Sales were down yesterday. Down again today. Is that a problem, or just a rough patch?

Daily numbers always wobble. The skill is knowing how much wobble is normal for your business. A drop inside the normal range needs no action. A drop outside it does.

Our position: do not react to one bad day. Learn your normal range first, then act only on a drop that falls outside it, or stays low.

The answer first

A drop is normal if it stays inside the range your own past days set. A drop is real if it falls outside that range, or if sales stay below the path your trend predicted for many days in a row.

Step 1: find your normal range

Take a few weeks of daily sales from a calm stretch. Work out the usual level and how far days stray from it. A control chart draws this as a middle line with a band around it. Days inside the band are ordinary. The free control chart tool draws it from your file.

Step 2: ask if the drop broke the pattern

Two signs say a drop is not just luck: one day far outside the band, or a long run of days on the low side. A single dip inside the band is just a dip.

Step 3: if it is real, ask how big

When the drop looks real, work out how much lower sales are than the path from before the drop predicts, with a range. That is the same causal impact check used for a sale. See Did my sale work?

A real example from our own data

We rebuilt our top article on October 23, 2026. We asked: did daily search clicks to it change after the rebuild, or is the difference noise? The data was 81 days of our own Search Console clicks to that page.

The check said the drop was real, not noise. Clicks ran about 9.9 a day below the path the page was on before the change (95% range 4.9 to 14.9 fewer a day, p = 0.0002). Over the 17 days since, that adds up to about 168 clicks. The method was interrupted time series by segmented regression, with the day to day link between days taken into account.

For scale, the plain averages were 19.3 clicks a day in the 22 days before and 13.6 a day in the 17 days after.

What we cannot rule out. This is one page and 17 days. Other site changes shipped the same week, and the page's average position slipped from about 3.2 to about 3.8. The season is not modelled, because we have two months of history and would need about two years. So the honest sentence is: clicks fell after the rebuild by more than chance explains. It is not proof that the rebuild alone caused it.

The whole check took about two seconds. When more days arrive, we swap in the new rows and run it again.

What to do when a drop is real

  • List every change in the days before it: prices, ads, site edits, stock-outs, an outage.
  • Check the other numbers near it, such as visits and the share who buy.
  • Run the check again as new days arrive. A real drop stays; a fluke fades.

Run this on your GA4 or Shopify data

From Shopify: open Analytics, then Reports, then Sales over time. Set the dates to cover at least a few weeks before the change. Choose daily. Export as CSV.

From GA4: open Explore and make a table with date as the rows and purchase revenue (or sessions, or whatever you are judging) as the value. Export it as CSV.

What your file needs: one row per day, one column for the date, one column for the number. Add a column for the thing that changed if you have one, such as the day the sale began.

What you get back: your normal range, the days outside it, and, for a real drop, the size of the gap with its range. Use the free control chart tool for the range and the causal impact tool for the size. For more on this kind of check, see interrupted time series.

Questions owners ask

How many days of history do I need?

A few weeks is a start. More is better, and you need a full year or two to account for the season.

Is one bad day ever a real drop?

Yes, if it falls far outside your normal range, such as a site outage. Most single bad days are inside the range.

What if my sales are seasonal?

Then last month may not be a fair base. Compare to the same season in earlier years, or see the page on growth and seasons.

Why quote clicks a day and not a percent?

A percent depends on the line you compare against, and that line moves. Clicks a day, with a range, is the plain number.