Keep-rate scoring: ranking customers by return behavior

September 25, 2026

Every apparel brand has a returns dashboard. Very few have a per-customer returns score. The dashboard tells you that returns cost 18 percent of revenue. The score tells you which customers are responsible, and that is the difference between a metric you watch and a lever you can pull.

A keep-rate score is exactly what it sounds like: the share of units a customer keeps, tracked per person over time. A customer who orders ten units and keeps nine has a keep rate of 90 percent. A customer who orders ten and keeps two has a keep rate of 20 percent. One number, updated with every order, tells your CX team who is cheap to serve and who is expensive.

How to build the score

Start with trailing twelve months of order and return data. For each customer, divide kept units by ordered units. That is the raw score. Then apply three refinements that make it usable in practice.

First, weight recent behavior more heavily. A customer who bracketed heavily last spring but buys cleanly now should not carry last year's score forever. A simple decay, like weighting the last 90 days at half the total, keeps the score honest about who the customer is today.

Second, set a minimum order count before the score counts. A first-time buyer who returns their only order has a keep rate of zero, but that is noise, not signal. Most brands find that three or more orders is the point where the score starts meaning something. Below that, treat the customer as unscored.

Third, segment by category before comparing. A denim buyer with a 60 percent keep rate might be your median customer in a fit-difficult category, while a 60 percent keep rate in accessories is a red flag. Compare customers to their category peers, not to a single global average.

What thresholds to set

Thresholds turn a score into decisions. A common starting pattern: above 75 percent is green, 50 to 75 percent is yellow, below 50 percent is red. Green customers get instant approvals and the friendliest policy. Yellow customers get approvals with the occasional nudge, like a fit-guide prompt before their next order. Red customers get human review on returns and, for the persistent few, a warning.

Resist the urge to set thresholds from industry benchmarks alone. Your margin structure decides what a red customer costs you. A brand with high average order value and low return shipping cost can tolerate lower keep rates than a brand selling thirty-dollar tees. Run the unit economics: cost per return times return frequency, minus the customer's lifetime margin. The threshold is where that math turns negative.

How to act on it without banning good buyers

The score's job is routing, not punishment. Green customers should never feel the system at all; their returns approve instantly and their experience stays generous. The whole point of scoring is that you can afford to be generous with the honest majority precisely because you are strict with the expensive few.

Yellow customers respond best to fit interventions, not policy threats. Show them size guidance tuned to what they actually kept. Suggest the cut that fits their history. Many yellow customers are not abusers at all; they are loyal buyers in a category your size chart describes badly. Fixing the information fixes the score.

Red customers need a conversation, not an instant ban. A warning that names the pattern, like ordering three sizes of everything for six months, gives honest customers a chance to adjust and gives abusers notice before consequences. The brands that do this well document everything, apply the policy consistently, and almost never need the ban because the warning works.

The pitfalls to avoid

Gift buyers will distort the score if you let them. A customer who buys gifts every December and returns half is not a bracketer; they are shopping for other people. Exclude gift-flagged orders or score gift and self-purchase behavior separately.

New customers need a grace period. Scoring someone's second order as if it were their twentieth produces false reds and angry emails. Keep new buyers unscored until they cross your minimum order threshold, and when in doubt, round toward generosity.

Finally, review the score's decisions regularly. Pull a sample of red-flagged customers each month and check whether the label still fits. Scores drift as behavior changes, and a customer who fixed their bracketing six months ago should not still be paying for it. A score that never forgives is a score your team will stop trusting.

Keep-rate scoring does not replace judgment. It focuses judgment where it matters: on the small slice of customers driving the outsized share of return cost, with the evidence attached and the honest majority left alone.