Glossary

What is a conversion anomaly?

A conversion anomaly is a conversion metric moving outside its expected range for that funnel and time period — the statistical signal that usually starts a conversion incident.

Definition

A conversion anomaly is a conversion metric moving outside the range that is normal for that funnel, that day of the week, and that traffic mix. It is a detection signal, not yet a diagnosis.

An anomaly says something unusual is happening. A conversion incident is what you open once the movement is real and worth investigating.

Why it matters

  • Fixed thresholds either miss real drops on small funnels or fire constantly on volatile ones.
  • Comparing against a funnel's own baseline accounts for weekday patterns and normal variance.
  • Good anomaly detection is what makes alerts trustworthy enough that people still read them in month three.

How to detect it

  1. 1Build a baseline per funnel and per stage from recent comparable periods.
  2. 2Require both a meaningful deviation and a sustained one before alerting.
  3. 3Account for volume: small samples swing widely and produce false positives.
  4. 4Suppress known causes — a paused campaign is expected, not anomalous.

Example

Illustrative scenario: a funnel that normally converts between 9% and 12% on weekdays sits at 6% for four consecutive hours. That is outside its range and sustained — an anomaly worth opening as an incident.

How Quarkad helps

Quarkad evaluates funnel metrics against each funnel's own baseline and raises an incident when a movement is both meaningful and sustained, rather than alerting on every fluctuation.

See how conversion incident intelligence works →

Related terms and pages

Frequently asked questions

Is every anomaly a problem?
No. Some are expected — a paused campaign, a holiday, a deliberate form change. Anomaly detection flags the movement; context decides whether it matters.
How much of a drop should trigger an alert?
It depends on the funnel's normal variance and volume. A 20% move on a high-volume funnel can be significant while the same move on a low-volume one is noise.

Find the change behind the drop.

Quarkad monitors your funnels and the changes around them, then tells you which change lines up with the drop.