Get a Scope
Free n8n template

AI shared inbox triage n8n template

What it does

An importable n8n workflow that polls a shared Gmail inbox every minute, drops auto-replies, bounces and internal mail in code, has Claude pick one of seven buckets as strict JSON, and applies the matching Gmail label. Sales inquiries become HubSpot contacts with a Slack alert. You need n8n, Gmail, an Anthropic API key, a HubSpot private app and Slack.

What this workflow does

  1. New email in shared inbox (Gmail Trigger) polls for inbox messages that do not yet carry ai-triaged or ai-unsure.
  2. Pre-filter (Code) skips auto-replies, bulk mail, bounces and senders from your own domains before any model call, and trims quoted history so Claude reads only the newest message.
  3. Classify with Claude (HTTP Request) calls the Anthropic Messages API with a strict JSON schema: one bucket, a confidence score, a reason, and contact name, company and phone when they appear.
  4. Parse and route (Code) handles refusals, truncated answers and API errors, rejects unknown buckets, and marks anything under the confidence threshold as unsure.
  5. Confident? (If) splits sorted mail from mail that needs a person.
  6. Apply bucket labels (Gmail) adds the bucket label and ai-triaged.
  7. Sales inquiry? (If) sends new leads to Create HubSpot contact, which creates or updates the contact by email, and Notify sales in Slack with a Gmail link.
  8. Label as unsure and Post inbox exception handle everything else, so nothing is silently dropped.

Before you import

  • A current n8n version and access to the shared Gmail account through OAuth.
  • An Anthropic API key. The template uses claude-opus-5-5; claude-haiku-5-5 is cheaper for high-volume sorting.
  • A HubSpot private app token with contact read and write scopes. Choose “Service Key” authentication on the HubSpot node.
  • A Slack bot added to the sales and exceptions channels.

Setup

  1. Import the JSON and connect the Gmail, Anthropic, HubSpot and Slack credentials.
  2. In Gmail, create one label per bucket plus ai-triaged and ai-unsure.
  3. In Parse and route, replace each REPLACE_WITH_LABEL_ID_... value with the Gmail label ID (it looks like Label_123, not the label name).
  4. In Pre-filter, replace REPLACE_WITH_YOUR_DOMAIN.com with your domains, and edit the bucket definitions to match how your team thinks about the inbox.
  5. Replace REPLACE_WITH_SLACK_SALES_CHANNEL_ID and REPLACE_WITH_SLACK_EXCEPTIONS_CHANNEL_ID.
  6. Tune MIN_CONFIDENCE (0.7 by default) after your first week of results.

Test it

Send the shared inbox four emails from a personal address: a pricing question, a support request, a fake invoice and a vague one-liner. Run the workflow manually and check labels in Gmail. The pricing question should create a HubSpot contact and a Slack alert. The one-liner should land in ai-unsure. An out-of-office reply should never reach the model.

Make it production-ready

  • Run label-only for a week by disconnecting the HubSpot node, as the guide recommends, then turn actions on bucket by bucket.
  • Add an n8n error workflow so Gmail or HubSpot failures alert someone.
  • The Claude call retries three times. Check HubSpot’s API limits for your plan if the inbox is busy.
  • Add a forward step for the billing bucket to your AP address once labels prove reliable.
  • Review the ai-unsure label daily. Repeated unsure messages usually mean a bucket definition needs rewriting.

Read the full guide on AI shared inbox triage for bucket design and mailbox access, or get help from Lumeor with Outlook or a custom routing table.

Next steps

01 Browse free n8n templates Fifty starter workflows you can import today. No email required. 02 Take the 5-minute AI readiness assessment Get a score and a recommended first project in your browser. 03 Get a written scope and estimate Tell us what you want to automate. We reply within two business days.