n8n template · AI Inbox Pack

Detect urgent emails with an AI score in n8n

This workflow exposes a scoring webhook: send it any email (subject, sender, body) and an LLM assigns an urgency score from 1 to 5, justified and paired with a suggested action. Above the threshold, Slack rings; below it, the email quietly waits for the daily digest.

$ n8n import:workflow --input=priorisation-urgence-emails.json
✔ Import successful — 9 nodes, valid connections

The problem it solves

The real cost of email isn't reading it — it's being interrupted by messages that didn't deserve it. Without prioritization, every notification is equal: the critical customer outage and the sales follow-up arrive with the same ding.

Give a model an explicit urgency rubric (blocking incident = 5, prospecting = 1) and you get reproducible, contextual triage. The webhook architecture makes the scoring reusable: shared inbox, contact form, CRM — anything that can send a POST can ask for a score.

How the workflow works, node by node

Node names are in French, exactly as they appear in the imported workflow — the logic is language-agnostic.

1. 01

   Email reçu (Webhook)

   Webhook

   HTTP POST entry point (path flowkit-email-score): it receives JSON with subject, from and body. The responseNode mode lets the workflow return the score to the caller.

2. 02

   Extraire les champs

   Set (fields)

   Normalizes the payload and truncates the body to 4,000 characters: clean input for the LLM, bounded token cost.

3. 03

   Scorer l'urgence

   LLM chain

   Applies a detailed urgency rubric (from 1, newsletter, to 5, blocking incident), with instructions not to be fooled by marketing 'URGENT' subject lines.

4. 04

   Modèle OpenAI

   OpenAI model

   gpt-4o-mini as an ai_languageModel sub-node: fast and cheap, ideal for scoring called on every email.

5. 05

   Score structuré

   Structured output

   Guarantees strict JSON output: integer score 1–5, one-sentence reason, suggested action starting with a verb. The next IF compares numbers, not text.

6. 06

   Urgent ?

   IF (condition)

   Compares the score to the threshold (4 by default). The only setting to touch to make the system more or less sensitive.

7. 07

   Alerte Slack

   Slack

   'True' branch: immediate alert with score, reason and suggested action — your team knows what to do before even opening the email.

8. 08

   Mettre en file pour le digest

   Supabase

   'False' branch: inserts the scored email into the emails_queue table with pending status, where the daily digest workflow will pick it up.

9. 09

   Répondre au webhook

   Webhook response

   Returns the score to the caller as JSON: the system that sent the POST can react too (label, CRM routing…).

What you need to run it

• ✔n8n ≥ 1.60 with the LangChain nodes
• ✔An OpenAI API key
• ✔A Slack bot (chat:write scope) for alerts
• ✔A Supabase project with the emails_queue table (SQL provided in the pack guide)
• ✔A system able to send emails to the webhook (forwarding rule, script, another n8n workflow)

Customization ideas

• →Adapt the rubric to your business: 'email from a client under contract = 4 minimum' is a one-line prompt change
• →Replace Slack with Teams, Telegram or SMS depending on your on-call channel
• →Add a 'department' field to the structured schema to route to the right team