# How do you score return fraud risk before the order ships? | WardrobeWatch

Source: https://wardrobewatch.com/blog/2026-09-26-pre-shipment-fraud-risk-score.html

September 26, 2026

Almost every return fraud control runs after the damage is done: the item ships, gets worn, comes back, and the brand discovers the problem during inspection. Pre-shipment scoring flips the order. It asks a simple question at checkout: how risky is this order, based on the customer, the cart, and the timing? The answer does not block the sale. It decides the terms.

## What the score is allowed to change

A risk score that cancels orders is a lawsuit waiting to happen. A risk score that adjusts policy is just operations. High-risk orders can go out with standard rather than free return shipping, a photo-evidence requirement on any return, and inspection notes queued for the warehouse. Low-risk orders keep the generous policy that converts browsers into buyers. Nothing about the purchase changes. Only the fraud surface shrinks.

The point is that fraud controls and conversion controls do not have to be the same policy. Blanket policies force a choice between converting honestly and protecting margin. Scored policies give each order the policy its risk profile deserves.

## The signals that work

Customer history dominates the score. Return rate over the trailing twelve months, wardrobing flags on past returns, chargeback history, and how long the account has existed together explain most of the variance. A brand-new account ordering three sizes of an occasion dress one week before a major local event scores differently from a three-year customer ordering one sweater.

Cart composition is the second layer. Full size runs of one SKU, occasionwear ordered alongside nothing else, and multiple units of high-resale items are the patterns that predict wardrobing and bracketing. Timing is the third: orders clustered right before event seasons, holidays, and festival weekends carry more risk than the same cart in a quiet month.

Device and payment signals round it out, but keep their weight modest. They add value at the margins, and they are the signals most likely to generate false positives on shared devices and family accounts.

## Keeping the score honest

Two disciplines keep pre-shipment scoring from rotting into profiling. First, score patterns, not people: the score should move on observable behavior across orders, not on demographics or geography. Second, audit the score's outcomes quarterly. Look at what the high-risk orders actually did. If a large share returned cleanly, the model is too aggressive and is costing you the customers you wanted to keep.

Also keep humans in the denial loop. A score can tighten policy automatically, but outright order cancellations or account restrictions should always pass through a human review. Automation sets the terms. People make the final call.

## Starting small

You do not need a machine learning pipeline to start. A spreadsheet that ranks customers by return rate and flags wardrobing signatures will catch the worst accounts this week. Add cart-level rules next: size runs, occasionwear clusters, event-season timing. Graduate to a proper score only when the manual process is working and you want it to scale.

The brands that win at return fraud are not the ones with the strictest policies. They are the ones that know, before the box leaves the warehouse, which orders deserve the strict policy and which deserve the generous one.

### Keep reading

[Can you catch wardrobing in rental and try-before-you-buy programs?](2026-09-25-wardrobing-rental-try-before-you-buy.md)

[How do you avoid false positives in wardrobing detection?](avoid-false-positives-wardrobing-detection.md)

[Returns fraud glossary](../glossary.md) · [WardrobeWatch vs Loop](../alternatives/loop.md) · [WardrobeWatch vs Happy Returns](../alternatives/happy-returns.md)

[Get a free returns fraud scan](mailto:hello@wardrobewatch.com?subject=Free%20returns%20fraud%20scan)