AI Customer Support & Sales

High-frequency pain points in customer service and sales: routing/prioritization, standard responses, record-keeping, and follow-up. AI can "classify first, then respond" and write records back to your systems.

1) Ticket / Lead Routing

• Have AI do "intent classification + priority + escalation criteria":

You are a customer service routing assistant. Based on the ticket text, output:
- Intent classification (billing / technical / usage guidance / complaint / other)
- Urgency (high / medium / low) with reasoning
- Whether human escalation is needed (yes / no, explain why)
- Suggested response template ID

• Output classification as JSON for automated write-back to ticket systems or Slack/Feishu notifications.

2) Standard Responses & Multi-Turn Follow-up

• Have AI provide "first-response template + follow-up questions + follow-up reminder timing" based on classification.
• For sales scenarios, request "objection handling scripts": List 5 likely customer objections and corresponding responses.
• For sensitive scenarios (compensation / compliance), require AI to flag "paragraphs requiring human confirmation" to avoid unauthorized promises.

3) CRM / Notes Auto-Sync

• Prompt example:

You are a CRM recorder. Input is a customer call/email summary.
Output fields: customer name / contact person / needs / budget / decision maker / timeline / next action / owner / risks.
Format: table, plus a 150-word summary.

• For sales pipeline, ask AI for "stage assessment + next step suggestion + potential blockers."

4) FAQ / Knowledge Base Maintenance

• Have AI extract high-frequency questions from chats/tickets and generate FAQ entries: question / standard answer / scope / last updated.
• For multilingual customers, require AI to generate bilingual answers and flag "localization-sensitive terms / statements requiring legal review."

5) Quality & Compliance

• Define "banned words / banned promises list": prices, compensation, legal terms must have human approval.
• Require AI to output "answer source / unconfirmed info" for human review.
• Before sending externally, have AI run an "emotion and politeness check + tone consistency" pass.

6) Practice

Take a real ticket or sales email. Have AI output: routing result + response draft + next action + CRM write-back fields. Check if it meets your service standards, then adjust the prompt.

---

7) Worked Example: Route a Billing Complaint Safely

A customer writes: “We were charged again, and it is blocking month-end reconciliation. If nobody fixes it today, we will not renew.”

Do not ask AI to reply immediately. Separate classification, evidence, drafting, approval, and write-back:

Analyse only. Do not reply to the customer.
Return JSON:
{
  "intent": "",
  "urgency": "low|medium|high",
  "customer_impact": "",
  "confirmed_facts": [],
  "missing_facts": [],
  "human_owner": "",
  "reply_must_not_promise": []
}
Use unknown for missing information. Do not infer cause or compensation.

After a human verifies billing records, draft the reply:

Use only confirmed facts.
1. Acknowledge the specific impact without accepting unconfirmed liability
2. State the action already taken
3. Give the next update time, not an invented resolution time
4. Mark anything requiring finance approval as [CONFIRM]

Minimum CRM Write-Back

8) Response Quality Scorecard

Score each item from 0–2:

• Facts: only confirmed information is used
• Empathy: the actual impact is acknowledged, not a generic apology
• Action: an owner and next update time are present
• Boundaries: no unapproved price, compensation, or legal conclusion
• Usability: the customer knows what happens next

Do not auto-send below 8. Finance, escalated complaints, legal, and security cases require human approval.

9) Failure Modes and Fixes

10) Chapter Deliverable

Build a reusable Service Case Pack: classification schema, escalation rules, response scorecard, and CRM write-back template. Continue to AI Workflows and Automation after the small process works reliably.