AI Document Classification

The AI Document Classification action examines each page of a document and classifies it against a Record Type's category / subcategory tree

This sends a document to be AI processed so each page can be matched to one of the valid classifications defined on a Record Type. Contiguous pages that share a classification are grouped, and the result is turned into separate, categorized documents. Like the AI Invoice Extraction action, this is processed immediately and NOT sent to a queue.

It works the same way as the existing Auto-Classify action - it can commit the classification immediately or stage the results for a later Auto-Classify Commit action.

Document

Active Document - Check this ON to classify the active document. This will ONLY work if the workflow is acting on a routed Document and not a Record.

From Category / SubCategory / Name - Specify the Category / SubCategory / Name of the Document to classify. The Documents in the active Record are queried and the first matching Document is processed.

NOTE ON (AI): All AI classification will incur extra fees. Check with your sales rep for more information.

Classification

Record Type - The Record Type whose category / subcategory tree the pages are classified against. Choose a specific Record Type, or leave it set to use the Record's matched Record Type. Setting descriptions on the Record Type's categories and subcategories helps the AI disambiguate similar classifications.

Commit Classify - Turn this ON to commit the classifications and split out the categorized Documents automatically. Turn it OFF to NOT commit; the action will instead stage a CLASSIFYDATA document in the Record showing how the classification would have happened, for a later Auto-Classify Commit action to apply.

Classify Level - Choose whether to classify at the Category + SubCategory level or at the Category level only.

Unclassified Pages - Controls what happens to pages the AI cannot confidently classify. When set to KEEP, an unclassified page is merged into the previous classified group, or placed in its own "Unclassified" document when it leads the source document. When set to DROP, unclassified pages are excluded from the results.

Auto-Generate Name - When ON, the AI generates a concise, descriptive document name for each classified group based on the visible content (such as a company name, document number, or date). When OFF, no name is generated.

AI Level

Quality - Specify the AI quality to use for this action.

Standard works well for most standard business documents. Cost: LOWER
Advanced works best for poor scans, dense legal documents, or highly similar categories that need more nuanced reasoning. Cost: HIGHER

Max Tokens - All newer AI processes are charged by token counts. You can set a maximum here to avoid having overly large documents charged at an extremely high token rate. The default for this setting is 16,384. Note that each page image consumes input tokens, so cost scales with the number of pages.

Advanced

Extra Instructions - Any extra instructions for the AI agent to follow. You do not need to ask it to classify the pages as it already does that. This is more for hints like "treat cover sheets as Unclassified" or "a remittance stub belongs with its invoice" and the like.

A typical fully automated pipeline is: AI Invoice Extraction → AI Data Matching (look up vendor) → AI Document Classification → (Auto-Classify Commit, if Commit Classify is OFF).