it is much less mysterious once you have seen the arithmetic.

Say your account runs at roughly twelve hundred impressions a day. Over thirty days that is somewhere near thirty-six thousand, and it sits there, give or take a slow weekend.

Then one pin gets picked up and does eighteen thousand impressions in a single afternoon. Your thirty-day total is now fifty-four thousand, and you feel rather good about yourself.

Nothing else changes. The publishing is identical, the pin still exists, the account is entirely healthy. But that total sits at fifty-four thousand for a month, and then on day thirty-one the spike falls out of the window and the total drops to thirty-six thousand overnight — a third of it gone, with a perfectly plausible-looking cliff on the chart to go with it.

If you happen to check on that day, you will look at that cliff and conclude something is broken. Nothing is broken. You are standing at the precise moment a good memory expired.

Four tells separate a window artefact from an actual decline, and you can check all of them in about two minutes:

• The drop arrived in one step rather than over several days. Real declines in distribution are a slope, not a cliff.
• It is roughly the size of a very good day you remember having about a month ago. That is not a coincidence.
• Your daily chart looks unchanged for the last fortnight — only the headline total moved.
• Saves and outbound clicks did not fall with it. A genuine distribution problem drags those along behind it.

If three of those four are true, close the tab and go and do something more useful. If none of them are, you may well have a real problem, and the order to work through an actual traffic drop is worth following properly rather than guessing at.

The same data, two ranges, two opposite conclusions

What makes this genuinely dangerous rather than merely annoying is that you can prove almost anything about an account by choosing a range — and you will do it accidentally, in good faith, whilst trying to be rigorous.

I have watched somebody talk themselves out of a strategy that was working, and somebody else double down on one that was not, in the same week, using the same dashboard.

Same account, same pins, same effort. Read on the last seven days it looks like a collapse. Read on ninety it looks like a steady climb with a bumpy patch in the middle. Neither reading is dishonest. Only one of them is any use for deciding what to do on Monday.

The rule that fixes it is unglamorous. Compare like with like, and never compare a rolling range to a calendar one. If you looked at a rolling thirty days in April, look at a rolling thirty days in May. If you reported on the month of April, report on the month of May. Mixing the two produces a change that is purely an artefact of the frame, and it will send you off rewriting descriptions that were fine.

Why last-7-days is nearly useless for judging a pin

I do not think seven days is a bad range. I think it is a bad range for the question people habitually point it at.

Pinterest is slow. A pin can sit almost untouched for a fortnight and then start moving, and a good share of what any pin will ever do arrives long after the week you published it in. Judging a new pin on seven days is like judging a plant by whether it grew today.

The second problem is that a week contains weekends, and which weekends depends on where you start it. Seven days ending on a Sunday and seven days ending on a Wednesday are not the same shape of week at all, and on a small account that difference alone can move a total by a fifth.

The third is plain arithmetic. Seven days on a small account is a small pile of events, and small piles bounce. One pin catching a little distribution rewrites the entire chart and tells you nothing about the account underneath it.

So I use seven days for exactly one thing: confirming nothing has gone catastrophically wrong. A broken link, a site that quietly unclaimed itself, a format that has stopped serving. It is a smoke alarm, not a thermometer, and I never ask it whether an idea was any good.

Ninety days is where Pinterest work actually becomes visible

If you want to know whether what you are doing is working, ninety days is the honest window — and I would far rather you looked at it once a month than looked at thirty days every morning.

Three months is long enough that no single spike can dominate it. It is long enough to contain a full cycle of slow pins catching on. And it is long enough that a real change of direction shows up as a direction rather than as noise.

It is also long enough to be uncomfortable, which is rather the point. A quarter of work that produced nothing is information you can act on. A week of work that produced nothing is a Tuesday.

One caveat sits on top of all of this: clicks and conversions carry their own lag, because somebody can save a pin in March and buy the thing in May. How the attribution windows work is a separate piece of the same puzzle, and it is worth knowing which window a reported number is using before you line two months of them up next to each other.

The four ranges I use, and what each one is for

This is the entire system and there is not much to it. The value is not in the choices — it is in never changing them.

1. Is anything on fire? Last seven days, once a week. I am looking for a cliff in outbound clicks, not a verdict on any individual pin.
2. Is this pin working? Thirty days from the day it was published, not thirty days from today. A pin deserves its own clock, because the account clock has no idea when it went out.
3. Is the account growing? Ninety days, once a month, compared with the previous ninety rather than with a feeling I had in June.
4. What should I make next? None of the above. That question is answered by demand and by what people are already saving, and no date range on earth will tell you.

The fourth is the one nobody says out loud, and it is the only one of the four that changes what you do rather than how you feel about what you already did. If you want the argument for which handful of numbers deserve to drive a decision at all, the case for the three that matter makes it better than I can here.

Two habits that removed nearly all of my false alarms

The first is that I write the range down beside the number. Every single time. If a figure appears in a document, a message or an email without its exact window attached, it will eventually be compared with something incompatible — and in my experience the person doing the comparing is usually me, three weeks later, in a hurry.

The second is that I read the daily trend line before the headline total. The total is a sum over a moving frame and can move for reasons that have nothing to do with reality. The daily line cannot mislead you the same way: if the last fortnight of bars looks like the fortnight before it, nothing has happened, whatever the enormous number at the top of the page says.

And I ignore the last two or three days of any chart, because Pinterest data settles rather than arrives. The dip at the right-hand edge is usually the data still filling in, and I have watched people panic about that specific dip on a more or less weekly basis. If the dashboard itself is still unfamiliar, the tab-by-tab walkthrough is the place to start before any of this.

When the drop is real

Sometimes it is real, and I would hate to talk somebody out of a genuine problem on the strength of a nice story about calendars. The pattern looks different, and it is not subtle once you know the shape:

• The daily line steps down and stays down for more than a fortnight.
• Saves and outbound clicks fall in step with impressions rather than holding up behind them.
• It is account-wide rather than confined to one board, one topic or one pin format.
• It survives a like-for-like comparison — same window length, same days of the week, previous period.

That combination is worth investigating, and it usually starts with distribution rather than design. Why pins stop getting impressions covers the likely causes roughly in the order they are worth checking.

What I would never do again is act on a single reading of a rolling window. The number was built to move, and it moves whether or not you have done anything to deserve it.

The reason I care about any of this is that most of the hours lost to Pinterest analytics are lost right here — staring at a total that shifted for structural reasons and trying to work out what you did wrong. Those hours would be far better spent on the one question a dashboard cannot answer, which is what to make next week. Working that out from what people are already searching for and already saving is the entire job PinPinterest does, and it is a much better evening than auditing a chart that was never accusing you of anything.

My client sent me another screenshot about four weeks later. Second cliff, same size, same shape — the recovery day itself ageing out of the window. She had worked it out before she messaged and was mostly writing to tell me so, which I have decided to count as the best result I have ever had from a two-page recovery plan nobody needed.