Changelog

With this update, you’ll find an include prediction explanations checkbox that tells you which data points had the highest impact on a given individual's predictions. Check out our blog post for more info.

When you create an outcome, the new bias section provides a summary of the bias that was discovered while the outcome was building. You’ll find tabs for different forms of bias–data, power, and predictions–as well as whether or not Faraday found the outcome to unfairly privilege certain sensitive subgroups over others. Click into any of the tabs to find the breakdown of given subgroups and their bias.

FIG is now able to update on an incremental basis, which should result in faster updates & more accurate predictions.

Users can now duplicate cohorts via the three dots on the cohort list view, or within the cohort itself.

When creating a dataset, you'll find previews of the data each of your properties represents while mapping fields. Similarly, when creating a cohort, you'll see a visual preview of the U.S. population baseline for Faraday traits that you add to the cohort.

Until now, the ability to filter a deployment by a percentile of scores existed in both the newer "filter" menu as well as the older "limit" menu. With this update, that option has been removed from "limit" to remove any confusion surrounding the redundant option. Users should continue to use the "filter" option and select "outcome percentile" to accomplish this task.

Faraday's Lookup API enables users to pull individual real-time insights via API.

The Outcomes view now includes various features that were previously stored in the technical model report, such as lift table and features of importance. The full technical model report can still be accessed via the three dots in the upper right when viewing an outcome.

Managed connections to your favorite martech platforms like HubSpot, Iterable, and Facebook are now easier than ever to use. If they're part of your subscription, head to Connections and enter your credentials.

Ever wondered what Faraday “scores” really represent? Starting now, we’re returning true probability values. That means that someone with a propensity score of 0.8 for a given Outcome truly has an 80% probability of achieving that Outcome. The calibration process we added also has the benefit of improving accuracy, which you will see by default in all new models. Check out our probability scores blog post for more details.