Skip to main content
AmigoFish

New Way Now: PUMA builds the world’s first AI concierge for retail with help from Google Cloud

From Google Cloud · · 3 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.

If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.

Watch on YouTube

Not rated yet

Description

*Featured in this video:* Ivan Dashkov, Director of DTC Technology Innovation at PUMA

*Executive summary:* PUMA, one of the world's leading sports brands, teamed up with Google Cloud to transform in-store discovery with Dylan — the world’s first “digital human” AI concierge. Brought to life through a seven-foot digital display in PUMA’s Las Vegas flagship, Dylan can interact conversationally in over 100 languages, provide expert recommendations across PUMA’s entire catalog, and alert associates to bring out shoes for customers to try in real time. Built on top of Google Cloud infrastructure, including Google Kubernetes Engine and Cloud Run, Dylan uses a multi-model architecture to deliver instant, personalized responses. With model garden in Gemini platform, PUMA can access and choose from over 200 cutting-edge models through one governed catalog, deploying lightweight Google Gemma models for split-second replies and NVIDIA Nemotron models for more complex questions. Dylan has already delivered big wins on the store floor, resolving 75% of inquiries without staff intervention while providing alternative avenues for every customer — from international tourists to introverted shoppers — to discover products on their own terms.

*Challenge:* PUMA wanted to give consumers the keys to their own shopping journeys by using AI to deliver consistent, expert-level product guidance. The goal was an interactive experience that could translate complex footwear specs and break down language barriers, while empowering every fan to explore PUMA’s offerings on their own terms.

*Solution:* PUMA worked in partnership with Google Cloud, alongside NVIDIA and LiveX.ai, to bring its seven-foot digital AI concierge to life at its Las Vegas flagship. Using Google Cloud infrastructure like Google Kubernetes Engine and Cloud Run, PUMA scales Dylan dynamically to support in-store demand without paying for idle capacity during quieter hours. Model garden in Gemini platform enables a multi-model architecture that allows PUMA to pair lightweight models like Google’s open Gemma for instant answers with NVIDIA Nemotron models to handle highly complex, multi-layered product questions. With multimodal and multi-language capabilities, Dylan can scan physical footwear shown by customers, interpret voice inputs, and search store inventory in real time to suggest tailored recommendations across PUMA’s full product catalog in more than 100 languages.

*Results:* Already, Dylan can resolve 75% of inquiries independently across the flagship’s full inventory, whether giving expert advice, finding a PUMA alternative for products shoppers already like, or recommending the best shoes for a night out on the Strip. With Dylan, PUMA is laying the groundwork to transform in-store discovery — delivering a new and powerful way for customers to discover and search for products on their own terms.

*Key takeaways and highlights from our interview with Ivan Dashkov, Director of DTC Technology Innovation at PUMA:*

→ “People want to use AI to interact with our brand. Now, with Dylan, we’re letting people come in, talk to our brand, and really dig deep into the product information. He finds the perfect shoe for you, and then he can alert a store associate to bring you the shoe to actually try on.”

→ “We're using model garden to use a lot of different models here. The Gemma model, we're using that for some of the easier questions and making sure there's no delay in the response from Dylan to the consumer. And then for the more complex questions, we're leveraging the Nemotron model from NVIDIA.”

→ “We're using a number of different technologies to bring Dylan to life. It's built on the Google Cloud infrastructure. GKE, AutoScale and CloudRun for when we need additional concurrency and compute power to hit higher demands, but also we can scale down. One of our goals is to be the first market and being able to build on stable infrastructure is a really great starting point for us.”…

Back to Google Cloud · Report a problem