Omniscore Overview

Omniscore uses two types of machine learning – unsupervised and supervised. The unsupervised machine learning focuses on short-term linkages and patterns, enabling it to catch emerging fraud attacks and anomalies that supervised machine learning cannot yet learn about due to the recentness of unseen attack types. Our supervised machine learning technology learns from historical data – decisioned orders and their outcomes.

The AI simulates how an experienced fraud analyst would review a transaction. The unsupervised machine learning aspect of Omniscore evaluates the transaction as a human would use instinct. The supervised machine learning aspect evaluates the transaction like the historical experience of seasoned fraud analysts. Together they allow us to calculate one highly-predictive transaction safety rating that can be relied upon for decisioning orders, so that there is less reliance on manual review and reactive fraud rules. The result is catching more true fraud and allowing more good transactions to generate revenue.

Transaction safety is the inverse of transaction risk. Omniscore is an indicator of a transaction’s safety ranging from .1 (unsafe) to 99.9 (safe). A safe transaction will have a relatively high Omniscore and an unsafe transaction will have a relatively low Omniscore.

Designed to make good decisions more intuitive, the Omniscore can be likened to U.S. academic letter grades that range from F to A. Most transactions will rate in the 80s and 90s (Bs and As). Transactions with issues will rate in the 60s to 70s (Ds and Cs). The riskiest transactions rate below 60 (F).

It is important to note that Omniscore is not a decision. It is a prediction of safety that is used by customers to decision a transaction, either automatically via creating a rule or manually while under review.

A Low/High Omniscore alert is generated when we identify a decrease in high Omniscore ratings and/or an increase in low Omniscore ratings on transactions within your merchant account. This means that increased indicators of risk were found, which can indicate a rise in fraudulent orders that might be worth investigating.

Since Omniscore is so accurate in predicting fraud, you can set one rule around it instead of creating large rulesets targeting fraud.

A suggested rule is to determine the decisioning threshold (at what value the Omniscore is set) based on the decline rate your expected fraud rate:

The decisioning threshold can be adjusted after analyzing decline and chargeback rates, and any other measures of performance important to the merchant.

A customer can create a No Change rule by using Omniscore. This allows the customer to watch how the score performs while not impacting actual transactions. While we're confident that Omniscore should be the only score used for fraud rule creation, some customers will want to test it. After three months (enough time for outcome data to be collected), the customer can compare performance with and without Omniscore.

You can create reports in Datamart to view current and previous Omniscore rates.

To view your current and previous transactions, create a Datamart report with an Omniscore less than 15 and their Current Status. Refer to the screenshot below: