How clustering works

Cluster-based rationalization uses a generative AI framework (GAF) scheduled job to create groups of similar control objectives before the user begins rationalization. The user then reviews these clusters rather than searching for similarities record by record.

The GAF job evaluates control objectives based on their name, description, and table name. The results are stored in the control objective cluster table. Clusters consist of normal groups, which contain similar control objectives, and outlier groups, which contain control objectives that the system could not confidently place in a cluster.

Clusters and outliers

After the GAF job runs, control objectives are organized as follows:

Clustered control objectives

Control objectives that the system has grouped together based on similarity. These appear in the Clusters module along with their cluster group and subgroup information.

Outliers

Control objectives that could not be confidently assigned to a cluster. Outliers appear in both the Clusters module (accessible through the Outliers filter) and the dedicated Unclustered Control Objectives module. A single main outlier group can contain multiple subgroups, each identified with a -1 suffix.

Users can rationalize both clustered and outlier control objectives. Although outliers are those the system was not confident about, users can still select and rationalize them if they determine that two or more outliers are duplicates.