Shopify storefront concierge vs operator AI (do not confuse them)

Buyers say “AI for my store” and mean three different jobs: talk to shoppers, run ads, fix ops. Mixing them is how you get a chatbot that pretends to be a media buyer.

Concierge does

• Parse shopper language into product filters
• Route to collections when they exist
• Suggest nearest alternatives when hard filters miss
• Cap costly alternative loops per session
• Log product gaps for merchandising truth

Operator desk does

• Morning brief, campaigns, stock/ads correlation
• Pending review before spend ships
• Multi-tenant store isolation for operators

Install reality

Concierge needs install/settings on the storefront side. It is not “flip a switch inside Ads Manager”. Treat it as a site capability with its own admin routes, separate from the operator shell.

How to evaluate a storefront concierge before you install it

Storefront concierge tools demo beautifully and fail in specific, predictable ways. These are the questions that separate the two, and they take an afternoon.

1. Decide which problem you are buying for. A storefront concierge answers shopper questions on your site. An operator desk helps you run the store. They share the word "AI" and nothing else. Buying one expecting the other is the most common disappointment here, so name the problem first: pre-purchase questions and support volume, or your own workload.
2. Test it against your hardest catalogue question. Not "do you ship to Ireland". Ask something that requires real catalogue knowledge: which of these jackets is warmest, what fits a 34 inch chest, what goes with this. If it answers plausibly but wrongly, it will do that to customers at scale, and confident wrong answers cost more than no answer.
3. Check whether it knows what is in stock, at variant level. Ask for something you know is sold out in one size. If it recommends it anyway, the integration is cosmetic. Recommending an unavailable variant creates a support ticket and a refund, which is the opposite of what you bought it for.
4. Find out what it does when it does not know. The correct behaviour is to say so and hand off, not to improvise. Ask it something genuinely unanswerable and watch. A tool that never says "I am not sure" is not confident, it is unsafe.
5. Check the escalation path to a human. How does a conversation become a support ticket, does it carry the history, and what happens outside working hours. A concierge that traps a frustrated customer in a loop is worse for the brand than no widget at all.
6. Ask exactly what it does with customer data. Where conversations are stored, for how long, whether they train a shared model, and what happens to personal details a shopper types into a chat box. Get this in writing before install, because it is your GDPR obligation and not the vendor's.
7. Understand the cost model before traffic arrives. Per conversation, per resolution, or flat. Then ask what happens on your busiest day, whether there is a cap, and what the cap does when it is hit. Usage-based pricing with no ceiling is a bill you cannot forecast on exactly the day you can least afford a surprise.
8. Measure against deflection and conversion, not chat volume. Tickets avoided and conversion rate for shoppers who engaged, against those who did not. Chat volume measures how prominent the widget is, nothing more, and every vendor dashboard leads with it.

If what you actually needed was help running the store rather than answering shoppers, that is a different product entirely, and the distinction is the whole point of this page.

In Ralph: concierge modules for install, sessions, intents, collections, alternatives, and gaps, separate from operator approval flows. Concierge docs.