What shipped
Home Depot has expanded Magic Apron, its AI shopping assistant, to more than 2,000 US stores, where it now handles millions of customer questions each month. Shoppers reach it through the Home Depot app or by scanning a QR code posted in-store, and it accepts text, voice-to-text, and photo uploads, responding in a shopper's preferred language. The expansion adds store-specific inventory and aisle-location lookups to capabilities the tool already had for general project advice and product recommendations.
What distinguishes this rollout from a typical chatbot deployment is the scope of what Magic Apron is now expected to answer. It can generate a map to a specific product inside a specific store, check whether that product is in stock at that location or a nearby one, and advise on whether a given product actually fits the shopper's intended use case. That last capability, judging suitability rather than just locating inventory, is the harder engineering problem and the one most retailers have not yet attempted at this scale, since it requires the system to reason about a project's requirements rather than simply match keywords against a catalog.
From digital assistant to store-floor tool
Jordan Broggi, Home Depot's executive vice president of interconnected retail, described the intent plainly: 'Magic Apron takes the best of our digital expertise and brings it into the physical store.' That framing matters because it signals Home Depot is not treating its online and in-store AI efforts as separate projects. The assistant was built first as a digital tool and is now being extended onto the store floor with the same underlying system, rather than a new in-store product built from scratch.
That sequencing choice has real operational advantages. A single assistant trained on the same catalog, the same project guidance, and the same product-fit logic across channels avoids the inconsistency that plagues retailers running separate digital and in-store AI initiatives. It also means improvements to the online assistant, driven by millions of digital interactions, compound directly into the in-store experience instead of requiring a parallel development and testing cycle for a store-specific version. Retailers that instead commission separate in-store and digital tools tend to discover the gap only after customers start noticing inconsistent answers between channels.
The trust gap the data reveals
Survey data cited alongside Home Depot's expansion complicates the rollout's apparent momentum. Research from SOTI found 62 percent of global consumers want personalized in-store experiences, and 58 percent say they do not mind retailers using AI as long as it delivers that personalization. Those numbers suggest a real, substantial audience ready to engage with tools like Magic Apron. But a YouGov survey found that one in three US adults will not use an AI shopping assistant at all, specifically because they fear it will push them toward purchases they do not need.
Those two data points describe the same market from opposite directions, and retailers need to plan for both simultaneously. A third of the addressable customer base actively distrusting AI-driven recommendations is a design constraint, not a rounding error that better prompt engineering will quietly fix. It argues for building in an obvious opt-out, clear disclosure of how shopping history and project data get used, and recommendation logic that is visibly weighted toward the customer's stated task rather than toward higher-margin upsells, with the weighting made legible enough that a skeptical shopper can actually verify it for themselves.
Why complement, not replacement, is the harder design problem
Home Depot has been explicit that Magic Apron is meant to work alongside store associates rather than in place of them, a positioning choice with obvious public relations value but also a real product design implication. An assistant designed purely to resolve a shopper's question at the point of purchase has different incentives than an associate whose job includes building a longer-term customer relationship. Keeping Magic Apron aligned with that second goal, rather than optimizing purely for conversion, is the harder and less visible part of the engineering work.
It is also the part most competitors underinvest in. Many retail AI assistants are built and measured primarily on conversion lift, which quietly pushes their behavior toward upselling regardless of how the product is marketed externally. Home Depot's stated positioning only holds up if its internal success metrics genuinely reward project-completion and associate-handoff quality, not just incremental basket size, and that distinction is invisible from the outside but decisive for whether shopper trust holds over time as the assistant scales into every remaining store.
The decision for other retailers building similar tools
Home Depot's approach offers a concrete template: build the assistant's core reasoning once, prove it digitally against high interaction volume, then extend the same system into stores rather than commissioning a separate in-store product. That sequencing reduces both engineering duplication and the risk of customers encountering inconsistent answers depending on the channel they happen to use. It is a build decision retailers evaluating buy-versus-build options for in-store AI should weigh carefully before committing to a vendor platform that only covers one channel.
The harder decision is cultural rather than technical: whether leadership is willing to measure the assistant on trust-preserving metrics instead of pure conversion, even when conversion-optimized behavior would look better on a quarterly dashboard. The YouGov finding that a third of shoppers already distrust AI assistants on sight means that getting this wrong does not just cap upside, it actively pushes a meaningful customer segment away from a channel that the business has already invested heavily in building, and that segment is unlikely to come back just because the next version ships more features.
