Holiday Pop-Up Markets: AI Stock Sync for Makers

Picture a Saturday in December. You've got a table at a holiday market in New Westminster, a Shopify store running at home, and forty hand-poured candles split between the two. By 2 p.m. the cedar ones are gone from your table. Your website still says six are left. Somebody buys two online that night, and Sunday morning you're writing an apology email.

That's the overselling problem, and it's the main reason Metro Vancouver makers ask us about AI around the holidays. The honest answer is that AI helps with part of it, plain software handles most of it, and a few pieces still need you. Here's how the parts fit.

Why stock drifts at a market table

Most makers lose count for three boring reasons, not one dramatic one.

First, the Wi-Fi. Community halls and outdoor markets are famous for weak signal. Shopify's own help page on offline features in Shopify POS is blunt: "Shopify POS can't sync orders and inventory with your Shopify admin when you're offline." It syncs again once you reconnect, but until then your online store is selling from an old number. The same page lists things that don't work offline at all, including creating new products, gift card transactions, discount codes, returns and exchanges.

Second, card payments taken offline carry their own clock. As of October 2026, Square's Canadian support article on offline mode says Register, Terminal and Handheld devices must upload offline payments within 72 hours, with 24 hours "highly recommended," and that expired offline payments "cannot be retrieved or reprocessed." It also says you're responsible for any declined or disputed payments taken while offline. So a sale you rang up isn't fully a sale until the device reconnects.

Third, the side channels. Cash sales jotted on a notepad, a friend holding three mugs for a coworker, an Instagram DM that says "save me the blue one." None of that touches your inventory system unless someone types it in.

What actually does the syncing (hint: it isn't AI)

Keeping one stock number across your POS and your website is a job for ordinary integration code, not a language model. Shopify already does this natively when your market table and your online store share the same product catalogue. Where makers run into trouble is when they mix systems, say Square at the table and Shopify online, or a wholesale spreadsheet on the side.

If you or a developer connect those systems, two details from Shopify's developer docs matter a lot.

Shopify's webhook best practices warn that "webhook delivery isn't always guaranteed" and that Shopify "doesn't guarantee ordering within a topic." Their advice is to run a background reconciliation job that periodically pulls fresh data through the API instead of trusting every event to arrive. In plain terms: your sync needs a safety net that rechecks the numbers on a schedule.

The inventorySetQuantities mutation reference (Admin GraphQL API version 2026-10, checked October 2026) adds another guardrail. Since version 2026-04, every update must state the quantity it expects to replace, and the change only goes through if that still matches; otherwise Shopify returns a stale-quantity error. The docs warn that skipping this check "can lead to inaccurate inventory quantities if multiple requests are made concurrently." That's exactly the market scenario: a table sale and a web sale landing at the same moment.

None of that requires AI. It requires careful plumbing.

Where AI earns its spot

So where does AI fit? In the messy, human parts that don't arrive as clean data.

• Turning scribbles into stock updates. Snap a photo of your cash tally sheet at the end of the day and have an AI assistant read it into a draft adjustment, item by item, for you to approve.
• Catching the side channels. An AI agent watching your Instagram and email inbox can flag "hold the blue mug for me" messages and suggest a reservation so that item stops showing as available.
• Packing lists. Before each market, it can look at what sold at past events and what's left in stock, then draft a list of what to bring and what to leave home for online orders.
• Low stock nudges in plain language. Instead of a dashboard you never open, a text that says "four cedar candles left across both channels, want to pull them from the website?"
• Answering customers after hours. When someone asks at 11 p.m. whether the large tote is still available, an assistant can check the live number and reply, instead of guessing.

Notice the pattern. AI reads, suggests and drafts. The actual stock change goes through the guarded API path described above, ideally after a human taps "approve."

Keep the AI on a short leash

This isn't excess caution. The OWASP GenAI Security Project lists "Excessive Agency" as one of its 2025 top risks for LLM applications: damage that happens when a model with access to other systems acts on an output that was hallucinated or manipulated. Their recommended fixes map neatly onto a maker's shop. Give the agent only the functions it needs. Limit its permissions, so it can propose inventory changes but maybe not delete products or issue refunds. And require human approval for high impact actions.

For a small shop, a sensible setup is read access to orders and stock, the ability to draft adjustments, and nothing that touches money without a person signing off. If the AI misreads a smudged "8" as a "3" on your tally sheet, you want to catch that on your phone, not discover it when a customer's order bounces.

Where this doesn't apply

Be honest with yourself about scale before you spend money.

If you do two markets a year and sell a dozen products, a shared catalogue in one POS system plus a quick manual recount after each event is probably enough. Building a custom AI workflow for that is overkill.

If everything you sell is one of a kind, like original paintings or vintage pieces, the cleaner fix is often to unpublish a market item online the morning you pack it, rather than sync anything in real time.

If your venue has no signal at all, no sync, AI or otherwise, can update your website until you're back online. The fix there is operational: hold back a buffer of stock from the website on market days, or mark market items as unavailable online for the weekend.

And if your product data is a mess, with duplicate listings, missing SKUs, or variants named differently in each system, AI will happily automate the mess. Clean the catalogue first.

Finally, the tooling isn't fully mature. Integrations between different POS and ecommerce platforms vary a lot in quality, and as Shopify's own docs admit, events can arrive late, out of order or not at all. Any setup that doesn't include a scheduled recheck is going to drift.

A practical starting point for this season

If you're selling at Burnaby or Vancouver markets this December, here's a sequence that works without a big build:

Start with one source of truth for stock. Turn on offline mode on your POS and know its upload deadline. Set a stock buffer on your website for market weekends. Then, if the side channels are still eating your counts, add an AI assistant that reads notes and messages and drafts adjustments for you to approve.

That order matters. The AI works best on top of a clean, boring foundation.

Sources

• Shopify Help Center, "Offline features in Shopify POS," accessed 2026. help.shopify.com
• Square Support Centre (Canada), "Process card payments with Offline Mode," accessed 2026. squareup.com
• Shopify Developer Docs, "Webhook best practices," accessed 2026. shopify.dev
• Shopify Developer Docs, "inventorySetQuantities" mutation, Admin GraphQL API version 2026-10, accessed 2026. shopify.dev
• OWASP GenAI Security Project, "LLM06:2025 Excessive Agency," 2025. genai.owasp.org

If your holiday markets keep turning into overselling apologies, book a free call with Autana. We'll look at how your table, website and inbox fit together and tell you straight whether AI is worth adding yet, or whether a simpler fix will do.