11.1 Semantic classification versus pattern rules
| Pattern / keyword condition | AI document type condition | |
|---|---|---|
| Looks for | A value: an IBAN, a national ID number, a word list | The kind of document, judged from its overall meaning |
| Output | Matches (with a count) | A probability between 0 and 100 % for each of your document types |
| Fires when | At least Min. count matches are found | The probability reaches the condition’s Threshold (%) |
| Languages | Whatever the pattern matches | Turkish and English (a type taught in one language also works partly in the other; teach both for best results) |
| Needs | Nothing beyond the policy | The AI module switched on, a trained document type, a supported agent |
| Where evaluated | All agents, Office add-ins, web add-ins, console Rule Test | Windows and macOS agents only (see section 11.10) |
| Typical use | “Contains a card number” | “Is a payroll file”, “is board minutes”, “contains a CV section” |
Key points:
- The product ships with no predefined document types. Every type is yours, trained from your documents on your server.
- Rules still decide. AI is only a condition type. Action (Recommend / Warn / Enforce), groups, exceptions, AND/OR logic and “highest label wins” work exactly as for other rules. A pattern rule and an AI rule can both match; the higher label wins.
- AI cannot be used in masking rules — AI finds documents, not text to mask.