Moving to Abuse-Specific Scores

Overview

Upgrading to abuse-specific risk scores will give you more accurate results. You can also see multiple scores if you face multiple types of abuse. For example, we can tell you that someone is likely to post abusive content but not likely to commit payment fraud.

This tutorial walks you through the process of switching from our legacy Sift Score to an abuse-specific score.

Step 1: Opt-In

You can opt-in to abuse-specific scores in your console. If you're fighting multiple abuse types, you should read our guide to adding additional abuse types next. Once you've opted in, your existing API calls can start providing the new risk score in addition to the legacy score. You'll also be able to view both scores in the console.

Step 2: Give Abuse-Specific Feedback

The more precise you are in your feedback about users/orders, the better scores we can give you.

If your team uses the Sift console, you'll set up Decision buttons. You can connect them to your backend to automate based on review decisions. They also provide us with the specific feedback of which users are good and bad for each fraud type you're fighting. As a first step, you can set up a feedback-only workflow.

If your team uses an internal tool to conduct reviews, you'll connect our Labels API to your tool. This way, when your team finds an abusive user, the appropriate abuse-specific label will be sent to Sift. It's important to be as accurate as possible with the labels you send so that you can get the best results possible. You should even send labels for abuse types you aren't signed up for. If you're fighting multiple types of fraud and can resend historical labels using the new format, we recommend doing so.

In addition to Decisions or the Labels API, some abuse types have events that provide additional feedback:

• payment_abuse : $chargeback
• content_abuse : $flag_content

Step 3: Monitor Both Scores

Once you're providing abuse-specific feedback and viewing both legacy and abuse-specific scores, you'll want to evaluate the new scores before switching. There are two reasons you'll want to do this. First, it may take a couple of weeks of giving abuse-specific feedback before the new scores are better than the legacy score. Second, since they're created separately, the scores at which you block, allow, review, etc may be different for your abuse-specific score than the scores you currently use.

To properly evaluate, we recommend reviewing a sample of your users/orders each day. Here are some options:

• Create a feedback-only workflow for your new abuse-specific score at the event you automate on. Have your team review a few of these users each day until you're comfortable with the new score. This will also allow you to provide abuse-specific feedback which will improve your score.
• Each day, download a CSV of your users/orders from the last hour or day. Review a random sample so that your team is looking at users with high, medium, and low scores. Repeat this process for a week or so, providing abuse-specific feedback and evaluating the new scores.
• Record both the legacy and abuse-specific score you get back in our API response and use these to analyze the performance of the two scores.

After a week or two of this evaluation, you'll be able to start using the new score for automation.

Step 4: Use the New Score

Once you've verified where you want to accept, review, block, etc with the new score, you'll just need to make one or two changes to begin automating off of it.

• If you automate using the score or Decision used in our API responses, you'll just need to look for the new score field.
• If you automate using Decision Webhooks, you'll need to ensure that you have them set up to provide abuse-specific feedback instead of legacy score feedback.