n8n Workflow: Classify & File Inbound Docs with OpenAI + Gmail

Automatically classify and file inbound documents with OpenAI + Gmail—only when you need a human.

This n8n workflow receives documents via a token-protected webhook, checks OCR and file metadata quality, then uses OpenAI to classify and extract key fields. Every outcome is stored in an n8n Data Table, high-confidence results are filed logically, and Gmail alerts are sent when OCR or AI results require human review.

What this workflow does

• Secure intake: Accepts a POST request to an n8n webhook and verifies it using a shared intake token.
• Normalize + key documents: Standardizes the payload (filename, MIME type, IDs, OCR text), generates a deterministic document key, redacts PII for the model, and uses a temporary in-memory claim to reduce concurrent reprocessing.
• Duplicate protection: Checks the n8n Data Table for an existing row with the same document key scope and immediately returns a “duplicate” response when already processed.
• Dead-letter invalid inputs: If the item is permanently invalid (e.g., unsupported MIME type or malformed identifiers), it records a dead letter in the Data Table and responds with the dead-letter reason.
• OpenAI extraction + validation: For acceptable OCR quality, sends a bounded, redacted excerpt to OpenAI to extract metadata (type, party, date, amount, currency). The workflow validates returned JSON against strict rules and confidence thresholds.
• File vs. review: Fully validated, high-confidence results are recorded as Filed with a logical archive path; other cases become Needs Review and trigger the appropriate Gmail alert before the workflow responds.
• Scheduled review processing: Runs hourly to find “Needs Review” rows where no alerts have been sent (per the workflow logic).

Use cases

• Inbox-to-archive automation for customer documents (forms, statements, other inbound files) using OCR + AI classification.
• Invoice- or record-like processing where only confident extracts are filed automatically.
• Operational support queues: notify staff via Gmail when OCR quality or AI confidence is insufficient.

Technical details

• Nodes/stack: webhook, if, code, data table, gmail, sticky note.
• Core mechanisms: deterministic document keys, PII redaction for OpenAI, strict JSON validation, confidence thresholds, duplicate/dead-letter handling, and Gmail alerts for “Needs Review”.