Rule Explainability
Rule Explainability lets you see exactly why Premail classified a message the way it did. Click any entry in the Activity tab to open the explanation view, which shows the AI's full reasoning for why it assigned a category and how confident it was.
The explanation view is useful when Premail does something surprising. If a message was archived that you wanted to keep, or left in your inbox when it should have been filtered, the reasoning trace tells you what the AI saw and why it reached the conclusion it did.
What will you see?
- The category the AI assigned and its confidence score
- The specific phrase or signal from the message that drove the classification
- Which rule matched and what action was applied
- An option to create a custom rule or add the sender to your allowlist directly from the explanation
What does a rule trace show?
Expanding a rule trace walks you through the decision in the order Premail made it: which rules were evaluated, what the model classified the message as, what confidence it returned, and how that compared against each rule's threshold. A rule that did not fire is as informative as one that did - the trace shows why it was skipped.
What should you do with a rule trace?
- Right classification, no action - the rule's confidence threshold is above the score returned. Lower the threshold.
- Wrong classification - the rule wording is ambiguous. Rewrite it more concretely in Custom Rules.
- An earlier rule claimed it - reorder or narrow the rule that fired first.
- The sender should never be filtered - skip the tuning and add them to your allowlist.
How do you test a rule change without waiting?
After adjusting a rule, use the Dropzone to drop the same message back in and confirm the new behaviour immediately, rather than waiting for similar mail to arrive.
Still have questions?
Contact Support (for Pro members)