Skip to main content
Email routing combines a service trigger with deterministic filters, optional AI classification, and one branch per destination. A typical workflow normalizes each new message, classifies it or applies fixed rules, chooses a branch, then labels, drafts, forwards, replies, or notifies as appropriate.

​
Start with the narrowest trigger

Use a Gmail or Outlook new-email trigger and configure its built-in filters when possible. Trigger filters avoid creating runs for messages the automation will never process. Remember that subject matching can be case-sensitive. Test with the exact capitalization and message shapes that arrive in the real mailbox.

​
Choose deterministic or AI routing

Use deterministic text and logic actions when the rule can be stated exactly:
  • sender domain equals a known customer or vendor,
  • subject starts with a fixed prefix,
  • a label is present,
  • body contains a specific identifier.
Use OpenAI Chat with structured output when the distinction depends on meaning rather than a fixed phrase. Keep the schema small:
Put hard safety rules before AI classification. For example, ignore your own automated senders and known newsletters before paying to classify the remaining messages.

​
Build the branches

Connect route to a case-selection action or compare it in separate branches:
  • sales can forward to the sales queue.
  • support can add a label and draft a reply.
  • billing can notify the billing team with the summary.
  • ignore can end without a side effect.
Use Conditional on side-effect actions so only the matching branch runs. Independent preparation branches can still run concurrently.

​
Avoid mail loops

A workflow that sends, forwards, or labels mail can trigger another email-based workflow—including itself. Filter automated senders, dedicated labels, or recognizable headers so generated messages do not loop.
Before enabling:
  1. Send representative messages for every route.
  2. Confirm the trigger data contains the body and attachments you expect.
  3. Inspect drafts before switching to automatic replies.
  4. Test the fallback branch for ambiguous messages.
  5. Review Run History for node errors even when a run status says completed.
See Service event triggers for account and event lifecycle behavior.