AI Customer Service Checklist for the Holiday Rush

Holiday weeks bring a flood of the same questions: where is my order, are you open on the 24th, can I still return this in January. AI customer service can answer those at any hour, which is why so many businesses switch on a chatbot in November. The ones that regret it usually skipped the same few steps. This checklist covers them, in the order we set them up.

Decide what AI customer service is allowed to answer

Sort every type of question into three groups before you write a single instruction:

1. Answer directly. Hours, locations, shipping cut-off dates, return windows, order status from your system, menu or service details.
2. Collect and route. Catering inquiries, quotes, appointment requests, new client intake. The AI gathers the details and hands them to a person.
3. Never answer. Refunds outside policy, disputes, complaints about staff, anything medical, legal, tax or financial, and anything involving another person's account.

The third group matters most. The CFPB's 2023 report on chatbots in consumer finance found that bots and highly scripted systems can fail to recognize when a customer is raising a dispute, because only certain words trigger the right path. Route your "never" list to a person by intent instead of by keyword: "I was charged twice," "this is wrong" and "I want my money back" all mean the same thing.

Ground it in your written policies

A chatbot that improvises policy creates obligations you did not intend. In Moffatt v. Air Canada, decided in February 2024, an airline chatbot told a customer he could apply for a bereavement fare retroactively within 90 days. The airline's actual policy did not allow that. Air Canada argued the chatbot was responsible for its own actions; the British Columbia Civil Resolution Tribunal rejected that, saying the chatbot is still part of the company's website and the company is responsible for all the information on it. The customer was awarded the fare difference plus interest and fees.

The practical steps:

• Write the holiday policies down first: holiday hours, shipping cut-offs, extended return windows, gift receipt rules, delivery areas.
• Give the AI only those documents as its source, and instruct it to say "I'm not sure, let me get a person" when the answer is not in them. Our post on RAG chatbots and knowledge bases explains how this grounding works.
• Date every policy and remove last year's versions so the bot cannot quote them.
• Make the site and the bot agree. If your returns page says 30 days and the bot says 60, the customer will hold you to the better one.

Always offer a way to a human

The CFPB report describes customers trapped in what it calls "doom loops." One complainant described running into "loop after loop of the same questions"; another called it "an endless loop with no way out."

• "Talk to a person" works at every step, typed in any reasonable wording.
• The bot states when people are available and what happens outside those hours: a callback, a ticket number, an email reply time.
• The handoff carries the full conversation, so the customer never repeats themselves.
• Someone owns the handoff queue every day of the rush, including weekends.

Protect customer data

The same CFPB report flagged privacy and security risks in chat logs. Before launch:

• Tell customers not to type card numbers or passwords, and mask them if they do.
• Verify identity before the bot shares anything about an order or account.
• Know where transcripts are stored, who can read them and how long they are kept.
• For home care agencies, healthcare software and other businesses handling health information, check whether your chat vendor will sign the agreements your privacy obligations require. Our guide to using AI with client data covers this in more depth.

Be honest about what it is

Tell customers they are talking to an AI assistant, and do not give it a human name and photo that suggests otherwise. Be just as careful in your marketing. When the FTC announced its Operation AI Comply enforcement sweep in September 2024, then-Chair Lina Khan said there is "no AI exemption from the laws on the books." Claims like "instant answers to everything" or "as good as our best agent" are claims you need to be able to back up.

Test it before the rush

Testing in the first week of December is too late. In early November:

1. Pull real questions from last December's inbox, DMs and call notes. Aim for a few dozen covering every category.
2. Run each one and mark the answer correct, wrong or should-have-handed-off.
3. Try to break it: ask for a refund outside policy, ask about another customer's order, ask a medical or tax question, write in another language, type angrily.
4. Fix the source documents as well as the prompt, then run the whole set again.
5. Set up a daily review of a sample of transcripts during the rush, and a way to switch the bot to "collect and route only" in one click if something goes wrong.

Set the pass mark before you run the test, so nobody argues with the result afterwards. A rule we suggest: with a set of, say, 40 questions, any wrong answer on a "never" topic blocks the launch outright, and every wrong answer in the "answer directly" group has to be traced to a missing or unclear line in a policy document and fixed there. A handoff the bot did not strictly need is a pass. A confident wrong answer is always a fail.

The checklist to copy

1. Questions sorted into answer, route and never.
2. "Never" topics detected by intent and sent to a person.
3. Holiday policies written, dated and loaded as the only source.
4. Website and bot answers checked against each other.
5. Human handoff available at every step, with hours and response times stated.
6. Full transcript passed on at handoff; a named owner for the queue.
7. Card numbers and passwords masked; identity checked before account details.
8. Transcript storage and retention known; vendor agreements in place.
9. The bot identifies itself as AI; marketing claims are ones you can support.
10. Real-question test set run, broken, fixed and rerun.
11. Daily transcript review and a one-click fallback.

How we set it up

At Pinn.Media we build AI assistants into the Pinn.Media CRM and our other products with this structure from the start: grounded answers, clear handoff into the same inbox your team uses, and transcripts you can review. If you want an assistant ready before the rush, book a discovery call at /contact#book.