Google is unleashing AI shoppers on enterprises — is your infrastructure ready?

AI shopping assistants, rather than elves, may be the ones bustling behind the scenes this holiday season.

At least, Google seems to be pushing in that direction: The tech giant has released a “major AI shopping update” in Gemini that can trigger AI agents to call stores, actively track pricing, and even purchase items on their own.

This signals a new shopping paradigm, but it may also tax enterprise systems and practices.

“Google’s update moves retail closer to intent-based shopping, where the experience feels less like hunting and more like being guided to the right answer,” noted Julie Geller, a principal research director at Info-Tech Research Group.

Shoppers can even have AI agents call stores

As a natural extension of its AI-powered capabilities, Google’s AI Mode can now process shopping questions in native language. That is, shoppers can describe what they’re looking for and receive an “intelligently organized response,” with images, pricing, reviews, and inventory info.

Responses are tailored and formatted to respond to user questions and needs, Google explains. For instance, a shopper looking for “cozy sweaters for happy hour in warm autumn colors” will receive a list of shoppable images; another on the fence about moisturizers, meanwhile, may get a table with side-by-side comparisons based on product reviews.

“Buyers will be able to get very personal recommendations, and aggregate vendors much like they do with Google already,” noted Jason Andersen, VP and principal analyst at Moor Insights & Strategy.

Going a step further, users can now shop right inside Gemini and, when searching for products “near me” in AI mode, can access a “let Google call” button. As they browse, Gemini will prompt them for more specifics, and on the backend, call nearby stores to determine availability, price, and information on any special promos. The shopper will then receive an email or text with inventory information on Google’s aggregate Shopping Graph. This features 50 billion product listings, two billion of which are updated every hour, according to Google.

These capabilities are currently only available to US-based users. Google’s Duplex technology underpins these new features, along with a “big Gemini model upgrade” to help the AI identify the best stores to call, suggest follow-up questions, and summarize key conversation takeaways. “Let Google call” rolled out in search this week in the US, in categories including toys, health and beauty, and electronics.

Rounding out the shopping experience, Google is now supporting full-on agentic checkouts. Shoppers can keep tabs on certain items via a price-tracking feature — size, color, amount they want to spend — and will receive a notification when the product comes into their price range.

Then, at least with some eligible merchants, shoppers can opt to have Google purchase the item via Google Pay. Google is rolling out the capability initially with a number of US merchants, including Wayfair, Chewy, Quince, and some Shopify retailers.

Google emphasizes that AI will always ask for permission before buying anything, and will only pay after a human approves the price and shipping details.

It says these new features are “giving merchants a new way to drive foot traffic,” while also freeing up shoppers’ time.

Enterprises should rethink infrastructure

Of course, this isn’t the first time we’ve seen agents integrated into the shopping experience; Walmart, Saks Fifth Avenue, Amazon, and others have been experimenting with AI-powered shopping capabilities.

However AI agents manifest, though, experts urge enterprises to rethink their infrastructure.

Google’s new agentic shopping features can strain enterprise e-commerce systems by “collapsing the discovery and checkout journey into a rapid chain of machine actions that all hit at once,” noted Info-Tech’s Geller.

What used to unfold step by step now fires almost simultaneously. When an agent checks pricing, inventory, reviews, and delivery options in a few seconds, any messy data or slow decision point shows up immediately, she pointed out.

“Most enterprise systems were built around human browsing patterns, so this creates pressure on the parts of the stack that aren’t clean or are loosely connected,” said Geller.

The real work for enterprises is making sure the core pieces “don’t trip over one another,” she said. This requires consistent product data, category structures that make sense, and decision systems that can operate “without pulling everything else down with them.”

“Guardrails around how quickly an agent can hit different endpoints matter too, because the traffic no longer looks anything like traditional browsing,” said Geller.

Operators should keep an eye out for unusual patterns and step in early. One session triggering a sudden cluster of requests, or disagreement among availability and delivery systems are signs that the system is “being pushed in ways it wasn’t designed for,” said Geller.

However, there is a positive side, she noted: Pressure from AI agents forces companies to clean up the fundamentals, and shoppers will “feel that right away.”

“Information is clearer, options feel more aligned, and the small contradictions that usually frustrate people start to fall away,” said Geller.

There has been some “nice uptake” of these types of agentic features for standalone e-commerce, such as Amazon’s Rufus, noted Moor’s Andersen. “But Google takes it across many sites,” he said.

Google is abstracting the agent from the e-commerce site into a graph, which shouldn’t (at least in theory) impact site performance or scale. But Andersen questioned how often the graph will update and whether it could potentially create new or different pricing incentives.

For example, will Google share with sellers (or their competition) that a certain number of customers have asked to be flagged when their item drops from $120 to $99 MSRP? “That would be incredibly valuable information,” said Andersen.

Further, seller behavior could change based on Google’s graph updates, resulting in more or fewer flash sales. It also creates challenges for distribution models.

“If I have several certified sellers, will there be a race to the bottom on my product, where an agent can pit different routes to market against each other,” Andersen questioned, “and how will Google prioritize the sellers?”

At this early stage, it’s difficult to know whether vendors will have the ability to opt out of the shopping graph, or if adoption will be slow enough so they can adapt as this new buying paradigm develops, Andersen noted.

Overall, he said, “this looks great for buyers, but for sellers, it could potentially be very disruptive.”Google is unleashing AI shoppers on enterprises — is your infrastructure ready? – ComputerworldRead More