Support automation covers a range, from sorting tickets to a bot that talks to customers directly. The risk rises with each step toward talking to customers on your behalf.
The examples here are public ones from tool makers' pages. None is ours, no client is named and we claim no result.
What can you automate in customer service?
Zapier's public template library (read 12 November 2026) lists patterns across the range: 'Classify and route support tickets in Zendesk', 'Triage support tickets into action items', 'Manage FAQs and requests from Slack with AI' and 'Resolve support tickets with guarded AI orchestration'. Its AI page describes agents that take action across tools, with examples such as routing tickets and handling requests, and a customer-facing chatbot.
Anthropic's engineering post Building effective agents (19 December 2024) names customer support as one of two particularly promising applications for agents, while advising that you find the simplest solution possible and add complexity only when needed.
Which steps are safest to start with?
Order them by how much a mistake costs, and start at the low end.
- Sorting and routing: label each request by topic and urgency and send it to the right queue. A person still replies.
- Summaries: give an agent a short summary of a long thread, with the original one click away.
- Drafts for approval: write a first reply from your own policy text, which a person edits and sends.
- Status answers: order or booking status, read from your own system, in a fixed format.
- Unsupervised answers: a bot replying to open questions on its own. Reach this step last, if at all.
What does the law say about chatbot answers?
The strongest public example is Canadian. In February 2024 the British Columbia Civil Resolution Tribunal decided Moffatt v. Air Canada. A customer asked the airline's website chatbot about bereavement fares and was told he could claim the discount after travel, which contradicted a page the chatbot linked. The tribunal rejected Air Canada's argument that the chatbot was a separate entity and held the airline responsible for all the information on its website, awarding about $650. Torkin Manes' summary (7 March 2024) says companies remain liable for what their AI tools tell customers and advises training systems for accuracy, internal policies and clear warnings about the limits of AI tools. It was a ruling of a tribunal, which is not a court.
In the United States, the FTC said in September 2024 that there is no AI exemption from the laws on the books, and that using AI tools to trick, mislead or defraud people is illegal. Neither source is legal advice for your case; ask a lawyer about your own rules.
How do you keep customers from getting bad answers?
- A fixed source of truth. Let the bot answer only from your own policy and product text, kept current, and have it say when it does not know.
- A visible route to a person. Offer it in every chat, in plain words, where visitors can see it.
- Honest labeling. Say that the customer is talking to an automated assistant, and what it can and cannot do.
- Limits on action. Do not let the bot give refunds, change contracts or make promises. A person does that.
- Review. Read a sample of conversations every week, and fix the source text when the bot gets something wrong.
What about security?
A bot that reads customer messages can be manipulated by them. OWASP's prompt injection page (LLM01:2025) says user prompts can alter a model's behavior in unintended ways, and lists mitigations: restrict the model's access privileges to the minimum necessary, use human-in-the-loop controls for privileged operations, and separate untrusted content. It says it is unclear that there are fool-proof methods of prevention. So give a support bot read access to what it needs and nothing else.
How do you measure it?
Measure what customers experience as well as how many chats the bot handled. Track resolution (did the customer's problem end?), the share of chats handed to a person, complaints about the bot, repeat contacts on the same issue, and time to a first useful answer. Read a sample of transcripts. If complaints or repeat contacts rise, narrow what the bot may do. Our guide to how to automate business processes with AI gives the measuring steps, and the guide to an AI chatbot for your website covers the website case.
When should a person always step in?
Complaints and anger, anything involving money, health, safety, legal rights or cancellations, requests from vulnerable customers, anything the bot flags as uncertain, and any request the customer repeats. Build these triggers in from the start, and make the handoff carry the conversation so the customer does not repeat themselves.
Conclusion
Automate support from the low-risk end: sort, summarize and draft before you let a bot answer alone. Keep a fixed source of truth, an easy route to a person and regular review, because the business is responsible for what its automation says.
PromotEdge International is an international strategy, technology, marketing and design company, and our delivery team is in India. See our AI chatbots, AI agents and AI automation pages.
Facts we rely on: see our company facts for what we state about ourselves and where each fact comes from.