Celonis Product Documentation

The Annotation Builder asset in Studio is introducing a new feature that generates a preconfigured "Validation Table" that allows analysts to evaluate the accuracy of the annotations being generated. This table includes the original data, the AI-generated annotations (e.g., predicted categories, reasoning, etc.), and columns where analysts can add their feedback directly into the table using free text fields and pre-populated dropdowns.

Analysts can generate the validation table automatically by clicking the Use in a View button on the Annotation Builder screen. The validation tables can be added as a new View in Studio or embedded in an existing Studio View. The table is generated automatically based on the annotations generated with no additional set up or configuration required.

In the validation table, analysts can review the individual annotations, add comments, and collect data to improve the prompts used by this Annotation Builder. If the analyst provides feedback in the validation table, the accuracy statistics for this Annotation Builder are updated automatically and displayed on the Annotation Builder overview screen.

With this release, you will be able to apply event count filters to Views that have process filters enabled. Event count filters enable you to select objects where an event occurs less or more times than a defined start and end threshold e.g. only objects where "Invoice paid" happens more than once.

In this example, the Sales Order Activity event log will be filtered to show only the Confirm order event. Using this filter, you can see the event count distribution where the selected event occurred zero, twelve, and twenty four times per object.

For more information about configuring advanced filters, see: Advanced filters.

And to learn more about process filters, see: Process filters.