Customer support: what to automate and which tools to use
Field: Sales and customer work. Role hub. Checked: 2026-08-18. Tools change over time. Sources for this review are at the bottom of the page.
At a glance
Bots and AI take over standard answers. Complex cases, empathy, and retention become more valuable because they still need a person. The main risk is a bot giving a confident but wrong answer to a customer.
What the data says (GDPval, Anthropic Economic Index, O*NET)
- Capability (GDPval): a typical work product is a "customer support conversation." AI performs well on standard cases.
- Use (AEI): retail and service work show high automation. Bots handle frequent questions.
- The honest scale: support and IT are the most common functions for agents, yet McKinsey (The state of AI in 2025, published Nov 5, 2025) found only 23% of organizations scaling an agentic system anywhere, and no more than 10% inside any single function. The instructive counter-case: Klarna brought people back into support in 2025 to fix quality.
- Mode: strong reduction in entry-level tasks. Bots take the standard support line. Complex cases, retention, and empathy remain. See the full data review and the role AI strategy.
What to automate first
- Answers to common questions. Answer from a trusted knowledge base, not from the model's memory.
- Ticket classification and routing. Sort requests automatically by topic and priority.
- Response drafts. An agent edits the draft before sending it.
- Ticket and history summaries. Bring an agent into the context quickly.
- Churn detection in customer success. Flag customers who may leave.
Task review: what AI can do and what you must check
| Task (O*NET) | Give to AI: method or tool | Keep or verify yourself | Prompt to start |
|---|---|---|---|
| Common questions | Answer from the knowledge base | Accuracy before sending | |
| Ticket classification | Sort by topic and priority | Disputed routing | Identify the topic and priority of this request |
| Response drafts | Suggested answer | Facts and tone | Give me two tone options. I will check the facts myself |
| Ticket summary | Short history | Customer context | Summarize the request history and the core problem |
| Complex or conflict case | Do not delegate | A person, empathy, and the decision |
Use a bot for standard questions from a trusted knowledge base. Send complex and retention cases to a person. Never send unchecked answers.
Tools by use case
- ChatGPT or Claude for complex or unusual responses.
- Zendesk AI, Embrace.ai for agents tuned to a brand, and Pylon for support and customer success.
One practical playbook
Use a bot for common questions and a person for complex cases
- Connect AI answers to the knowledge base, so the system answers from a source instead of inventing one.
- Set the routing rule: a standard case gets an automatic draft, while a complex case goes to a person.
- Human check: review drafts before sending. Route emotionally difficult cases to a person at once.
- Use the saved time for retention and complex cases.
- Output: the queue is smaller, and customer trust remains intact.
Where not to use AI
Red flags
- Unchecked answers: a bot may invent terms or instructions and lose the customer's trust.
- Customer personal data: remove identifying details and do not send it to outside services.
- Emotionally difficult cases: use a person, not a template.
Prompt patterns: weak and better
| Weak | Better |
|---|---|
Reply to the customer |
Give me two tone options. I will check the facts myself |
Solve the problem |
Give me three possible causes of this complaint and a way to check each one |
Where to move your effort
Move time from standard work into retention, conflict, complex decisions, and empathy. Human work becomes more valuable. See Chapter 7.
The levels ladder in this role
The five levels from Chapter 5, in the language of this profession. Mark where you stand in your main tasks this week.
| Level | What it looks like here |
|---|---|
| 1. AI user | I answer the customer faster with an AI suggestion |
| 2. Validator | I check the bot's answer before the customer sees it. A confidently wrong answer costs more than a slow one |
| 3. Orchestrator | I built the line where the bot takes the routine and I take the hard cases and escalations |
| 4. Outcome owner | I own customer retention, not the number of tickets closed |
| 5. System builder | I set up the knowledge base and escalation rules so the line runs without manual control |
Where people usually get stuck. The routine tier moves to bots entirely, so level 1 shrinks fastest here. Hard cases, conflict, and retention are what gain value.
Next: my level of usefulness → a plan for one level up.
Ready-made skills and plugins for this role
You can turn repeatable procedures such as answering from a knowledge base, routing tickets, and writing an escalation summary into a portable skill, or use an existing one. Browse skill banks, then see turn a workflow into a skill to build your own.
Where to go next
Map your week → run the integrated profession audit → make a 90-day plan. Or open the workbook.
Sources for this review: OpenAI GDPval (2025): https://openai.com/index/gdpval/ · Anthropic Economic Index: https://www.anthropic.com/economic-index · Stanford AI Index (2025): https://hai.stanford.edu/ai-index · McKinsey, "The State of AI": https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai. Plus the book, Chapters 2, 3, and 7. Version: 2026-08-18.