- Why 14 events in 28 days falling to 8 is better read as "a gap of 6, or 0.21 a day" than as "down 43%"
- How wide the ordinary swing is: if your real rate stays at 10 per 28 days, you land at 8 or below about 33% of the time and at 14 or above about 14% of the time
- That GA4 applies data thresholds — "data may be withheld when the number of users or events in a given period is small" — so at low volume you cannot tell a drop from a suppression
Key questions, answered
- Q: My numbers dropped from last month. Did something break?
- A: At a few events a month, you cannot tell yet. Even if your true average holds at 10 per 28 days, counting alone puts you at 8 or below roughly a third of the time and at 14 or above roughly one time in seven. A move from 14 to 8 sits inside that ordinary swing. Go looking for a cause once several periods in a row point the same way.
- Q: What should I watch while the numbers are still small?
- A: Pick one metric and read it as a raw count over 28 days. Three conditions: your own work can move it, it fires at least once a week, and your own return visits cannot inflate it. Rates like conversion rate swing hardest when the denominator is small, so they do not get to be the headline yet.
This article is educational content about how to read your own numbers in the first months of running an affiliate site. Every analytics figure below is our own data from Kingfin's Japanese site (kingfin-jp.com), with the pull date and the period stated each time. It does not predict what your site will show, and nothing here guarantees that reading numbers this way will grow them. The statistics are offered as a guard against misreading noise, not as a forecast.
"Nothing again today" goes on for a while
You open analytics before the coffee is poured. Zero. Yesterday was zero too. The day before that there was one, so today counts as a decline.
For the first few months, that is mostly what the screen gives you. Whether the zero means you did something wrong or simply that nothing has happened yet, the screen will not say. What accumulates is not information. It is the time you spend looking.
We did exactly this. For weeks after launching Kingfin's Japanese site, the morning routine was: open GA4, compare with yesterday, confirm nothing moved, close it. It took a while to notice that the checking was eating the hours that were supposed to go into writing.
What follows is our own data, unedited, and the rules we now use to read numbers that never reach one a day. You cannot make a small number big by reading it differently. You can stop it from making your decisions for you.
In raw counts: 14 in 28 days
Here is what we have. GA4 on kingfin-jp.com, counting clicks out to the external signup page (outbound_kingfin), with the date we pulled each figure and the window it covers.
| Pulled | 28-day window | Sessions | Signup clicks | Per day |
|---|---|---|---|---|
| July 7, 2026 | Jun 9 – Jul 6 | 224 | 14 | 0.50 |
| Aug 11, 2026 | Jul 14 – Aug 10 | 215 | 8 | 0.29 |
| Aug 22, 2026 | Jul 25 – Aug 21 | 173 | 8 | 0.29 |
None of the three reaches one a day. Sessions fell from 224 to 173; signup clicks went from 14 to 8. Revenue across this stretch was zero. There is no version of this worth hiding.
Look at the middle row: 14 dropped to 8. The first sentence that came to mind at the time was "we're down 43%." Written that way, it sounds like something broke.
"Down 43%" and "a gap of 6" are the same sentence
Fourteen to eight is −42.9%. Close to a 40% collapse. The same change stated as a count is a gap of 6, which over 28 days is 0.21 a day.
Neither version is false. But they send you in opposite directions. "Down 43%" sends you hunting for a cause. "Six fewer over four weeks" lets you wait.
The smaller the denominator, the wilder the rate. Two to three is +50%. One to zero is −100%. One the following month reads as infinite growth. At this scale, every month is a crisis if you let percentages narrate it.
This cuts both ways. When you read "conversion rate improved by 40%" somewhere and no denominator is given, you may be looking at a single extra click. A rate without its denominator is not a number you can read — yours or anyone else's.
Your rate can hold steady and still print 7 through 13
One step further, because this is the part that stops the end-of-month slump.
Suppose your real rate is exactly 10 per 28 days and never changes. You still will not measure 10 every time. You do not control when people arrive, so the count itself carries spread.
Spread in rare, independent events is handled by the Poisson distribution. That is the only time the name appears here. What matters is the consequence: at an average of 10, the spread — the standard deviation — is the square root of 10, about 3.16.
- Measuring 8 or fewer … about 33% of the time
- Measuring 14 or more … about 14% of the time
- Landing between 8 and 14 … about 70% of the time
This assumes events arrive independently. Real traffic clusters by weekday and by when you post, so actual swings run wider than these, not narrower.
One period in three comes in at 8 or below. One in seven comes in at 14 or above. Our "14 → 8" was well inside the range you get from a rate that never moved.
The expensive mistake is assigning a cause to that swing. You change your writing in a month that happened to land at 8, then call it a win in a month that happened to land at 12. Nothing was tested, but you walk away with a conviction. A conviction built on noise is worse than no data, because it steers the next decision.
So how many periods before you act? A precise answer needs assumptions, but in practice three periods moving the same way is early enough. A single period's change is something to record, not something to explain.
Sometimes the number did not drop — it was withheld
There is a second thing that only happens at low volume: the data is not shown at all.
Google documents this as data thresholds. From [GA4] Data thresholds (Analytics Help):
Data may be withheld when the number of users or events in a given period is small.
Rows containing search-term information can likewise be removed when the total number of users is not high enough.
The same applies to reports that include demographic details and to audiences defined from them. While volume is low, a fall on screen and a row that was suppressed look identical.
Three things help:
- Widen the window. Read 28 days instead of 7, so fewer rows fall under the threshold
- Stop stacking filters. Country by device by source shrinks every row's user count
- If you truly need raw rows, use the BigQuery export — but note Google's own caveat that data derived from Google signals is not included in that export, so it will not match the interface
The third one is usually a problem for later. Early on, the first two are enough.
Four ways to read a small number
Turned into things you can change tomorrow.
| What people do | What to do instead |
|---|---|
| Open the dashboard every morning | Pick a weekday and open it once |
| Read the last 7 days | Read 28 days; treat period-over-period as a footnote |
| Ride the conversion rate up and down | Stack raw counts month by month |
| Keep five metrics in view | Choose one and ignore the rest |
Stacking counts sounds dull and works well. Once the months sit side by side as bars, your eye stops asking about any single bar and starts asking whether the shape leans right. The month-over-month percentage quietly drops out of view.
Choose exactly one metric
Ours is outbound_kingfin — clicks leaving for the external signup page. Not signups.
We did not make signups the headline metric for a simple reason: the count gets smaller and stops being readable. If clicks run at 8 a month, whatever happens after is fewer. In a world of zero or one, there is no up or down to read.
Three conditions for picking yours:
- Your own work can move it. Writing, changing a path, posting. A metric disconnected from today's work will not change what your hands do
- It fires at least weekly. Anything that shows up once or twice a month swings too hard to read, no matter what you try
- You cannot inflate it. If your own return visits or your own clicks register, the number grows every time you go check on it
Once one metric satisfies all three, you can leave the others alone for a long while. Deciding not to look is itself what gives the hours back.
One thing to do today
Pick the weekday you will open analytics, and put it in the calendar. That is the whole instruction.
For us, the writing time came back after that. The numbers still do not reach one a day. But we stopped panicking in the months that land at 8, and the hands went back to the work. Whatever growth comes, it comes after that.
If the screen in front of you says zero right now, that is not today's score. It is a count that has not had enough chances yet. Set the day you will next look, and close it.
Frequently asked questions
[Disclaimer] This article is educational content from the Kingfin English Editorial Team. It does not solicit any investment or service. The analytics figures are our own data from kingfin-jp.com, with pull dates and periods stated, and are not an indication of what any other site will show. The statistical figures assume events occur independently; real data clusters by weekday, season and campaign, so the stated probabilities are approximations. Statements about Google Analytics reflect Google's published help pages as of writing and are subject to change. OlympTrade, promoted through Kingfin's affiliate programme, is an FX and binary-options trading service and is not registered as a financial instruments business operator in Japan. Trading always carries the risk of losing your capital, and no result — including affiliate earnings — is guaranteed. Invest only funds you can afford to lose, at your own discretion and risk.
