SegmentSage answers

When a Segment Quietly Doubles: A Drift Investigation in Four Checks

A segment that doubles in a week is not a marketing insight. It is a symptom. Segments are queries over data, and when the count lurches, something underneath moved: a feed changed format, an event started firing twice, a definition was edited in a tool nobody thought of as production, or the business genuinely changed and the segment is the first place it showed. Teams that celebrate or panic before distinguishing those causes waste the week. There is a short order of checks that separates a data incident from a market signal, and it starts with the segment's own history rather than with theories.

Check the definition before the data

Open the segment definition and its change history first. Look for edits in the window where the count moved: an operator broadening an OR to an AND, a property swapped for a similarly named one, a time window widened during a campaign build and never restored. In tools without change history, compare the current definition against the last exported or screenshotted version, and resolve, today, to export definitions on a schedule from now on. A surprising share of drift investigations end here, in five minutes, with a definition that no longer says what its name claims.

Then check the inputs, event by event

If the definition is untouched, profile the inputs. For each event or attribute the segment depends on, plot daily volume for the last ninety days and look for the step change: an event doubling on a release date points at duplicate firing, a property arriving in a new format, uppercase where the rule expects lowercase, silently fails matches in one direction and floods them in the other. Talk to whoever shipped in that window. Data changes that break segments are rarely announced as data changes. They arrive as an app release, a tag manager publish, or a warehouse model refactor with a changelog that never mentioned your segment by name.

Separate new members from returning ghosts

Break the movement into who entered and who failed to leave. Growth from new members, with normal exit rates, suggests an upstream volume or matching change. Growth because exits stopped, a churn or lapse segment that no longer lapses anyone, points at a broken date comparison or a frozen attribute, and it is the more dangerous pattern because the segment keeps mailing people it exists to stop mailing. Sample twenty members by hand and verify each against source records. Twenty checked records beat any dashboard argument about what the count ought to be.

Write the incident into the segment's record

Whatever the cause, close the loop in writing: the dates affected, the members wrongly included or excluded, the campaigns that used the segment in the window, and the fix. Decide explicitly whether sends made on the corrupted segment need a follow up, an apology in rare bad cases, usually just a note in the campaign record. Then add a size alert: segments above a materiality threshold get an automatic flag when their count moves more than a set percentage week on week. Drift will happen again. The goal is to hear about it from your own monitoring before a stakeholder hears about it from a customer.

Keep reading

Frequently asked questions

Why would a segment suddenly double in size?

Common causes are an edited segment definition, a duplicated or reformatted event feed, a matching rule broken by a data format change, or members no longer exiting. Genuine market change is possible but should be the last hypothesis checked.

What is the first check in a drift investigation?

The segment definition and its change history. Definition edits during campaign work are a frequent and quick to confirm cause.

Should I pause campaigns using a drifting segment?

For segments with a compliance or suppression purpose, yes, until membership is verified. For ordinary marketing segments, pause if sampled members do not belong.