Generate empathetic customer replies with Claude AI and auto-escalation
Needs changesNeeds changes. This template uses 1 thing(s) that n8n 3.0 removes: AI Agent v1 (old agent modes). 1 breaking item(s) found. Static check, last run 2026-10-19.
What to replace before upgrading
| Removed in 3.0 | Replace with |
|---|---|
| AI Agent v1 (old agent modes) AI Agent (Empathy)1 (v1.6) | Latest AI Agent node (Tools Agent). For SQL Agent, use a Postgres or MySQL tool with a recent AI Agent. |
Nodes used (12)
- Edit Fields (Set)
n8n-nodes-base.set - Code
n8n-nodes-base.code - AI Agent
@n8n/n8n-nodes-langchain.agent - Anthropic Chat Model
@n8n/n8n-nodes-langchain.lmChatAnthropic - Simple Memory
@n8n/n8n-nodes-langchain.memoryBufferWindow - Structured Output Parser
@n8n/n8n-nodes-langchain.outputParserStructured - If
n8n-nodes-base.if - Webhook
n8n-nodes-base.webhook - Schedule Trigger
n8n-nodes-base.scheduleTrigger - Slack
n8n-nodes-base.slack - Google Sheets
n8n-nodes-base.googleSheets - Manual Trigger
n8n-nodes-base.manualTrigger
What it does
🧠 Overview Generate empathetic, professional reply drafts for customer or user messages. The workflow detects sentiment, tone, and risk level, drafts a concise response, sanitizes PII/links/emojis, and auto-escalates risky or low-confidence cases to human review. --- ⚙️ How It Works 1. Input — Manual Test or Webhook Trigger 2. AI Agent (Empathy) — returns { sentiment, tone, reply, confidence, needshandover } 3. Post-Process & Sanitize — removes URLs/hashtags, masks PII, caps length 4. Risk & Handover Rules — checks confidence threshold, risk words, and negativity 5. Routing — auto-send safe replies or flag to Needs Review --- 🧩 Setup Instructions (3–5 min) - Open Set Config1 and adjust: - MAXLEN (default 600) - ADDFOLLOWUPQUESTION (true/false) - FORMALITY (auto | casual | polite) - EMOJIALLOWED (true/false), BLOCKLINKS (true/false) - RISKWORDS (e.g., refund, lawsuit, self-harm) - Connect Anthropic credential to Anthropic Chat Model - (Optional) Replace Manual Trigger with Webhook Trigger for real-time use > Tip: If you need to show literal angle brackets in messages, use backticks like <example> (no HTML entities needed). --- 📚 Use Cases 1) SaaS Billing Complaints - Input: “I was billed after canceling. This is unacceptable.” - Output: Calm, apologetic reply with refund steps; escalates if refund is in RISKWORDS or confidence < 0.45. 2) Product Bug Reports - Input: “Upload fails on large files since yesterday.” - Output: Acknowledges impact, requests logs, offers workaround; routes to auto-send if low risk and high confidence. 3) Delivery/Logistics Delays - Input: “My order is late again. Should I file a complaint?” - Output: Empathetic apology, ETA guidance, partial credit policy note; escalates if language indicates legal action. 4) Community Moderation / Abuse - Input: “Support is useless—you’re all scammers.” - Output: De-escalating, policy-aligned response; auto-flags due to negative sentiment + risk keyword match. 5) Safety / Self-harm Mentions - Input: “I feel like hurting myself if this isn’t fixed.” - Output: Immediate escalation, inserts approved resources; never auto-sends. --- 🚨 Auto-Escalation Rules (defaults) - Negative sentiment - Message matches any RISKWORDS - confidence < 0.45 - Mentions of legal, harassment, or self-harm context --- 🧪 Notes & Best Practices - 🔐 No hardcoded API keys — use n8n Credentials - 🧭 Tune thresholds and RISKWORDS to your org policy - 🧩 Works on self-hosted or cloud n8n - ✅ Treat outputs as drafts; ship after human/policy review --- 🔗 Resources - GitHub (template JSON): https://github.com/yskmtb0714/n8n-workflows/blob/main/empathy-reply-assistant.json