Macy's, Sephora, and Zalando are all now deploying AI shopping assistants in production. Macy's rolled out "Ask Macy's" across its digital platforms and says customers using it spend 4.75x more than non-users. Sephora launched an app inside ChatGPT, connected to its 80-million-member Beauty Insider program. Zalando expanded its AI assistant to all 25 markets, combining proprietary models with OpenAI's LLMs.
The signal is clear: this category is moving from pilots and demos into real retail operations. But as Macy's Chief Customer & Digital Officer Max Magni put it: "Nobody has cracked the code."
That quote gets to the real issue. The deployment wave is real, but the architectural question behind it is still mostly unanswered in public: how do you build one of these systems so it actually works, so your team can control it, and so it survives contact with real customers?
I spent the last few weeks building a reference implementation for premium floral e-commerce — a category where the purchase is emotional, time-sensitive, and easy to get wrong. My conclusion was simple: the production problem is not "how do we add a chatbot?" It is "how do we design a decision system the business can trust?"
About the author
Cyril Noirot
Lead Data Scientist
Freelance data scientist. I design and ship decision systems — forecasting, pricing, marketing measurement, optimization.
Seen in practice
Anonymized case studies where these ideas were applied to real decision problems.