Skip to main content
Own toolchain · 01

n8n workflows with LLM classification

Incoming webhooks are validated, classified, stored, answered: all in a single n8n workflow with an error branch.

one workflowper integration instead of a separate service
WebhookPOST /lead{ }Validate✦ClaudeclassifyCRMappend: rowGGeminireplyGmailsend: replysuccess{ }Error: data{ }Error: CRMTelegram alerterror

Pain

Classic integrations cost weeks: a separate service per connection, manual error handling, no overview of running flows. LLMs add another layer that doesn't fit cleanly into hard-wired pipelines.

Solution

  • Receive every incoming event through a single webhook endpoint.
  • Validate the payload structurally (Zod-style JSON-schema checks) before any LLM is called.
  • Let Claude do the classification (heat score, intent, spam detection) and store the result in the CRM.
  • Let Gemini draft a personalized reply that goes out via Gmail.
  • Mirror an error branch under the success path: any node can fall through to a failure path that ends in a Telegram alert, no silent failures.

How it works

  1. 01
    Webhook (POST /lead)
    Entry point for forms, external tools, mail filters.
  2. 02
    Validate ({ })
    Structural check: missing fields and wrong types are rejected immediately.
  3. 03
    Claude · classify
    LLM scores heat (Hot/Warm/Cold) and assigns an intent tag.
  4. 04
    Save to CRM
    Lead is written into Sheets or Notion with classification.
  5. 05
    Gemini · draft reply
    Personalized reply draft based on classification plus lead context.
  6. 06
    Send via Gmail
    Reply goes out, CRM entry is flipped to 'contacted'.
  7. 07
    Error branch
    On node failure, the parallel path takes over, logging into a Telegram alert.

Outcome

New integrations are built as a workflow instead of a separate service. Workflows run in production; failures become visible instead of disappearing.

Why this stack

n8n
Visual workflow editor with production features (retry, logging, versioning).
Claude
Classification with traceable reasoning, relevant for CRM audits. Internal funnel: for clients we swap in an open-source model on our own GPU or a provider their data protection team approves.
Gemini
Fast and cost-effective generation of personalized reply drafts. For clients the same provider choice applies as for classification.
Webhooks
Universal adapter: almost any tool can POST.
More about our AI & Automation services

Related use cases

Send
→
HOTMüller GmbH
WARMSchmidt KG
COLDSpam filter
Own toolchain · 04

Website with built-in lead tracking

Contact form with validation, automatic email notification, CRM sync and lead classification at intake. No more lost enquiries.

every leadwith an owner, reminder and escalation deadline
Next.jsResendCRM sync
Learn more→
Telegram voice message on a smartphone being transcribed, classified and stored as a task list on EU server infrastructure
Own toolchain · 03

Task dashboard with voice input

Telegram voice note → Whisper transcript → automatically classified tasks in the backlog. Cron jobs surface reminders.

voice instead of typingclassified tasks at no extra effort
Whisper (self-hosted)Telegram Botn8nLlama / Mistral
Learn more→