AI
Boxing Day AI Price-Drop Emails for Burnaby Shops
Boxing Day in Metro Vancouver is loud. Every inbox fills up at once, and the shops that come out ahead aren't always the ones with the biggest discounts. They're the ones that get the right markdown to the right customer quickly, without breaking any rules.
This post covers how a Burnaby or New Westminster retailer can use AI to send price-drop alerts this December, what it really costs, and the two Canadian laws you need to follow while you do it.
Why price-drop alerts matter in December
It helps to know how much of the season actually happens online. In its retail trade release for December 2025, Statistics Canada said retail e-commerce sales "increased 3.6% to $4.3 billion in December, accounting for 6.1% of total retail trade." The same release put British Columbia retail sales at $9.6 billion for the month, down 0.5% from November.
So online sales are a slice, not the whole pie. Most of your Boxing Day revenue probably still comes through the front door. A good alert should bring people into the store, not just get a click.
The idea is simple. A customer shows interest in a product, the price drops, and they get a short note about it. The hard part is doing that for hundreds of products and thousands of customers in the 48 hours around December 26.
How the setup actually works
Here's the setup we'd recommend, roughly in this order:
- Price feed. Pull current prices every night from your POS or online store.
- Plain code decides what counts as a drop. A simple script compares today's price to your documented regular price and flags real drops above a threshold you set. The AI model doesn't make this call.
- Match to interested customers. Use wishlists, abandoned carts, past purchases in the same category and in store signups, but only for contacts with valid consent.
- AI drafts the email. The language model writes a short message in your shop's voice, using the product name, the real prices passed in as fixed data, and the customer's first name.
- Automated checks. A script confirms that every price in the draft matches the source data exactly. Anything that doesn't match gets held back.
- A person approves a sample. Someone at the shop reads a batch before anything goes out.
- Your email platform sends it, with your business address and a working unsubscribe link.
The most important choice here: don't let the AI do math or decide what the regular price was. Language models are good writers, but they aren't a ledger. Give them the numbers, tell them not to change anything, and then check the output against your records anyway.
What it costs to run (illustrative)
As of November 2026, Anthropic's models overview lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, and Claude Haiku 5.5 starting at $0.10 per million input tokens and $0.50 per million output tokens, all in USD. The same page says Batch API requests are 50% off.
Illustrative example only: say you send 2,000 personalized alerts. Each one uses about 1,500 input tokens (instructions plus product data) and 300 output tokens. That's 3 million input tokens and 600,000 output tokens. On Sonnet 5.5 that's about $6 for input plus $6 for output, so roughly $12 USD, or about $6 through the Batch API. Your real numbers depend on prompt length and which model you choose. Prices change too, so check the current page before you set a budget.
The model is rarely the expensive part. Setup time, cleaning up product data and email platform fees usually cost more.
Law one: CASL decides who you can email
Canada's Anti-Spam Legislation covers every commercial electronic message you send. Section 6 of the Act says the message needs the recipient's consent, express or implied. It also has to identify the sender, give a way to contact them, and include an unsubscribe mechanism. That contact information has to keep working for at least 60 days after the message is sent.
Implied consent covers less than many owners think. Under section 10, an existing business relationship includes someone who bought from you within the two years before the message was sent, or who made an inquiry within the six months before it. A customer whose last purchase was three years ago probably falls outside that window. So does someone who asked about a couch back in May.
Unsubscribes also run on a deadline. Section 11 says unsubscribing has to cost the recipient nothing, and the request has to take effect "without delay, and in any event no later than 10 business days after the indication has been sent." If your AI workflow pulls contacts from a spreadsheet that doesn't sync with your email platform's unsubscribe list, you've built a compliance problem. Fix that connection before you automate anything.
Law two: the Competition Act decides what 'was' means
This is the rule that catches retailers. The Competition Bureau's page on ordinary selling price, last modified December 18, 2025, says you can't advertise a discount against a regular price the product never really sold at, and you have to be able to prove your reference price.
The Bureau describes two tests. Under the volume test, more than 50% of the product's sales were at that price or higher within a reasonable period, usually about a year before or after the promotion. Under the time test, the product was offered in good faith at that price or higher for a substantial period, again usually about a year. If the regular price meets neither test, the discount claim can be considered misleading.
For an AI alert, that means the 'was' price has to come from your sales records, never from the model. Say an item was marked up in November just so it could be marked down on Boxing Day. An automated email repeating that claim to 2,000 people doesn't lower your risk. It multiplies it. This isn't legal advice, so talk to a lawyer if your pricing history is messy.
Where this doesn't apply
Automation isn't the right answer for every shop, and it's better to know that now than on December 27.
Small lists. If you have 150 subscribers and 30 items on sale, just write one good email yourself. AI personalization won't add much at that size.
Unreliable price history. If your POS doesn't keep accurate regular price records, you can't be confident you meet the Bureau's tests. Automation would just spread the errors faster.
Missing consent records. If you can't tell which contacts gave express consent and which ones bought within the last two years, sort that out first. No tool fixes this for you.
Late starts. If it's mid December and you've never built this before, it's probably too late to do it well for this Boxing Day. A rushed setup with no testing is how wrong prices end up in thousands of inboxes. Start with one simple segment, or plan for next year.
Model errors. Even strong models can occasionally change a number or make up a product detail. Automated checks catch most of these, and human review catches more, but neither is perfect.
Inbox fatigue. Everyone emails on December 26. More messages don't automatically mean more sales, and too many alerts will push people to unsubscribe.
A sensible place to start
Pick one category with about 20 products. Send alerts only to wishlist and abandoned cart customers who clearly have consent. Test it on a small early December sale and track opens, clicks and in store redemptions. If those numbers hold up, expand to more categories before Boxing Day.
Sources
- Statistics Canada, "Retail trade, December 2025," The Daily, 2026. https://www150.statcan.gc.ca/n1/daily-quotidien/260220/dq260220a-eng.htm
- Government of Canada, Justice Laws Website, Canada's Anti-Spam Legislation, section 6, accessed 2026. https://laws-lois.justice.gc.ca/eng/acts/e-1.6/section-6.html
- Government of Canada, Justice Laws Website, Canada's Anti-Spam Legislation, section 10, accessed 2026. https://laws-lois.justice.gc.ca/eng/acts/e-1.6/section-10.html
- Government of Canada, Justice Laws Website, Canada's Anti-Spam Legislation, section 11, accessed 2026. https://laws-lois.justice.gc.ca/eng/acts/e-1.6/section-11.html
- Competition Bureau Canada, "Ordinary selling price," 2025. https://competition-bureau.canada.ca/deceptive-marketing-practices/types-deceptive-marketing-practices/ordinary-selling-price
- Anthropic, "Models overview," Claude Platform documentation, 2026. https://platform.claude.com/docs/en/about-claude/models/overview
Want help setting up price-drop alerts that use your real prices and respect your customers' consent? Autana Solutions is based in Burnaby and New Westminster and works with shops across Metro Vancouver. Book a free call and we'll look at your data, your list and your timeline, and tell you honestly whether automating makes sense this year.
Want an AI employee for your business?
We install a 24/7 AI worker for businesses in Vancouver, Burnaby, and beyond. Book a free Discovery Call.
Book a call →

