Support Triage

1. The Problem

Customer support queries arrive via email, website chat, WhatsApp, Instagram DMs, and phone — simultaneously and unpredictably. Triaging them manually across all channels means either dedicating a staff member full-time to monitoring, or accepting the reality that some channels are slower than others and customers who use the slow ones get a worse experience. In a competitive market, response time is a product feature. The business that replies in 10 minutes wins the customer the business that replies in 3 hours is still considering.

2. Integration & Webhook Setup

Follow the exact steps below to configure and deploy this automation inside your OpenClaw workspace.

1. 1Create a new agent named `support-triage`. Channel connections required: email (Gmail or Outlook), website chat (Intercom, Drift, or Crisp all supported), WhatsApp (via Twilio Business API), and Instagram (Meta Business API — requires Facebook Business account verification, typically 2–3 business days).
2. 2Build your Knowledge Base in the Context Payload: add your 30–50 most common support questions and their correct answers. The more complete this is, the higher the agent's first-contact resolution rate — aiming for 70–80% without human involvement.
3. 3Configure your `ESCALATION_RULES`: define which query types always go to a human (refund requests above a threshold, legal complaints, safeguarding concerns, VIP accounts). Set your escalation notification channel (Slack or SMS to a named team member).
4. 4Connect your fulfilment or order management system (Shopify, WooCommerce, or custom via API) so the agent can respond to 'Where is my order?' queries with live tracking data rather than a holding message.
5. 5Set your `RESPONSE_PERSONA`: configure the agent's name (many clients prefer a named persona like 'Aria from Brightleaf Support'), tone, and sign-off. The agent never claims to be human if directly and sincerely asked.

3. The Context Payload (context.json)

Save this file as: .openclaw/agents/support-triage/context.json

{
  "automation_id": "14",
  "title": "Support Triage",
  "level": 2,
  "tier": "Agency Foundation",
  "setup_time": "3–4 hours",
  "estimated_api_cost": "~$15–$35/mo",
  "client_price_range": "$500–$2,000",
  "agents": [
    {
      "role": "orchestrator",
      "model": "claude-3-5-sonnet",
      "temperature": 0.2,
      "max_tokens": 4096
    }
  ],
  "memory": "session",
  "output_format": "structured_json",
  "human_review_gate": true,
  "documentation_standard": "required"
}

4. Execution Commands

Run these commands from your openclaw-workshop/ directory to validate, test, and schedule this automation. Commands are taken directly from The OpenClaw Income Engine, Appendix F.

# ── STEP 1: Test all channel connections ──
$ openclaw run support-triage --test --verbose
$ 
# ── STEP 2: Test knowledge base matching with sample queries ──
$ openclaw run support-triage --test-kb --queries "Where is my order?,Do you offer refunds?"
$ 
# ── STEP 3: Run in classify-only mode (no responses sent) ──
$ openclaw run support-triage --mode classify-only
$ 
# ── STEP 4: Run live on one channel first (email only) ──
$ openclaw run support-triage --channel email
$ 
# ── STEP 5: Activate all channels ──
$ openclaw webhook activate support-triage --all-channels
$ 
# ── Add a new FAQ entry without redeploying ──
$ openclaw config edit support-triage
# Add new object to knowledge_base array
$ 
# ── Review escalation rate over last 7 days ──
$ openclaw run support-triage --metrics --days 7
$ 
# ── Test a specific message against the routing logic ──
$ openclaw run support-triage --test-message "I want a refund, this is outrageous"