Traffic went up 20%. Is that real, or just a normal week?

You open your analytics on Monday morning and traffic is up 20% on last week. Should you celebrate, investigate, or ignore it? Most of the time the honest answer is "it depends on how much your traffic normally moves", and that can be measured.

This guide explains how to tell a real change from noise without a statistics degree.

Compare like with like

The most common mistake is comparing today with yesterday. Most websites have a strong weekly rhythm: a business site is busy on Tuesday and quiet on Saturday, a recipe site is the other way round. Comparing Saturday with Friday mostly measures the weekend.

A better baseline compares a day with the same weekday in previous weeks. If the last eight Mondays averaged 1,000 sessions, that average is what a normal Monday looks like.

Know how much "normal" varies

Even with the right baseline, numbers wander. If the last eight Mondays were 960, 1,030, 990, 1,010, 970, 1,040, 1,000 and 1,000, a Monday with 1,050 is unremarkable. If they were all within a few sessions of 1,000, the same 1,050 is surprising.

There are two ways to estimate how much variation to expect:

1. From your own history. Look at how far the same weekday usually strays from its average. The standard deviation of those eight Mondays is a good measure.
2. From the size of the numbers. Counts of independent events vary roughly by the square root of their size. A page with 100 visits a day will easily swing by 10 to 20 from chance alone, while a site with 10,000 visits a day should rarely move by more than a couple of hundred without a reason.

A simple significance check

Put the two together into a z-score: how many "normal variations" away from the baseline today's number is.

z = (today − baseline) / typical variation

Using the square-root rule for the typical variation, a Monday with 1,284 sessions against a baseline of 1,002 gives:

z = (1,284 − 1,002) / √1,002 ≈ 282 / 31.7 ≈ 8.9

As a rule of thumb:

The square-root rule assumes visits are independent, which is not quite true in practice: one newsletter or one viral post moves many visits at once. It therefore tends to overstate certainty. Using your own history is safer: if your Mondays typically stray about 67 sessions from their average, the same day gives z = 282 / 67 ≈ 4.2. That is a smaller number, but still well above 3, so the change is real.

Small numbers need extra care

Conversion rates are where most false alarms come from. If a page converted 3 visitors out of 40 last week and 1 out of 38 this week, the rate fell from 7.5% to 2.6%, which looks dramatic. With numbers that small it is almost certainly noise. Before you react to a change in a rate, check how many visitors and conversions sit behind it. As a rough guide, be wary of any rate built on fewer than a few dozen conversions.

Then ask what drove it

A real change always has a cause, and the fastest way to find it is to break the total down:

1. By source. Did one referrer, campaign or AI assistant account for most of the difference?
2. By landing page. Did one page suddenly get far more entrances?
3. By device or country. A drop limited to mobile, or to one country, usually points at a technical problem rather than demand.

If the extra 282 sessions all came from one source landing on one page, you have your explanation and probably your next action.

How we automate this

This is exactly the routine our analytics runs for you every day. Each metric is compared with its weekday baseline, changes are only reported when they pass a significance test, the main driver is identified by breaking the change down by source, page, device and country, and the result is written up as a short brief with a recommended next step. Every number in that brief is checked against the underlying data before you see it.

See how a brief is written or start a free trial to get one for your own site.