Agentic Commerce Is No Longer a Pilot. Shopify, NVIDIA and Google Cloud Explain What Comes Next
Shopify, NVIDIA and Google Cloud told Shoptalk Europe 2026 that agentic commerce is moving past pilots. Winners concentrate AI investment, use hybrid models, and align leadership across commercial and technical teams.
- 1BCG data cited on stage: AI leaders invest twice as much as peers in 50% fewer initiatives, concentrating bets rather than scattering pilots.
- 2Shopify says brands in its billion-product feed have seen conversion double, AI search traffic rise eightfold and orders climb thirteenfold, with the Universal Commerce Protocol keeping retailers as merchant of record.
- 3NVIDIA's Azita Martin: hybrid frontier plus fine-tuned open-source models are now required in production, and joint work with Shopify on Nemotron has delivered 4.5x faster responses at 6x lower inference cost.
The mood at Shoptalk Europe 2026 in Barcelona shifted noticeably on day three. The question was no longer whether to invest in agentic commerce. It was what separates the retailers already generating commercial value from those still running experiments that go nowhere.
Martin Barthel, Managing Director and Partner at BCG, framed that question directly when he brought together Deann Evans of Shopify, Azita Martin of NVIDIA and Prue Mackenzie of Google Cloud for the session titled "The Tech Companies Driving Industry Innovation." Each brought a different piece of the stack. Together, they made the same argument: the gap between leaders and followers is widening fast, and the differentiator is not technology. It is leadership, data infrastructure and the discipline to concentrate investment rather than scatter it.
A BCG study Barthel cited during the session found that AI leaders invest twice as much as their peers but in 50% fewer initiatives. In practice, that means fewer chatbot pilots and proof-of-concept experiments, and more concentrated bets on the two or three areas where AI can genuinely move the business. That ratio, he suggested, is the clearest signal of what separates companies generating real returns from those accumulating impressive pilot portfolios.
Shopify: The Billion-Product Feed and What It Takes to Show Up in It
Deann Evans, Managing Director for EMEA at Shopify, opened with a number that reframes the entire agentic commerce conversation. Shopify's catalogue now holds over a billion products. The brands included in its feed, she said, have seen conversion double, traffic from AI search rise eightfold and orders increase thirteenfold. AI-driven customers convert 50% more often and spend 14% more per order. These are Shopify's own figures as presented by Evans on stage and have not been independently verified.
The practical implication, Evans argued, is that the most important question for retailers right now is not whether to experiment with AI. It is whether their data, policies and checkout experience are ready for agentic commerce. Being in the feed matters. Being optimised within it matters more.
Central to Shopify's agentic effort is Sidekick, an AI tool Evans described as evolving from acting like the retailer to acting with them. She cited Oumu, a Danish design brand, as an example using Sidekick across merchandising and analytics.
Evans also detailed the Universal Commerce Protocol, an open standard co-developed with Google and backed by a technical council that includes Stripe, Etsy, Amazon and Salesforce. The protocol allows AI agents to interact with a retailer's product catalogue, inventory and post-purchase processes across any platform using a common standard, with retailers remaining the merchant of record throughout. Shopify merchants are enrolled by default. Retailers outside Shopify can join through the company's Agentic plan.
On the internal side, Evans referenced a memo from Shopify chief executive Tobi Lütke in March 2025, which she said established AI use as a baseline expectation across the entire company. She pointed to River, an internal agent built into Slack that understands Shopify's codebases and databases, and noted that around half of Shopify's code is now AI-generated, with a human reviewing all of it. Her practical recommendation to other organisations: tie every AI initiative to a measurable output, not usage metrics.
NVIDIA: Why the Hybrid Model Is Not Optional
Azita Martin, Vice President and General Manager for AI in Retail and FMCG at NVIDIA, described the market as splitting in two. Most retailers are still at the chatbot and single-function agent stage. A smaller group has moved to multi-agent systems capable of autonomous reasoning and action at scale. The gap between the two is growing, and Martin was direct about what that means for jobs and organisations that wait too long to close it.
"AI is not going to take your job," she said. "But somebody using AI will take your job."
On regional differences, Martin said Asia Pacific retailers are furthest ahead, citing NVIDIA's work with Coupang in South Korea, Shopee in Singapore and Mercado Libre in South America. The United States is catching up. Europe is still behind.
NVIDIA's approach to closing that gap is what the company calls blueprints: pre-built combinations of frontier and fine-tuned open-source models for specific retail tasks, including catalogue enrichment, shopping assistants and supply chain forecasting. As retailers move from experimentation into production, Martin argued that a hybrid model becomes necessary. Frontier models alone are not enough. Fine-tuning an open-source model on a retailer's proprietary data improves accuracy and, critically, means the resulting model belongs to the company that trained it, not to the model provider.
Five Things Friday
Five useful signals from the people shaping global retail. Every Friday.
On the Shopify partnership, Martin said joint work using NVIDIA's Nemotron models has delivered four and a half times faster response times and six times lower inference costs. She was explicit that NVIDIA's core value is its acceleration software, not its hardware alone.
Google Cloud: Otto Was Founded in 1949. It Has Europe's Most Advanced Agentic Commerce Solution.
Prue Mackenzie, Director for Retail, CPG and EMEA Strategic Industries at Google Cloud, described her role as taking Google's underlying technology and translating it into commercial applications for specific retail verticals. Her clients include LVMH, Carrefour and Otto.
She opened with Otto as an illustration of what innovation culture actually looks like in practice. Founded in 1949 with three employees and a mail-order catalogue in Germany, Otto is today the first client Google Cloud has in Europe with a live, voice-activated, conversational agentic commerce solution. The point Mackenzie made is that being furthest ahead in AI does not require being the youngest or the most digitally native. It requires a culture of innovation and leadership from the top.
The more detailed case study Mackenzie walked through was virtual try-on. A luxury fashion client approached Google wanting to launch a high-fidelity virtual try-on experience on a tight timeline. The challenge was that the client's standards were far higher than Google's consumer-facing models could meet: customers expected an exact match on print, colour and fit. A slightly off result was not acceptable.
Google's response was to embed one of its applied AI engineers directly with the client for two weeks to fine-tune the model. After that, the rollout was iterative: internal testing first, then a limited launch in a single country before expanding. The technical work eventually delivered the quality required.
What remains unresolved is the commercial model. Generating a single virtual try-on image can require seven separate model calls. That is expensive for a product category where margins are already under pressure. Mackenzie was candid that the industry still has significant work to do on the commercial side to make these experiences viable at scale, even where the technical capability now exists.
The Macy's deployment offered a more straightforward proof point. Revenue from customers using Google's agentic commerce tools was 4.7 times higher than from those who did not, according to figures Mackenzie cited on stage.
Mackenzie also shared an example where things had not gone well. A client's e-commerce team was fully invested in an agentic search initiative but the technology team was not involved in the project. Without access to the right data infrastructure, the initiative stalled despite strong commercial intent and significant investment. Her conclusion was direct: top-down alignment between commercial and technical teams is not a nice-to-have. It is the prerequisite for any agentic commerce project to deliver.
The clearest signal of how seriously organisations are taking this came from a comment made in the room the previous morning. The head of e-commerce at L'Oréal said the company is looking to hire a dedicated head of agentic commerce. Mackenzie described that as significant because agentic commerce, done properly, cuts across the entire business and cannot sit with a single team that has no mandate over the others.
The Common Thread
Barthel closed by drawing together what all three conversations had in common: cost discipline in how AI is deployed, a deliberate hybrid approach combining frontier and open-source models, and clear ownership at leadership level rather than treating agentic commerce as a technology project.
The BCG finding he cited made the resource question concrete. Leaders invest twice as much as followers. They do so in half as many initiatives. That concentration is not caution. It is the mechanism by which they generate commercial returns while others accumulate experiments.
All statistics cited in this article are as presented by the respective speakers on stage at Shoptalk Europe 2026 and are attributed accordingly rather than independently verified.
Related coverage
Marketplace
FTC filing puts figures on alleged Amazon ad surcharges
Nov 6, 2026
Marketplace
From Keywords to Intent: Retail Search Gets a Reset
Apr 13, 2026Marketplace
NRF Keynote: Gymshark Ben Francis on Building the Next Generation of Retail Brands
Feb 3, 2026
Marketplace

