Data quality & fairness

One of the most challenging steps when building machine learning models is sourcing "feature data" — rich data about the subjects of your predictions. This step can be especially difficult, especially if you take privacy seriously. You may only have a single interaction with a consumer: this creates a cold start problem when you want to predict consumer behavior.

Faraday Identity Graph

This is where the Faraday Identity Graph (FIG) comes in. Faraday licenses opt-in permissioned consumer data from the world's most respected compilers. This means that you get access to hundreds of powerful features on nearly 300 million U.S. adults.

Identity resolution

Faraday uses FIG to build machine learning models automatically. Based on the identities you declare when registering datasets, we (fuzzy) match the people in your data to the known identities in FIG.

As a result, each of the people in your data gets "enhanced" with all the data in FIG. This allows Faraday's machine learning algorithms to discover robust, useful patterns in your historical data that can be applied to others to predict their behavior.

Fairness

We are committed to mitigating bias in our use of data and machine learning. We've written about it in the past.