Anomaly Watch - Growth Ops Console

Loading demo scan

This view runs the same z-score detection as production, but on synthetic Northwind Cloud metrics. In the shipped system, GA4 fed these charts daily.

Watch summary ·

Anomaly feed · most severe first

Flagged days

Ranked by absolute z-score. Critical = |z| ≥ 3.5 · Warning = |z| ≥ 2.5 · Notice = |z| ≥ 2.0

Per-metric trends · anomaly markers

Top pages · last 7 days vs prior 7 days

Which URLs are driving the numbers

Top 10 pages by views. The change column compares the last 7 days against the prior 7 days - anything red is well above normal, anything muted is well below. This is where to look when something's been flagged above.

Top countries · last 7 days vs prior 7 days

Where the traffic is coming from

Top 10 countries by sessions. Sudden traffic from an unexpected region is often a bot tell - and seeing a single country dominate a spike usually answers "is this real?"

The math · how each anomaly gets determined

The formula

z=(x−μ)÷σ

x = yesterday's value for this metric
μ = the mean (average) of the previous 14 days
σ = the standard deviation of those 14 days - how much the metric usually wiggles
z = how many standard deviations yesterday is from the recent norm

When we flag it

\|z\| ≥ 2.0	Notice	●
\|z\| ≥ 2.5	Warning	●
\|z\| ≥ 3.5	Critical	●

Sign of z tells direction: positive = spike, negative = drop. The sensitivity dropdown in the header sets the threshold.

Worked example · "Sessions" on May 14

Yesterday's sessionsx = 603,562

Mean of prior 14 daysμ = 132,191

Std dev of prior 14 daysσ = 116,243

Compute:(603,562 − 132,191) ÷ 116,243

Result:z = +4.05σ

|z| ≥ 3.5 →CRITICAL spike

Translated: yesterday's number was four standard deviations above the typical range. In statistical terms that's roughly a 1-in-15,000 event under normal conditions - so something really changed. The dashboard flags it without ever needing to read the actual chart.

How this works

The idea, in plain English

Every day, this dashboard compares yesterday's number to what we'd expect based on the previous two weeks. If the number is much higher or much lower than expected, it gets flagged. Same instinct you'd use eyeballing a chart - "that day looks weird" - except the system runs it across every metric, every day, without missing.

Why a two-week comparison window?

Marketing traffic naturally rises and falls by day of the week - Mondays look different from Sundays. Two weeks captures both weekdays and weekends, so the comparison adapts to those patterns without overreacting to recent campaigns or seasonal shifts. It's long enough to be stable, short enough to stay current.

What the severity tiers mean

Critical · a "once in years" kind of move · investigate immediately

Warning · a "few times a year" move · worth a quick check

Notice · a "monthly-ish" move · outside normal but not alarming

What this catches

Traffic spikes from campaigns or viral pages, traffic drops from tracking breaks or site outages, unusual conversion patterns, sudden geographic shifts - anything far outside the recent norm. Especially useful as a safety net the week of a launch or a redesign.

What this doesn't catch

Gradual trends that build over several weeks - those get absorbed into the rolling baseline by design. Issues that happen only at certain hours (the system checks once a day, not hourly). Problems in segments that aren't on the watchlist, such as a specific channel cratering while overall sessions hold steady. For those, pair this view with the channel and page reports below - or open a focused investigation on the day flagged here.