The Maze: Michaels has put a number behind conversational commerce. The U.S. arts-and-crafts retailer says shoppers who use its Google Gemini-powered Ask Mike assistant convert at more than twice the rate of shoppers using traditional site search. The assistant has handled roughly 75,000 conversations since a quiet May launch, and 27% of interactions end with a product click or add-to-cart action. The headline is conversion. The mechanism is more interesting: Ask Mike turns a project into a basket.
*Source: Michaels. Ask Mike sits beside the retailer's product catalog and invites shoppers to describe a project instead of typing isolated product keywords.*
Traditional search starts with an item; Ask Mike starts with an outcome. A shopper can ask for help planning a child's birthday party, choosing materials for DIY curtains or framing a photo. The assistant then asks about the project, colors, materials and budget before recommending products. Michaels says more than 60% of interactions focus on product discovery. That matters in crafts because customers often know what they want to make, but not every item required to make it. Keyword search solves for “floral fabric.” Conversation can solve for “finish the room.”
The commercial move is from finding one SKU to assembling a mission. Michaels says assistant users bundle core supplies and finishing touches in one session, while conventional-search users more often look for a single item they already know. That can widen the basket without relying on a blunt “frequently bought together” row. The assistant can discover the customer's constraint, then recommend complements that fit the same project. Google Cloud's retail-agent pattern supports multi-turn product filtering, comparisons and cart actions. The chat is only the interface. Structured catalog data is the engine.
Six weeks to production sounds fast because the slow work happened earlier. Google Cloud credits Michaels' existing cloud foundation, modernized systems and organized product catalog. A large language model cannot reliably recommend a party kit if product attributes, availability, variants and relationships are messy. Retailers copying the visible chatbot without fixing those inputs will get a charming layer over weak search. Michaels also brings category knowledge: it knows which questions to ask and which supplies belong together. The moat is partly model access, but mostly clean product data plus merchandising logic.
The conversion claim is promising, not yet a verdict. Michaels says Ask Mike users convert at more than double the traditional-search rate, but it has not disclosed the baseline, traffic split, absolute revenue lift or test design. Users who choose a project assistant may already have higher intent or larger missions than someone typing a product name. The next proof should be incremental: randomized conversion, average order value, margin, repeat use, returns and recommendation errors. Otherwise, a useful early signal risks becoming AI theater with a shopping cart attached.
Why it matters: Onsite search has usually been treated as navigation: help shoppers find the item they named. Ask Mike treats it as merchandising: understand the job, shape the basket and keep more of the project spend with one retailer. That is especially powerful in categories where customers buy systems rather than single products—crafts, beauty routines, home projects, electronics setups and events. Michaels owns the storefront and transaction; Google supplies the AI layer. Watch whether the assistant expands from a self-selected tool into product-page prompts, whether basket economics hold up, and whether the conversion gap survives a controlled test.
Sources: Digiday | Michaels newsroom | Google Cloud documentation | Retail Dive
