Model-Based Fraud Scoring is a technique used by merchants to predetermine the level of risk involved with an order in a card-not-present transaction. Each order that is placed is given a score. Merchants are then able to use this score to decide whether they want to reject, accept or review the order for other fraud checks. Often detail beyond the score helps determine which specific review tools or techniques should be used for each case of suspected of fraud the merchant decides to review.
Multi-merchant risk models use many tools and signals to determine an overarching “fraud score” for each transaction. This is usually a numeric score, or sometimes an Approve, Deny or Review recommendation, produced by many given factors to provide a ranking to determine the level of risk presented with a given order. The score is the result of proprietary algorithms and the data that is fed into the scoring model.
The value of model-based fraud scoring comes in two primary areas:
The fraud scoring service can incorporate many point tools within the score, including IP geolocation, proxy detection, Device Identification and more.
Merchants can benefit from fraud scoring service’s network data, leveraging shared velocities and blacklists. In other words, a bad actor that has been seen at another organization’s website can be identified and blocked before doing damage against another organization.
How does it work?
Merchants first send all transaction details they have received from the order they are investigating to a provider who feeds it into a scoring model.
The more data that is provided, the more accurate the result of the score will be. How each vendor uses all given variables and inputs within their model to determine a fraud score is unique to each vendor.
The vendor then issues a numeric score back to the merchant.
Model-Based Fraud Scoring can hold a lot of value considering the many tools and techniques that are implemented to determine a risk score. It is also a much more economically efficient way determine fraud checks than it would be to have them done manually.
There are many differences to consider when comparing providers. More information on fraud vendors is provided on our model-based fraud scoring technique page.
It is also important to consider the differences between model-based fraud scoring and a true custom modeling solution. Model-based fraud scoring applies the same model to different organizations, often with different models for various industries and risk profiles. A model-based fraud scoring solution may have a machine learning component to it that adjusts the model for a given merchant over time, but organizations start with the same base models. A custom modeling solution, however, is completely bespoke to each organization. This requires a much longer time to implement the model and requires extensive historical data analysis.
An important question to ask vendors when determining which model-based fraud scoring vendor to choose for your business is to understand if or how Machine Learning used in the model. Is Machine Learning incorporated into the model that will maintain and update the model overtime to make sure it is kept relevant? Machine Learning models branch out into supervised and unsupervised models. This is especially important to know in determining how the data is fed into the model.
To learn more about the understanding of the fraud solution marketplace and the fraud solution providers servicing that market, we’ve developed a Fundamentals For Selecting A Fraud Solution Provider training session where you can learn how to determine the ideal fraud solutions for your business.