Macy's Deploys Conversational AI to Reduce Abandonment

Macy's launched a conversational shopping assistant called "Ask Macy's" built on Google's Gemini Enterprise for Customer Experience. According to Google Cloud's case writeup, the agent was developed in roughly four to five weeks and connects to a catalog of more than 2.5 million SKUs. PYMNTS reports that during beta testing, revenue per visit among customers who used the assistant was about 4.75 times higher than among nonusers, and that two months after launch the assistant serves thousands of shoppers daily across Macys.com and the Macy's app. PYMNTS, Google Cloud, and Retail Dive describe the assistant as multimodal, handling text and images, and offering a virtual try-on capability.

Per PYMNTS and Google Cloud's case study, the assistant frames conversations with follow-up questions rather than returning a grid of search results, narrowing recommendations by eliciting preferences such as fit, fabric, occasion, or color, and supporting image uploads for virtual try-on. Google Cloud positions Gemini Enterprise for Customer Experience as the underlying platform. Constellation Research notes the deployment scaled quickly from a small share of site traffic to broad availability within roughly a week.

Industry pattern: agentic, multimodal assistants typically combine catalog retrieval, conversational state tracking, and reranking of product lists based on progressively collected attributes. Companies deploying them often face attribution challenges, because reported per-user revenue uplifts can reflect selection bias when early adopters are higher-intent shoppers. Isolating net lift requires randomized experiments or intent-matched cohorts rather than comparing raw user-versus-nonuser averages.

Google's product blog introduces the Universal Commerce Protocol (UCP), an open standard Google says it co-developed with retailers and platforms and lists Macy's among endorsers. Reporting frames UCP as infrastructure for agentic commerce spanning discovery and checkout, which could reduce integration friction between conversational agents and merchant payment systems if broadly adopted, per Google Cloud and Google product materials.

• •Whether the uplift persists in intent-neutral A/B tests, and how the assistant affects overall cart abandonment and average order value.
• •Latency, reliability, and the cost of serving multimodal inference at production scale.
• •Whether emerging standards such as UCP simplify checkout integration.

Reporting so far does not publish randomized-control results or full methodology for the 4.75x figure, and outlets note Ask Macy's users skew toward already high-intent shoppers, per PYMNTS.