The AI Document Split action examines one document containing mixed documents — invoices, receipts, statements, purchase orders, correspondence, all fed through the scanner in a single pass — works out where each document starts and ends, and splits the original document into separate documents that stay in the same Record. Each new document keeps the original document's Category and SubCategory and is given a descriptive name built from what the AI can see on the page, such as the company, the document kind, a number, or a date.


This makes it different from its two siblings. AI Invoice Split and AI Generic Split split a document into a new Record per document; this action keeps everything in the Record it started in. It is also different from AI Document Classification, which sorts pages into a Record Type's category tree — classification groups consecutive pages of the same kind together, so a run of ten invoices stays one document there. AI Document Split is for exactly that case: it separates documents even when neighboring documents are the same kind, and it needs no Record Type and no category tree at all.


Use it when a Record collects paperwork as one scan and you want each piece filed as its own named document — an employee file scanned in one pass, a loan record's supporting documents, a day's receipts scanned as a single batch.


AZURE ONLY: This is an AI action and runs only on Azure-hosted / AI-licensed sites.


How the split is found

The action watches what each page says about itself — what kind of document it is, who it is from, its document number, and its date. When a page clearly asserts a different identity than the page before it, a new document starts; pages that agree, or that say nothing, continue the current one. Nothing needs to be printed on the pages for this to work — no separator sheets, no barcodes, no page counts — and there is nothing you have to configure to make it split. When in doubt the AI keeps pages together rather than over-splitting.


The original stacked document is removed once the split completes, and the Record's other documents are not touched. If the AI decides the stack is really one document, it is simply renamed in place.


What the action captures

The AI has to read each document's header to find the boundaries, so the same call also captures each document's key values: the generated Name plus DocType, Vendor, DocNumber, Date, and Amount. Nothing is written to the Record. The captures are held in memory for the rest of the workflow run as a JSON array — one object per new document, including the document's ID and page range — under a Temp Variable

 you can read back with [DMTEMP(SPLITDATA)].

To use the values, follow this action with a JSON Loop over that variable. Inside the loop each captured value is available as OBJ:Name, OBJ:Vendor, OBJ:Amount and so on, and the actions you place there decide where the data goes — line items, an e-form table, Record fields, a REST call, or anywhere else. If the workflow does nothing with the variable, the values simply expire with the run; the split itself is already saved.


Documents whose boundaries the AI was unsure of get a Confidence of "low" and an AIWarning property in their JSON object, which the loop can filter on to route those for review.


Options

Document

Active Document — On to split the Record's active document when there is one. Otherwise the Cat/Sub/Name selector below chooses the source, and the first matching document is used.


Cat/Sub/Name — Which document on the Record to split. Leave blank to take the Record's first document.


Captured Data

Temp Variable Name — The name the captures JSON is stored under, SPLITDATA unless you change it. Leave the default so the JSON Loop that follows can always use [DMTEMP(SPLITDATA)]; change it only if one workflow runs two splits and needs to keep their results apart.


Extra Capture Fields — Optional. One line per field to capture beyond the standard set, or a constraint on how a standard field is read (allowed values, format, where it appears on the page). Each extra field appears as a property on that document's JSON object. Header values only — line items cannot be captured during a split.


AI Level

Quality — Standard suits most business documents; Advanced costs more and is best for poor scans, dense pages, or subtle boundaries. Vision detection needs a capable model, so there is no Economy option for this action.


Advanced

Extra Instructions — Boundary rules specific to your documents, for example "Receipts are usually single pages" or "Terms and Conditions pages belong with the document before them". When in doubt the AI keeps pages together rather than over-splitting.



Processing several documents at once

The action splits one document per run. When a Record collects more than one Document, wrap the action in a Document Loop to process them all. Set the Document Loop's Category, SubCategory and Name to select the proper Documents, and inside the loop turn the split action's Active Document toggle ON so each pass splits the document the loop is currently on. The loop takes its list of documents before it starts, so the documents the split creates are never picked up and split again by the same loop, and running over a document that is already a single document is quick and harmless — it is simply renamed in place.


Keep the JSON Loop inside the Document Loop as well, placed right after the split. Each split replaces the Temp Variable with its own captures, so a single JSON Loop placed after the Document Loop would only ever see the last stack's values. With both loops nested, each stack is split and its values are written out before the next stack begins. Make sure to handle the values inside the loop - i.e. add to a table or line items. 



TIPS

  • If every document in the stack is the same kind and you want a new Record per document, use AI Generic Split (or AI Invoice Split for invoices) instead — those actions exist for that shape of work.
  • If you want the pages sorted into a Record Type's document categories rather than separated document by document, use AI Document Classification.
  • Put the JSON Loop immediately after this action in the same workflow. The captures live only for the workflow run — they are not saved to the Record, so a later, separate workflow cannot read them.
  • If the split is close but not right, describe the exception in Extra Instructions before reaching for Advanced quality — a sentence about continuation pages or single-page receipts usually fixes more than a bigger model does.



NOTE ON (AI): Detecting splits with vision AI consumes tokens and may incur extra fees. Workflow tokens are billed separately from End-User or Admin tokens — check with your sales rep for more information.