Michaels shipped an AI shopping assistant in six weeks

The headline number in the Michaels announcement is 75,000 conversations. The number that should get your attention is six weeks. That's how long it took a national craft retailer to get an AI shopping assistant from concept to production — and it reframes the question from "can we afford this" to "is our product data good enough to support it."

What actually happened

Michaels announced on July 21, 2026 that "Ask Mike," live since May on Michaels.com and its iOS and Android apps, was built with Google Cloud's Gemini Enterprise for Customer Experience. The reported results:

• Nearly 75,000 conversations since release
• Over 60% of interactions were product discovery
• Six weeks from concept to production-grade deployment, per Paul Tepfenhart, Google Cloud's global director of retail industry strategy and solutions

Instead of typing exact product names or clicking through filters, shoppers describe a project — a centerpiece, a kid's birthday craft, a gift — in plain language and get products back. Michaels says AI-driven product overviews and contextual prompts on product detail pages are next.

Why an AI shopping assistant matters for your business

That 60% figure is the tell. This isn't a support bot deflecting tickets. It replaced site search for most of the people who used it, which means it inherits site search's failure mode: it can only surface what your catalog actually describes.

"I'm making a centerpiece for a fall wedding" only resolves to a shopping cart if your product data knows materials, colors, sizes, occasions, and skill level. Michaels shipped in six weeks because the expensive part was already done — a structured catalog with real attributes on hundreds of thousands of SKUs. If your products are a name, a price, and a paragraph a vendor wrote in 2019, a conversational layer on top of that will confidently recommend the wrong thing, and it'll do it fast.

So budget the catalog, not the model. Cleaning attributes, filling gaps, and building a retrieval index over your own product data is where the time and money go. The assistant on top is genuinely a few weeks of work now.

One more thing worth noticing: Ask Mike runs on Michaels' own site. That's the opposite trade from handing your catalog to a third-party shopping agent. Michaels keeps the conversation, the intent data, and the customer. You should too — the same product data feeds both, but only one of them tells you what your customers are actually asking for.

Sources: Michaels press release, Retail Dive.