Intelligent Document Processing (IDP) Workflow
Turning incoming documents (invoices, forms, contracts) into clean data: classify, extract, validate against business rules, send low-confidence fields to a person and post the rest straight into your systems.
4 teams10 steps2 decisionsFree download
Free to download. Edit it in Flow Studio to change the teams and steps, ask AI for changes, or export to PowerPoint, Word or draw.io.
Step by step
The same flow as a numbered list, ready for an SOP, a BRD or a ticket.
| # | Step | Who | Next |
|---|---|---|---|
| 1 | Document arrivesStartEmail, upload, scanner or SFTPInbox · step 2 | Inbox | step 2 |
| 2 | Identify the document typeClassifier; split bundles; reject unreadable filesDocument AI · step 3 | Document AI | step 3 |
| 3 | Extract the fieldsLayout OCR + LLM to schema; per-field confidenceDocument AI · step 4 | Document AI | step 4 |
| 4 | Check against rules and recordsVendor and PO match, totals, tax ids, duplicate checkDocument AI · step 5 | Document AI | step 5 |
| 5 | Confident and valid?DecisionEvery field above its threshold and all rules passDocument AI · Yes: touchless: step 9, No: step 6 | Document AI | Yes: touchless: step 9, No: step 6 |
| 6 | Check the flagged fieldsSide-by-side view with highlights; correct only flagged fieldsReviewer · step 7 | Reviewer | step 7 |
| 7 | Valid after review?DecisionReviewer · Yes: step 9, No: step 8 | Reviewer | Yes: step 9, No: step 8 |
| 8 | Returned to senderEndInbox · Done | Inbox | Done |
| 9 | Post into the systemAPI with idempotency key; attach the originalBusiness system · step 10 | Business system | step 10 |
| 10 | ProcessedEndBusiness system · Done | Business system | Done |
When to use it
Use it to automate accounts payable, onboarding forms, KYC documents or contract intake with document AI, as a replacement for template-based OCR or manual keying.
Who's involved
- Inbox2
- Document AI4
- Reviewer2
- Business system2
Yours works differently?
Describe your own process in a few sentences and Flow Studio draws it, with your teams and your rules.
Describe your processarrow_forwardFor the technical team
The build decisions behind this flow. Each step above also carries its technical detail.
- 01Ingest: email inbox, SFTP, scanner or upload; de-duplicate by file hash.
- 02Classify: document type model or LLM classifier; split multi-document PDFs.
- 03Extract: layout-aware OCR plus LLM to a JSON schema per type; per-field confidence and source coordinates.
- 04Validate: master data lookups (vendor, PO, tax ids), arithmetic checks, duplicates within 90 days.
- 05Post: ERP API (e.g. SAP, NetSuite, Dynamics) with idempotency keys; corrections logged for retraining.
How to get it right
- 1Set confidence thresholds per field, not per document. A wrong bank account matters more than a wrong reference.
- 2Validate against your own data: supplier exists, PO matches, totals add up. Rules catch what extraction misses.
- 3Feed every human correction back into your test set and prompt examples.
- 4Measure touchless rate and field accuracy weekly, by document type and supplier.
Questions
What is intelligent document processing?
Software that reads documents the way a person would: it classifies them, pulls out the fields you need with AI, checks them against rules and your systems, and routes only uncertain cases to people.
How accurate is AI document extraction?
Well-tuned systems commonly reach high field accuracy on clean, common documents, and lower on handwriting or unusual layouts. That's why per-field confidence thresholds and human review of low-confidence fields matter.
More process flow templates
- AI feature integrationTaking an AI feature from idea to production: pick the use case, test models against real examples, add guardrails, pass a security review, and roll out in stages with monitoring.
- RAG pipelineHow retrieval-augmented generation answers questions from your own documents: documents are split, embedded and indexed, then each question pulls the best passages and the model answers with citations.
- AI agent with tools and MCPHow an agentic AI system works through a task: it plans, calls tools through MCP servers with scoped access, asks a person before risky actions, checks results and loops until the goal is met.
- MCP server integrationConnecting an internal system to AI assistants through a Model Context Protocol server: design the tools, secure them, test them, review them and roll them out with monitoring.