How to extract key information from emails

Two layers of information live in every email. The headers and metadata (from, to, subject, dates, attachments, message-id) are trivially structured. The body is the hard part — it's prose that may contain order numbers, dates, dollar amounts, action items, signatures, and quoted history from a long thread.

Headers and structure: easy

Python's stdlib email.parser handles the headers cleanly:

from email import message_from_bytes from email.policy import default with open("message.eml", "rb") as f: msg = message_from_bytes(f.read(), policy=default) print(msg["From"], msg["Subject"], msg["Date"]) for part in msg.iter_attachments(): print(part.get_filename())

For Gmail or Outlook: hit the API directly (Gmail API messages.get with format=full, Microsoft Graph /me/messages). Both return parsed headers and the body in HTML and plain text.

Body extraction: harder than it looks

Three problems compound:

• Replies and forwards quote prior messages, often with broken delimiters.
• Signatures vary per sender and per device.
• HTML emails have inline styles, tracking pixels, and footer disclaimers that aren't part of the actual content.

The talon library (open-sourced from Mailgun) handles signature and quote extraction in Python — not perfect but the best free option. talon.signature.extract_signature and talon.quotations.extract_from work well enough for most flows.

Pulling structured fields out of the body

Once you have the clean body, the typical needs are:

• Order/invoice/ticket numbers — regex on known prefixes works for in-house systems.
• Dates and times — dateparser (Python) handles natural-language dates in 200+ locales.
• Dollar amounts — regex catches them; libraries like price-parser handle currency edge cases.
• Action items — best done with an LLM, not regex.
• Signatures and contact details — talon plus a contact-extractor like phonenumbers and email-validator.

When the email body is the document

Common patterns where the body itself contains the unstructured data you want:

• Customer support: extract intent, sentiment, urgency for routing.
• Sales: extract company name, contact role, deal stage signals from prospect replies.
• Operations: extract shipping notifications, status updates, and exception alerts from system emails.

An LLM with a target schema beats handwritten regex for all of these once volume is more than a few dozen messages a day.

When the value is in attachments

If the email is a transport for an invoice PDF or a signed contract, the email body matters less than the attachment. ExtractFox's API takes file uploads or URLs — wire your inbox processor (Make, Zapier, custom) to fetch attachments matching a rule, hand them to the right extractor, drop the results in your accounting or CRM system. The email becomes the trigger; the document becomes the data.