## TL;DR

Email is the messiest intake channel: forwards, duplicates, zipped attachments, and non-invoice content. Ingest by extracting attachments (including from zips), deduping on content hash plus sender, classifying invoice vs non-invoice, and threading replies to the original invoice. Treat the inbox as untrusted input, not a clean feed.

## Steps

1. Extract attachments, recursing into zips.
   Expected: All candidate documents.
2. Dedupe on content hash and sender.
   Expected: No double intake.
3. Classify invoice vs statement vs correspondence.
   Expected: Right handling per type.
4. Thread replies to existing invoices.
   Expected: Context preserved.
5. Quarantine suspicious attachments.
   Expected: Security.

## When to use

- Email AP intake
- Dedup design
- Multi-channel intake

## When not to use

- Portal intake
- EDI intake
- Invoice processing itself

## Compatibility

Email APIs (Gmail, Outlook); ERP-agnostic downstream.

## Variant phrasings

### AP email ingestion

### invoice inbox automation

### email attachment dedup invoices

## Root cause

The same invoice arrives by email multiple times through forwards and CCs. Without content-hash dedup, each copy becomes a payable.

## Edge cases

- Password-protected zips need a sender protocol
- Embedded images in the email body can be invoices too
- Phishing via invoice email is common; verify senders

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_ci9v1qagsoAlXbdL1beKRQ