Meetings
Docs
Project management
Chat
Procurement
Spreadsheets
BIProcess business documents
Someone opens each PDF or image, identifies its type, retypes fields, checks totals and reference data, attaches the source, and routes incomplete or unusual records.
Your agent classifies the document, extracts the approved schema, validates fields and arithmetic against system data, stages clean records, and routes exceptions with source evidence.
Document arrives
intake folder or mailbox
Your AI agent
Classifies, extracts, validates, and stages the record while preserving the source file and a complete exception trail.
Required
Files
Reads the source document, metadata, sender, and intake location.
Writes the retained source and processing status.
Recommended
Accounting
Reads reference records, valid values, duplicates, and business rules.
Writes a staged record after validation.
Optional
Workflow automation
Writes the exception route, owner, and retry state.
Upload files to a controlled folder and write extracted fields to a review sheet. Keep stable document IDs and the original file.
Paste this into your agent or automation tool. Choose one document type and have its schema, validation rules, one clean processed example, and three real files covering clean input, a scan, and a duplicate or invalid document ready.
```text Setup prompt theme={null} Help me set up a business-document processing workflow for one document type. It should extract and validate each file, stage a clean record, and route every exception with evidence. It must never approve or pay anything. 1. Ask me which intake, file, and destination systems the document moves through. 2. Ask me for the document type, field schema, required fields, and source ID rules. 3. Ask me for format, total, date, duplicate, reference, and policy validations. 4. Ask me for one clean processed record and one useful exception example. 5. Build the workflow so it preserves the source, classifies the file, and extracts only approved fields with page references and extraction confidence. 6. Stage a record only when every required check passes; route unreadable, missing, duplicate, conflicting, or invalid files with the failed rule and source evidence. 7. Test it on my three real files and show me the record or exception for each. ```What good looks like
The clean file should stage correctly, the scan should preserve uncertain fields for review, and the invalid or duplicate file should route with the exact failed rule. Correct any extraction, total, duplicate, or destination error and rerun the same files.
Choose your trigger
Run it when a new file arrives in the named intake location from a known source. Quarantine unsupported types and unreadable files before record creation, and retain the original file throughout the workflow.
What runs without you
The workflow can stage documents that pass every required validation and route the rest to the named exception owner. The process owner still approves consequential downstream actions. Review a weekly sample of clean and failed files, and stop unattended staging if any validation failure disappears silently.
Pairs well with triage operational exceptions - the documents that fail validation become the exception queue.
