Holiday Returns Rush: An AI Return Agent for Burnaby

Boxing Day is coming, and so is the week after it. That's when the gift receipts come out, the phone keeps ringing and the same five questions land in your inbox over and over. Can I return this without a receipt? Is it store credit or a refund? I bought it online, can I bring it into the Burnaby location?

None of these questions is hard. There are just a lot of them, and they arrive when your staff are already stretched. That's the case for a small AI agent that answers return questions from your actual policy, around the clock. It's also a case for being careful, because a wrong answer here can cost you real money.

Why BC shops get to write their own rules

Start with the legal ground, because it shapes everything the agent should say. According to Consumer Protection BC, "In BC, there are no laws that deal directly with refund, exchange and return policies." Businesses set their own terms. Some shops do store credit only, some do exchanges, some do full refunds on unopened items, and some say all sales are final.

There's one exception worth knowing. The same Consumer Protection BC article says that if an online order hasn't arrived within 30 days of the expected delivery date, the customer can cancel and get a refund no matter what the store's policy says. That article dates from January 2020, so confirm the current rules with Consumer Protection BC before you bake them into anything.

So the policy is yours. That's freedom, but it also means the only source of truth is whatever you wrote down. If it's vague, or lives in one manager's head, no AI tool will fix that for you.

The Air Canada lesson, decided right here in BC

The most useful case for any shop thinking about a customer facing chatbot was decided by BC's own Civil Resolution Tribunal. In Moffatt v. Air Canada, 2024 BCCRT 149, issued February 14, 2024, a customer asked Air Canada's website chatbot about bereavement fares. The chatbot said the discount could be claimed after travel. The airline's actual policy, on another page of the same website, said it couldn't.

Air Canada argued it couldn't be held liable for what the chatbot said. The tribunal called that "a remarkable submission" and wrote: "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." The airline was ordered to pay a total of $812.02, including $650.88 in damages.

The dollar amount is small. The principle isn't. If your return bot tells a customer in New Westminster they've got 60 days when your policy says 30, you should expect to own that promise. The tribunal also rejected the idea that customers should have to double check the chatbot against another page of the site.

What a return policy agent actually does

Think of it as a narrow assistant, not a general chatbot. It answers questions about returns, exchanges and refunds, using only your written policy and, if you connect it, your order system. Everything else gets handed to a person.

A sensible first version does four things:

• Answers policy questions in plain language, quoting the exact line from your policy it's relying on
• Checks an order date against your return window when the customer gives an order number
• Explains what to bring in store (receipt, original packaging, payment card) and your holiday hours
• Hands off to staff for anything outside the policy: damaged goods disputes, exceptions, angry customers, or anything it isn't sure about

That last point matters most. Holiday returns are full of edge cases. Gift returns with no receipt, items bought on sale, a final sale tag someone missed. The agent's job is to clear the easy 80 percent of questions (an illustrative figure, not a measured one) and route the rest cleanly, with the conversation attached so your staff don't start from zero.

How to keep it from making things up

Large language models can produce confident, wrong answers. Anthropic, which makes the Claude models, says so plainly in its developer guide on reducing hallucinations: even the most advanced models "can sometimes generate text that is factually incorrect or inconsistent with the given context."

The same guide lists practical fixes, and they map neatly onto a return bot. Give the model explicit permission to say "I don't know." Tell it to use only the documents you provide and not its general knowledge. Have it cite a supporting quote for each claim, and drop any claim it can't back up. For a return bot, that means every answer about a return window or refund type shows the policy sentence it came from.

The guide is also honest about the ceiling. It says these techniques "significantly reduce hallucinations" but "don't eliminate them entirely," and advises that you "always validate critical information, especially for high-stakes decisions." For a shop, a refund decision is a high stakes decision. That's why the agent explains the policy but doesn't approve refunds on its own.

A few more guardrails we'd recommend:

Keep the policy in one place, versioned, and update it before your holiday hours change. Test the agent with your twenty most common questions plus your ten weirdest ones before launch. Read a sample of real conversations every day during the rush week, not after it.

Privacy: what your customers should know

Return conversations involve names, order numbers and sometimes payment details. Canada's federal, provincial and territorial privacy regulators published joint Principles for responsible, trustworthy and privacy-protective generative AI technologies in December 2023. Three points apply directly to a shop using a return agent.

First, accountability. The principles say "accountability for decisions rests with the organization, and not with any kind of automated system." That lines up with the Air Canada decision.

Second, openness. Make sure people "are aware that they are interacting with a generative AI tool." A simple line at the top of the chat does it.

Third, limiting collection. The principles recommend using de-identified information in prompts "where possible and reasonable." In practice, the bot doesn't need a full card number to check a return window. Don't ask for it.

Where this doesn't apply

An AI return agent isn't the right move for every shop, and it's better to say so up front.

If you get a handful of return questions a week, a clear policy page and a well written email template will do the job for far less effort. If your policy is still mostly "it depends, ask the owner," write it down first. An agent can only be as consistent as the document behind it.

High value or regulated goods are a different story too. If returns routinely involve inspection, serial numbers, warranty claims or safety issues, keep a human in the loop for every case and use AI only to collect details before handoff.

The tooling also has rough edges. Connecting to an older point of sale system can be fiddly, and some smaller platforms don't expose order data easily. And as the Anthropic guide itself warns, no prompt technique fully removes the risk of a wrong answer. If you can't commit someone to reviewing conversations during the rush, wait until you can.

Finally, nothing here is legal advice. If your policy has unusual terms, run them past a lawyer before you automate them.

A realistic timeline

If you want an agent live before Boxing Day, the work is mostly on your side: a clean written policy, a list of real customer questions from last year, and a decision on what gets escalated and to whom. The build itself is the smaller part. Starting in early December is tight. Starting now is comfortable.

Sources

• Consumer Protection BC. "The scoop on refunds, returns, and exchanges." 2020. https://www.consumerprotectionbc.ca/2020/01/the-scoop-on-refunds-returns-and-exchanges/
• Civil Resolution Tribunal (British Columbia). Moffatt v. Air Canada, 2024 BCCRT 149, Reasons for Decision. 2024. https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do
• Anthropic. "Reduce hallucinations," Claude developer documentation. Accessed October 2026. https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations
• Office of the Privacy Commissioner of Canada, with provincial and territorial privacy authorities. "Principles for responsible, trustworthy and privacy-protective generative AI technologies." 2023. https://www.priv.gc.ca/en/privacy-topics/technology/artificial-intelligence/gd_principles_ai/

If you run a shop in Burnaby, New Westminster or anywhere in Metro Vancouver and want a return policy agent that answers from your real policy and knows when to hand off, book a free call with Autana. We'll look at your policy and your return volume and tell you honestly whether it's worth doing this season.