Intelligent document processing overview

Michelle Welch

Updated October 19, 2026 21:27

Intelligent document processing (IDP) in integrator.io turns documents — native and scanned PDFs, Word, Excel, and PowerPoint files, TIFF scans, and images — into structured, validated data inside your flows. An AI agent reads each document and returns the fields you define in a JSON schema; the flow validates the result, delivers it to your ERP, CRM, or any other connected application, and sends anything uncertain to a person for review.

Use IDP when documents are how data reaches you: vendor invoices, purchase orders from partners that don't use EDI, remittance advice, shipping documents, receipts, and onboarding forms. There are no templates to build per layout and no models to train — you describe the fields you need, and the agent finds them in documents it hasn't seen before.

How intelligent document processing works

IDP is part of the AI agent step, so a document becomes data in the same flow that validates and delivers it. Every document moves through five stages:

Ingest — A flow picks up documents where they arrive — a file server, Celigo Storage, cloud storage, a mailbox, or an application's API — and passes each file to an AI agent.
Read — Celigo extracts the text and tables from the file, using OCR for scanned pages, before the AI model reads it.
Understand and structure — The AI agent returns the fields you define in a JSON schema, with formats standardized and repeating data such as line items returned as one array. Celigo validates every response against the schema.
Validate and route — Flow steps check the data against your business rules and systems of record, and branching routes each document type to its own path.
Deliver or review — Records that pass post to their destination. Records that don't wait in error management, where a person corrects and retries them.

What intelligent document processing reads

With intelligent document processing turned on, an AI agent reads native and scanned PDFs, Word, Excel, PowerPoint, and RTF files, TIFF and BMP scans, images, and plain-text files, up to 20 MB each. The same file types work on every model — a Celigo AI model or your own through a bring-your-own-key (BYOK) connection. For the full list, how each type is read, and every limit, see Supported file types and limits for intelligent document processing.

What you can build

Common patterns, each built from standard flow steps:

Accounts payable invoice intake — Extract vendor invoices, match the purchase order, screen for duplicates, and create vendor bills in your ERP.
Orders from partners that don't use EDI — Read purchase orders that arrive as PDF or Excel files and create sales orders through the same destination your EDI orders use. See Process orders from trading partners that don't use EDI.
Cash application — Extract who paid, how much, and which invoices from remittance advice, match open invoices, and post payments.
Logistics documents — Capture bills of lading, packing slips, and proof-of-delivery documents and match them to shipments and orders.
Supplier and employee onboarding — Extract and check W-9s, certificates of insurance, and new-hire forms before records are created.
Mixed inboxes and drop folders — Classify every incoming document and route each type to its own extraction agent and destination. See Classify documents and route them by type.

Start from a Marketplace template

Marketplace templates install as editable flows that you can adapt to your documents and systems. In Marketplace, filter By type and select AI Agents.

AI Powered Document Processing — Classifies files dropped into Celigo Storage into 11 document types and routes each to its own extraction agent.
AI AP Invoice Ingestion Agent (Email/PDF to NetSuite Vendor Bill) — Vendor invoices to NetSuite vendor bills, with purchase order matching and duplicate screening.
AI based Email Remittance for Cash Application Management — Remittance emails to NetSuite customer payments, with the Cash Application Manager integration app.
AI EDI Document Classify-and-Route Agent — Identifies EDI and non-EDI files that trading partners drop and routes each to the right flow.
AI Onboarding Document Review (HR intake - Slack) — Checks new-hire documents for completeness and drafts follow-up requests.

Security, privacy, and governance

Intelligent document processing runs inside the same governed platform as your flows:

Account-level opt-in — Off until an account owner or administrator turns it on. See Turn on intelligent document processing.
No copies of your documents — Celigo reads each file for the run and doesn't keep a copy. The extracted content goes to the AI model the agent uses.
Execution logs — Each agent run records the model, token usage, AI credits, and errors. See Monitor AI agent and guardrail activity with execution logs.
PII controls — A guardrail after the agent can detect or mask personal data before it leaves the flow. See Validate extracted data and route exceptions for review.
Human oversight — Exceptions wait in error management until a person retries or resolves them.

For how these controls fit together across AI workflows, see Enhancing AI enablement with TRiSM: Trust, risk, and security in action.

AI credits and AI agent actions

Reading a document with intelligent document processing doesn't consume AI credits.
Each file that intelligent document processing reads counts as one AI agent action.
The agent's model consumes AI credits as in any other agent run; larger documents use more. With a BYOK connection, your provider bills model usage to you.

See Track your AI agent actions and credits usage.