.png)
Two things shipped this year that are worth talking about even months later: Shopify's Winter '26 Edition turned on Agentic Storefronts by default across every store on the platform, syndicating product catalogs straight into ChatGPT, Perplexity, and Microsoft Copilot without a merchant lifting a finger. Tobi Lütke put it plainly at launch: "We're making every Shopify store agent-ready by default." A few months later, Klaviyo's Composer started drafting entire email flows, segments, and push notifications from a single plain-language prompt, copy included.
Both of these are genuinely useful tools, and this isn't a piece arguing otherwise. But neither one has anyone checking whether the output actually fits the merchant in front of it. An AI agent reading a catalog doesn't know that one SKU is a loss leader and another is where the real margin lives. A generated email flow doesn't know that a brand's last big sale flopped because the segment was wrong, not the copy. That gap between fast and right is where Concierge lives, and this piece is a straightforward walkthrough of what that process actually involves, since it's one thing to say "a team, not just software" on a homepage and another to show what that means in practice.
Concierge is the done-for-you build: a small team works directly with a merchant to design, build, and launch a quiz specific to their catalog, rather than handing over a template and a login. It costs more than the self-serve builder and takes longer, two to three weeks instead of an afternoon, and that trade only makes sense if the extra time is actually going somewhere. Here's where it goes.
The first week isn't spent writing quiz questions. It's spent reading the business.
That means going through the catalog itself, not just a product list but the actual structure underneath it, which SKUs are entry points, which are upsells, where the margin sits. It means looking at returns data where a merchant has it, since a pattern of returns on a specific product often points to a mismatch between what customers expected and what they got, which is exactly the kind of mismatch a good quiz question can prevent. And it means reading support tickets, because customers describe confusion in their own words there in a way that never shows up in analytics, someone asking "which one is actually for sensitive skin" in a support chat is telling you something a bounce rate can't.
Generic quiz logic assumes a generic catalog: a handful of clean categories, one product per answer combination, no edge cases. Real catalogs rarely work that way. A supplement brand might have three products that all reasonably answer the same question, differentiated only by a dosage preference most shoppers don't know they have an opinion about yet. A pet food brand's inventory might shift weekly in ways a static rule set can't follow.
Building logic that fits means starting from the catalog's actual shape rather than forcing the catalog into a template's shape. That's slower. It's also the difference between a results page that feels like a recommendation and one that feels like a filter with extra steps.
This is the part of the process that's hardest to describe in the abstract, so it's worth being concrete rather than making a general claim about judgment.
A question can test perfectly clean on paper and still confuse real people the first time they see it, wording that makes sense to whoever wrote it but not to a shopper seeing it cold. A recommendation rule can be logically correct and still be wrong for an edge case the logic didn't anticipate, a product that's technically the right match by the rule but the wrong one for anyone who actually knows the catalog. Catching these before launch, not after a few hundred responses have already gone through, is most of what the review step in Concierge is actually for.
"We're making every Shopify store agent-ready by default." — Tobi Lütke, CEO, Shopify
That quote isn't a criticism. Being agent-ready by default is a real capability and most merchants should want it. It's just a different problem than whether the recommendation a customer receives is actually right for them, and that second problem is the one a person still has to sit with.
A few questions worth asking any quiz vendor, Sensez included, before committing:
None of these questions require picking Sensez to be useful. They're a reasonable filter for anyone evaluating this category of tool right now, when it's genuinely harder than it used to be to tell a thoughtfully built product from a fast one.
Concierge setup starts with a conversation about the catalog, not a demo of features. Book a call to talk through what a build would actually look like for a specific business.
