Insights

Europe’s AI Reset Is Not a Growth Story. It Is an Operating-Model Story.

European digital commerce is growing again, but the mechanism has changed. The winners will be the retailers who rewire commerce around agents, not the ones who redecorate the storefront.

Eric Pilkington, Chief Executive and General Manager, UST

European e-commerce is doing something it has not done in years. It is growing again. EuroCommerce puts the market at €842 billion in 2024, up 7 percent nominal and 4.6 percent real, and on track for €901 billion in 2025. Eurostat confirms the same picture from the supply side: 24 percent of EU businesses sold online in 2025, generating 19 percent of enterprise turnover. The consumer is still cautious. The channel is not.

It would be easy to read this as a demand story, and easy to be wrong. The number that matters is not the growth rate. It is why the growth is happening. In the last cycle we moved from catalogs to online, from online to mobile, from mobile to platforms. This cycle is different in kind. The interface between the consumer and the retailer is changing. The economics of growth are changing. The basis of competitive advantage is changing with them.

DIVIDER

What the European data actually says

The share of shopping that begins with a machine is no longer a projection. Adobe’s February 2026 analysis of a trillion visits to US retail sites, a fair leading indicator for Europe, showed AI-referred traffic up 1,200 percent year over year during the holiday quarter, and up 393 percent for the year. By May 2026, that traffic had doubled again in twelve months, converting 54 percent higher than the baseline, and spending 53 percent more time on site. OpenAI’s own signals research shows roughly 50 million shopping-oriented conversations a day on ChatGPT alone, about two percent of prompts across a base of 900 million weekly users. At UST, our July 2025 view of this shift argued that the traditional customer journey is collapsing, and that agents will pick winners on five dimensions: price, availability, reliability, service quality, and trust signals. A year on, the retailer data is doing the arguing.

There is a pattern underneath the numbers. AI adoption is high where cognition is hard, and reversal is cheap, and it stays low where the consumer wants to keep authority. HBR’s research on AI shopping agents finds the same distinction: consumers happily delegate research, comparison, and product discovery, but resist full automation of purchase decisions until trust catches up. Consumers are willing to outsource thinking. They are not yet willing to outsource authority. The retailer’s job is to earn a place inside the delegated cognition step, then defend it when authority follows.

When roughly one in twenty visits already comes from an agent and converts at half again the rate of a human visit, the retailer’s traditional levers of placement, advertising, and interface design lose weight. The question stops being how a retailer converts a human. It becomes whether an algorithm can find, understand, and defend the retailer’s offer against every other offer in the category, at machine speed, on evidence. This is what we mean at UST by AI Agent Optimization: the retailer’s product, price, availability, and policy have to be readable, ranked, and trustworthy to a machine before they can win a human.

DIVIDER

The wins are earnings reports, not consulting slides.

The clearest evidence that this is not a pilot cycle sits in European retailers’ 2025 and early 2026 disclosures. Zalando, Europe’s largest online fashion platform, reported GMV of €17.6 billion, up 14.7 percent, and revenue of €12.3 billion for 2025, with EBIT of €591 million and 2026 guidance of €660 to €740 million. Inside those results, the company disclosed that AI-generated content in its studio production rose from zero to 90 percent in a year, campaign production shrank from six weeks to days, content output grew 70 percent, delivery precision improved by 22 percentage points, and engineering shipped 20 percent more code. OTTO Group, in its 2025/26 annual results, reported €13.8 billion in revenue and EBIT of €641 million, up from €276 million a year earlier, with 42 million active customers and a stated €350 million commitment to AI investment. Its voice-mode shopping assistant already sustains 11 turns in a session against roughly 2 for text search and drives 37 percent higher average order values than classic search.

These wins are not any single lever, and that is the point. They compound. Better personalization increases engagement and media yield. Richer media data improves demand signals. Better demand signals sharpen pricing, assortment, and inventory. Stronger operations feed back into the customer experience. Each interaction generates data that improves the next decision. UST’s own field research on the next retail architecture, which we published in 2025, put the combined effect at roughly a 5-point gross margin improvement, 10 to 15 percent inventory reduction, and 30 to 40 percent design-cycle compression when the layers are wired together rather than deployed in isolation. HBR’s February 2026 study of 800 public companies deploying generative AI reached a related and sobering conclusion: the leaders capturing durable gains do it by rewiring value creation, while headline productivity gains from off-the-shelf tools are already being competed away in the price of the service. Compounding is a system property, not a tool feature.

DIVIDER

The gap that decides the winners

There are now two curves that leaders should be watching. The first is the capability of AI agents. That curve is compounding on the demand side. Adobe’s traffic index for AI-referred visits to retail sites has doubled in twelve months. The Economist has argued that in shopping specifically, agents are moving from novelty to a serious sales channel, with retailers and platforms already experimenting with sponsored responses inside AI answers. The second curve is retailer control readiness, meaning the ability to see, price, publish, decide, and fulfill in the way agents demand. That curve is flat by comparison. Eurostat data on European businesses shows only about a fifth used any AI technology in 2025, with adoption concentrated in large firms and specific functions, and Adobe’s own analysts note that most retail sites are still not machine-readable enough to be reliably surfaced by an agent. UST’s Thinking Ahead survey of 502 senior decision-makers found that 82 percent expect AI to automate significant portions of operations and 75 percent expect it to reshape business models within five years. Still, a companion review with HFS Research concluded that the chief impediment is strategy and operating model, not technology.

MIT Sloan Management Review has made a similar point at the operating layer. Its 2025 work on the emerging agentic enterprise found that 76 percent of executives now view agentic AI as a coworker, and estimated that agentic AI could reach 35 percent adoption in two years, versus roughly 70 percent for generative AI over three years. Its work on agentic AI at scale is blunter: 69 percent of an expert panel says accountability and management systems have to be rebuilt for a workforce that includes machine agents. The gap between those two curves is where advantage will be won or given away. Retailers that close it become the answer the algorithm picks. Retailers that do not will keep optimizing a shelf that agents have already replaced.

DIVIDER

Five shifts, and how to read them correctly

European e-commerce is being reshaped by five shifts at once: agentic commerce, commerce becoming media, retail media as a margin lever, intelligent omnichannel, and the systemic challenge from Chinese platforms. Each one is real. Read as a list, though, they invite the wrong response: a portfolio of pilots. Read as a system, they force a different conclusion. What is actually shifting is not the channel. It is the operating model underneath it.

1. Agentic commerce reshapes who the customer is

HBR has now framed this explicitly: brands are preparing for three interaction patterns at once, brand agents talking to consumers, consumer agents talking to brands, and full agent-to-agent commerce in the middle. Research from China, where Meituan’s Xiaomei and similar agents already handle discovery and transaction, shows the pattern arriving faster than most European boards assume. The retailer’s customer is now two customers at once, one human and one agent. Both have to be served, and only the agent reads structured data.

2. Commerce and media collapse into one surface

Product, content, and transaction now share the same feed. HBR research on AI shopping agents finds that traditional marketing signals, brand recognition, celebrity endorsements, and emotional advertising largely do not move a machine buyer, which reads structured evidence and price signals instead. That is a P&L rewrite, not a campaign brief. It requires first-party data, always-on content, and creative systems that produce and test thousands of variants without a campaign meeting.

3. Omnichannel becomes intelligent orchestration

Customers do not think in channels, and the retailer’s systems should not either. Unified pricing, promotions, and service become a system-of-record problem. AI is the coordination layer that turns dozens of disconnected touchpoints into one continuously optimized journey.

4. Chinese platforms compress the timeline

Manufacturer-led brands moving up the value chain, low-cost platforms compressing delivery, and marketplace alliances have already reset expectations on price, speed, and integration. The Economist has argued that Europe still has a distinct opportunity in the applied and industrial layer of AI, but only if firms move fast enough to matter. That is the sharpest reason not to treat AI adoption as a pilot cycle.

5. Growth is now a capability question, not a demand question

5 to 7 percent annual growth in a cautious consumer environment is a capability signal. It reflects who can capture demand faster, price sharper, and fulfill more reliably at any given moment. The macro is not doing the work. The stack is.

DIVIDER

Intelligent Commerce: the response that adds up

At UST Evolve, we call the response Intelligent Commerce. It is not a rebrand of AAO, and it is not a slogan for personalization. It is an operating stack for the agent era, built on the recognition that AI agents are replacing the shelf. Retailers win by becoming the answer the algorithm picks, not the brand fighting to be seen.

The stack has four layers, and none of them are optional. The agent surface is the machine-readable foundation of product, price, availability, and policy. It is what an agent can find, understand, and cite. This is where AI Agent Optimization lives as a tactical discipline: rich structured metadata, consistent taxonomy, evidence-backed claims, and real-time signals that a machine can trust. Without it, an offer is invisible to the first customer in the funnel.

The decision layer is where merchant AI agents run pricing, assortment, promotions, and content in a continuous loop rather than as periodic decisions. This is where the compound effect shows up in the P&L: higher gross margin, fewer markdowns, fewer stockouts, and faster design cycles, as Zalando and OTTO have already reported in public results. The stronger effect is that the retailer starts learning at a different clock speed than competitors who still plan in quarters.

The media and content layer is where commerce and media stop being separate businesses. Content is the storefront. Retail media is a P&L for scale with dedicated leadership and monetization logic, not a pilot bolted onto merchandising. In a feed-based world, always-on creative production and monetized attention are the same discipline.

The operating model is the layer that most transformations skip and then wonder why nothing compounded. MIT Sloan’s recent research on AI value at scale calls this the “AI spine”, a cross-functional structure that owns data, models, and workflow together rather than in silos, and its research on compounding gains from generative AI finds that organizations with structured feedback loops are six times more likely to see performance improve over time, though only 15 percent currently use AI to help the organization learn. Customer-journey P&Ls, three-month product cycles, and human-AI hybrid teams are what turn the other three layers into a flywheel. Without them, retailers buy AI, keep old decision rights, and the AI politely does what the org chart tells it to do.

DIVIDER

What to do in the next two quarters

The good news is that this is not abstract. The Economist has noted that the economy-wide AI productivity boom has yet to show up in the aggregate data, which means the firm-level dispersion is enormous. That dispersion is the opportunity. There is a short list of moves that separate retailers on the compounding curve from retailers on the pilot treadmill.

1. Make the business agent-ready. Structure product data, pricing, availability, and fulfillment signals so an AI agent can find, evaluate, and transact on the assortment in real time. Treat this as core infrastructure, not SEO.
2. Fund fewer bets, deeper. Concentrate spend on the two or three workflows where the value stack is largest for the specific business: on-site conversational commerce, AI pricing and assortment, or retail media. Get one workflow to production-grade before starting the next. The Economist has documented that AI operating costs are already rising fast enough to punish scattered investment.
3. Stand up retail media as a real P&L. Assign dedicated leadership, monetization logic, and integration with merchandising and pricing. In a feed-first world, always-on content and monetized attention are the same discipline.
4. Move to a three-month product cadence. Kill twelve-month roadmaps. Every ninety days, ship a working capability, measure the delta on a specific KPI, and reallocate.
5. Rewire decision rights before rewiring the tech. AI insight without a decision owner produces reports, not results. Assign a customer-journey P&L owner who can move pricing, content, and inventory on the same authority.

DIVIDER

The reset is not the growth. The reset is the model.

European retail’s AI story is not that growth returned. It is that the mechanism of growth changed. Consumers already delegate cognition to machines. They will delegate more control as trust, infrastructure, and evidence catch up. The retailers that will benefit are the ones that stop optimizing a channel and start operating a stack.

Intelligent Commerce is that stack. It is what turns 7 percent digital growth, an €842 billion European market, and a rapidly compounding agent-referred channel into results that show up in a specific company’s EBITDA. The board question in 2026 is not whether AI matters. It is whether the operating model is built to compound when it does.

Eric Pilkington is Chief Executive and General Manager of UST, where he leads Evolve, the firm’s intelligent-commerce platform for retailers and consumer brands. He writes on AI, operating models, and the economics of enterprise transformation.