Building an Enrichment Model requires labeled data to allow the classifier to learn from good (or Golden) examples of each label (or class) within the data set. Labelling takes time, is often labor intensive and can be inaccurate (humans can disagree on the label "is that a deer or an antelope?").
For situations where you have a data source, know a little about how that data is broken down but do not have any labels it is possible to "BOOTSTRAP" or perform a "ZERO SHOT TRAINING" to generate an Enrichment Model based on what you know as a first step, this can then be used on new entries in this or new data sources.
Once created you can use Active Learning to look at these initial labels and correct them or add new label classes if you find you missed a few sample labels. These updates can then be used to retrain the Enrichment and its performance compared to the 'older' version of the model, allowing you to decide on whether to keep this new version or not.
Once happy with your new Enrichment you can apply it to new Entities within the same project.
Follow our Bootstrapping guide to create your Enrichment without labels