Where to set the serial returner threshold

Set the threshold on behavior over time, not a single return rate. A customer with twelve orders and four returns is a different case from one with two orders and two returns, and your policy should treat them differently.

Return rate without context misfires. New customers naturally test fit, and a strict rate threshold flags them right when they are deciding whether to trust your brand. Thresholds that ignore order history punish exploration. Weight the rate by tenure: a 30 percent return rate across fifteen orders means something very different from the same rate across three.

Weight the signals that indicate intent, not just volume. Wardrobing markers carry more meaning than raw counts: items returned worn, tags removed, returns timed right after events, the same category returned repeatedly. A customer returning five dresses the week after wedding season is telling you something that a return count alone never will.

Warn before you deny. A warning converts a meaningful share of serial returners into profitable customers, because many simply did not know the pattern was visible. Denial is the last step, not the first. Reserve it for the accounts that kept going after a clear warning, and document the pattern so the decision holds up if questioned.

Thresholds also need a decay function. A customer who bracketed heavily two years ago and has ordered cleanly since is not the same risk as one doing it now. Most scoring systems treat old behavior as permanent, which fills the warn list with people who already corrected. Let stale signals fade. It keeps the enforcement list short, current, and defensible, and it gives reformed customers a reason to stay.