What should AI customer service answer in a small online store?

Which questions should AI answer?

Start with the questions your customers ask again and again. Read through your last month of messages, and you'll likely find the same short list:

• Shipping and delivery times. How long it takes, where you ship, and what it costs.
• Returns and exchanges. Your policy, in your words, every time.
• Hours and contact. When a person is around, and how to reach one.
• Payment. Which payment methods you accept.
• “Where is my order?” Store owners call it WISMO. AI can answer it well only if it can see the order's live status. Without that, it can only point to the tracking link.
• Product questions. Sizes, materials, or what's inside a gift box, if the AI knows your catalog.
• After hours. At night and on weekends, a customer still gets an answer, or at least knows when a person will reply.

What should still go to a person?

• Money. A refund, a credit, or a discount outside your policy. A decision that costs money needs someone who's responsible for it.
• Judgment. An exception to the rules, a bulk or corporate order, or a custom quote.
• Upset customers. A gift that arrived damaged, an order that went wrong, or a customer who's simply had enough. That's when people want a person, not a script.
• Changes to an order. A new address or a different item changes the order itself, so a person should make the change.
• Anything it doesn't know. If the answer isn't in what the AI can see, it should hand off, not guess. A confident wrong answer is worse than “Let me get a person.”

Why is a wrong answer worse than no answer?

Because your customer believes it, and you answer for it. In February 2024, a Canadian tribunal ruled on a case where Air Canada's website chatbot told a customer they could apply for a bereavement fare after their trip. The airline's own policy page said it didn't apply once travel was done. Air Canada argued it couldn't be held liable for what its chatbot said. The tribunal disagreed:

“

It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.

Moffatt v. Air Canada, 2024 BCCRT 149, British Columbia Civil Resolution Tribunal, Feb 14, 2024

The tribunal found negligent misrepresentation and ordered the airline to pay the customer the fare difference, with interest and fees. It's a Canadian small-claims decision, not US law, but the lesson travels: the chat on your site speaks for your store. So the rules for a good AI answer are easy to say and harder to build:

• Promise only what has already happened. “Your order is being packed” should mean someone started packing it, not just that the order exists.
• Say when it's stuck. If it's about to repeat itself, it should say so and offer a person.
• Answer from real data. From the order, the catalog, and your written policies, not from a guess.

What makes a good handoff to a person?

A handoff is where a chat can lose a customer. A good one has three parts:

1. 1The context travels. What the customer asked, what they want, and the order number, so nobody asks them to start over.
2. 2Someone knows they're waiting. A customer who asked for a person shouldn't wait unseen. An alert goes to the team.
3. 3The reply reaches them. If the customer has left the chat, the answer should still find them.

At Hasod, the online gift store in Israel where we built MOVA, when a customer has waited 5 minutes for a person, the team gets a text, between 8 AM and 11 PM Israel time. And if someone replies after the customer has left the chat, the reply follows them: a browser notification where they allowed one, an email if the chat has their address, or, as a last resort, a text.

What do the rules say about chatbots?

In California, a law in force since July 1, 2019 makes it unlawful to use a bot to talk with someone in California online with the intent to mislead them about its artificial identity, knowingly deceiving them in order to push a sale (or influence a vote). A business that discloses that it's a bot, clearly and conspicuously, isn't liable under that law (California Business and Professions Code, sections 17940 to 17943).

The simple way to stay clear is to say it plainly in the chat: this is an AI assistant. This isn't legal advice, so check the rules in the states where you sell.

How do you know it's working?

• Read the chats. Pick a few every week and read them start to finish. You'll catch a wrong answer before a customer complains about it.
• Count the handoffs. How many chats went to a person, and why. A topic that keeps coming back is a page your site is missing.
• Watch the repeats. A question the AI answers all the time is a sign the answer should be easier to find on your site.
• Check the waits. How long customers waited for a person after they asked for one.

What does AI customer service cost?

AI customer service is often priced per conversation it resolves: $0.99 per outcome at Intercom, on top of helpdesk seats, and about $1 per automated interaction at Gorgias, on top of a helpdesk plan, as published on Oct 18, 2026. The full picture is in how much an AI employee costs in 2026.

With MOVA, Maya is part of the base. At Basic, inside the $149/month base, she answers general questions in web chat, up to 300 chats a month. Premium, for $59/month on top of the base, adds answers about a specific order from its live status. Master, for $99/month, adds photos, a secure payment link through your payment processor where it allows it, and the handoff with a summary.

How does Maya handle this at MOVA?

Maya, MOVA's customer service agent, answers in web chat on your site from the day it goes live. What she can do depends on her level:

• Basic, in the base. Answers general questions in web chat on your site: hours, shipping, returns, and payment. Up to 300 chats a month.
• Premium. Answers questions about your store, including about a specific order, using the order's live status.
• Master. Reads a photo the customer sends before answering, and hands the chat to a person with a summary and a draft reply, by email, and by text where your provider allows it.

At Hasod, Maya says an order is being packed only after someone has started packing it. When she's about to repeat herself, she tells the customer, offers a person, and flags the chat for the team.