Scope
This guide covers automation for Predictions & Calibration. Availability depends on your tenant package, role and configured providers.
Purpose
Compare model predictions with what actually happened.
Refresh supported propensity and prediction outputs.
Before you start
Correct tenant and project selected
Required source or provider is available
An accountable owner is identified
Test data can be separated from live activity
Procedure
- Complete the manual workflow successfully before enabling repeat execution.
- Refresh supported propensity and prediction outputs.
- Define the trigger, eligibility conditions, action and intended output.
- Add a failure path, owner, retry rule and stop condition.
- Run a controlled test and inspect every generated record or delivery event.
- Keep the automation bounded: Expose sample size, uncertainty and model limitations.
Verify the result
Compare predicted versus actual using calibration measures. Confirm that the intended record, state, run or report is visible to another authorised operator and that its next owner is clear.
Completion checkpoint
The capability is in the intended state, evidence is current, exceptions are owned and the next action can be understood without a separate status message.
Controls and limitations
Expose sample size, uncertainty and model limitations. Predictive, attribution and recommendation outputs support operator judgement; provider and data limitations remain relevant.