This article explains Loop's machine learning fraud model powered by Loop Intelligence, which automatically flags potentially fraudulent returns based on computer details, geographic anomalies, order attributes, and customer history. It covers how merchants can review fraud risk assessments in the Loop admin with fraud signal analysis, use the "Report as Fraud" and "Mark Safe" buttons to provide feedback, filter returns by fraud risk in the dashboard, and configure Workflows to take actions on high-risk returns like manual review or excluding return methods.

Powered by Loop Intelligence - Loop’s proprietary foundation model built entirely in-house to understand how every order, shopper, and return connects. It’s the brain behind smarter fraud detection, forecasting, and the future of post-purchase intelligence.

The video below covers the following:

• Interacting with high-risk or reported fraudulent returns in the Loop admin.

• The Fraud Risk report that Loop offers.

• Using Loop's Workflows feature to further support merchants' fraud assessments.

Aspects of the video are explained in more detail below.

Loop has developed a fraud model to help merchants identify potentially fraudulent returns and prevent fraud before it occurs. Loop's fraud model, available to all Loop merchants, is a machine learning (ML) classification model that automatically flags returns based on characteristics typical of fraud or abuse. This flagging creates a dataset that can be compared against new and future returns to determine if they might be fraudulent.

In short, the fraud model helps Loop evaluate new returns as they come in, and it lets merchants know when a return resembles past fraud or abuse.

The model looks at a variety of feature categories to determine whether or not a return appears fraudulent, including:

• Details about the customer's computer that can provide evidence of attempted identity concealment.

• Geographic anomalies on the shipment and return.

• Order and return attributes such as return reasons, refund amount, number of returned items, and others.

• Customer history including past cases of fraud or patterns of good returns behavior.

Loop's fraud model sorts returns into risk tiers. Most returns are low risk. Returns that carry characteristics typical of fraud or abuse are flagged as high risk. Above high risk sits a stricter tier, critical risk — reserved for the returns Loop is near-certain are fraudulent, such as those submitted by a shopper who has been flagged for theft on the Loop network. Critical risk is a narrow subset of high risk, surfaced separately so you can act on the most confident cases without second-guessing.

Note: Loop also uses shopper history across its network to protect trustworthy shoppers from false flags. A return from a shopper with a long, diverse, and clean history that would otherwise trip the high-risk threshold can be automatically downgraded to low risk.

Setting up the fraud model requires no work from the merchant; the model will automatically evaluate all new returns. However, if a merchant wants the model to take action on high-risk returns, they may set up a workflow to do so.

Loop released the Report as Fraud (RAF) button in February 2024, which allows merchants to report fraud to Loop. This feature is separate from Fraud Tools and is on for all Loop merchants, including those that don't use Fraud Tools.

The returns submitted by way of the RAF button make up Loop's training datasets. Any newly reported returns will be used in future training sets. This allows Loop to adapt as fraud indications change.

The following sub-sections explain what merchants can expect to see in the Loop admin regarding the fraud model.

Each return is evaluated before it is submitted. When Loop identifies a high-risk return, merchants will see a Potential fraud alert tile on the return details page. This tile shows a summary of the primary signals that triggered the flag and provides a link to the full report.

The tile also includes two feedback buttons at the bottom:

• Mark safe: Indicate that this return is not fraudulent.

• Report fraud: Confirm that this return has been identified as fraud or abuse.

Clicking View full report → opens the Fraud report modal, which provides a more detailed breakdown of why the return was flagged. The modal is organized into two sections:

• Primary signals: The factors that most strongly influenced the high-risk score. Each signal includes a plain-language explanation with specific contextual details; for example, noting that a shopper has a 100% historical return rate. The number of primary signals is shown in the section header.

• Low impact: Additional factors that were considered but had lower or no influence on the score. These are listed without detailed explanations and can be expanded or collapsed.

The Report as fraud button at the bottom of the modal allows merchants to provide feedback on the return. Loop uses this feedback to continue improving the fraud model.

When a return meets the critical-risk bar, Loop surfaces it more prominently than a standard high-risk return. Instead of the fraud card in the right-hand sidebar, a critical-risk return displays a full banner at the top of the return detail view, signaling that this is a "review now" case rather than routine review.

Loop's recommended action on a critical-risk return is to reject and report as fraud. As with high-risk returns, you retain both feedback actions — Mark safe and Report fraud — so you can always override Loop's recommendation.

Note: Critical risk is a narrow, high-confidence subset of high risk. It's a strong signal, not a certainty — you're always in control of the final decision on the return.

Merchants can easily find high-risk and confirmed fraud returns in the returns dashboard.

High-risk returns for which the merchant hasn't yet provided feedback will have a yellow fraud icon (a shield with a magnifying glass):

Returns that have been confirmed by the merchant as fraud, regardless of the initial evaluation, and Critical-risk returns will have a red fraud icon (a shield with an exclamation point):

Merchants can filter their returns by high fraud risk. This filter can be stacked with shipping status and other filters.

In the Loop admin, merchants can go to Returns management > Policy settings > Workflows to use fraud risk as a workflow condition.

With this condition, merchants can implement a limited set of actions:

• Exclude return methods (for example, Happy Returns). Refer to the FAQs below for some Happy Returns limitations.

• Exclude Keep Item (other "Exclude outcome" options will not work).

• Send the return to manual review (rejecting the return will not work).

• Change the processing event.

In addition to the fraud high risk condition, you can build workflows on a separate critical fraud risk condition. This lets you apply stricter automation to the near-certain cases than you do to the broader high-risk pool. For example, you might route critical-risk returns to manual review before a label is issued, while high-risk returns simply process on inspection. The two conditions can coexist in your workflow configuration, and critical risk works with nested conditions if you want finer control.

Note: Loop evaluates the fraud risk of a return just before it is submitted, meaning the fraud high risk condition can’t be used to change anything during the return itself.

To learn more, visit the Workflows article.

Can I set up a fraud workflow if I use Happy Returns? Yes, with some caveats. If you want to change a high fraud risk return to manual processing, you must also exclude Happy Returns as a shipping option. Happy Returns must be processed on scan, so in order to manually process high fraud risk returns they must be returned box and ship.

Do I have to build a fraud workflow to get value out of this feature? No. Loop Fraud Tools will monitor and evaluate each return even if no workflows are acting on those evaluations. That said, the fraud model in and of itself does not take any actions on returns. If you would like for Loop to take action on high-risk returns, the only way to do so is by configuring a workflow.

How does Loop protect or anonymize my data for use in the fraud model? While Loop does use cases of a store’s confirmed fraud to train our fraud model, no users or stores can see data from another user or store. They can only see the model’s evaluations and reasonings that have been compiled and generalized from various stores.

Can I ‘opt-out’ of having my shop’s data shared into the ML fraud model? Loop only uses reported fraud to train the fraud model. If you do not report fraud or use Fraud Tools, your data will not be used to train the fraud model.

For additional questions, please reach out to support@loopreturns.com.