A Bloomreach custom prediction is a model you define yourself from four parts. A target filter describes the customers who did the thing you want to predict. An eligible-customer filter decides who gets scored. Then come the features the model may learn from, and the algorithm settings. The output per customer is a probability (yes or no), a number (regression), or a best-fit class. Custom predictions sit alongside the prebuilt templates for purchase, churn, email open, and in-session intent.
Updated
Category
AI, Bloomreach Campaigns
Also known as
custom predictions; Bloomreach custom model
In practice
Before opening the setup, write the target as an event with a time window. Narrow eligibility to the customers a campaign can reach, and write the campaign that will read the score first. Judge the model by lift against a holdout group, not by its accuracy metric.
Something out of date?