Lift score on predictions

Outcome predictions now include a lift score alongside probability and percentile. Lift score tells you how many times as likely a person is to convert as average: 1.0 is average, 2.0 is twice as likely, and 0.5 is half as likely. That makes scores easier to read and share with people outside your data team: instead of "Jane has a 6% chance of converting," you can say "Jane is 3x as likely to convert as average." It's calculated from the model itself, so it stays accurate when your models refresh. It's the same ratio as the Lift column in an outcome's lift table, which reports it for groups of top scorers. The lift curve and Average lift over random on the metrics tab show lift over random, which is lift score minus 1.

Lift score is available on identified, referenced, and aggregated deployments. Hashed deployments don't include it. The column is fdy_outcome_{outcome_id}_propensity_lift, and aggregated deployments also get count_ and avg_ versions of it. The Lookup API returns the same key.

New deployments show lift score under Structure. If a deployment created before September 23 doesn't include lift score, recreate it once the pipeline has rebuilt. Editing its structure won't add it. Some structure formats pick up lift score on their next build; formats that only send identity columns don't include it. See lift score for details.