Changelog

A new dataset type has been added—merge dataset—which allows you to merge two sets of data using a join key as an identifier. This type of advanced dataset can be useful if you have customer data in two different tables–for example, customers in one, and orders in another–and would like to combine them for more streamlined usage.

If an individual isn’t recognized at a particular address, Faraday now can still accurately enrich their data using the aggregate traits of others in the same household. This update extends our existing postcode-level aggregation to the address level, ensuring even better coverage for modeling.

Previously, when connecting your data to Faraday, individuals we didn’t recognize (those without a match in the Faraday Identity Graph, or FIG) were simply dropped from your data. Now, data about these unrecognized identities are ingested and stored normally. You can now review how your data is being enriched with FIG traits via the new enrichment column in Datasets. The higher the fill rate, the more identities we were able to enrich with our built-in consumer data.

These changes may result in larger cohort counts due to Faraday being better able to handle partially-matched identities. On the other hand, your pipelines that target the everyone cohort as a population to include may be smaller due to identity resolution improvements, where previously individuals may have been incorrectly counted twice. Overall, these changes should increase prediction accuracy.

Faraday’s UI now provides API shortcuts, allowing users to seamlessly transition from manual setup to programmatic workflows. Each resource’s three-dot menu includes a new "API Shortcuts" option, which opens a modal displaying pre-built API requests for that resource.

Faraday’s Outcome abstraction allows users to define propensity objectives which predict whether a person is likely to join a given Cohort (often, to experience a given event). In other cases, users want to determine which variant of a given event a person is likeliest to experience. For example, which product they are likely to buy next. This is what we call product recommendation.

A new POST /recommenders endpoint has been added to create recommenders, where you define parameters like stream name, eligibility, predictors, and bias mitigation.

View our documentation to learn more.

While creating an outcome, the bias section provides options for mitigating either of the two sensitive dimensions released at launch: age and gender. To mitigate bias, simply toggle on either (or both) age and gender, and choose a mitigation strategy from equality or equity. Equality aims to neutralize bias, whereas equity aims to invert it. For more info, check out our blog.

A new connection has been added for Secure File Transfer Protocol (SFTP), allowing users to securely transfer CSV data to and from Faraday.

Now, aggregated deployments in the UI can support census tract and block group zoom levels, which can be useful for canvassing & market sizing use cases.

Introducing a new look for the outcome performance dashboard to make it easier to interpret the results you can expect when using the outcome in the real world. In this new look, Faraday will place your outcome’s score on a grid whose Y axis is associated with Faraday’s ability to predict the outcome, and the X axis is associated with the business value expected due to the predictive lift of the outcome. The further toward the top right corner, the better your real-world results.

At the end of 2023, the Explore map will be deprecated due to low engagement and insights that are no longer as relevant as they once were.