A shopper opens an AI assistant and asks it to order household basics, find a birthday gift under $50, and check whether the shoes they viewed last week have dropped in price. Thirty seconds later, the agent has handled all three requests. The shopper hasn’t browsed a webpage or seen an ad banner. In agentic commerce, a retailer can make a sale and never receive a visit.
Retailers’ digital growth has long depended on visits. Over the past decade, much of that growth has come from marketplaces, loyalty and personalization programs, and retail media. Each reinforces the others. Marketplaces expand retailers’ assortment without adding inventory while generating data on how people shop across different trip missions. That data improves personalization and loyalty initiatives, and the resulting traffic supports retail media, whose high margins fund further investment. Together, they form the digital flywheel.
External AI agents—such as ChatGPT, Claude, and Gemini—change how that flywheel works. They increasingly sit between retailers and shoppers, influencing which products are considered and where purchases are made. Retailers, however, retain assets that agents do not control, including brand trust, proprietary data, access to third-party sellers and advertisers, and established loyalty relationships.
Retailers now have two jobs: remaining a destination that shoppers choose directly and winning shoppers’ evaluation when agents choose for them.
The Digital Flywheel Already Drives Retail Growth
We surveyed 50 of the top 300 global retailers and 100 of the top global brands, as well as interviewed retail and technology leaders. That research showed that each part of the digital flywheel contributes to online sales, customer acquisition, and profit.
- Marketplaces account for 26% of these retailers’ total online gross merchandise value (GMV), on average, rising to 35% among the top quartile. New marketplace customers generate 23% of retailer traffic and 30% among the top quartile.
- Loyalty and personalization programs account for an attributed 11% of retailers’ GMV, on average, rising to 15% among the top quartile.
- Retail media draws an average of 28% of its revenue from third-party marketplace sellers, increasing to 42% among the top quartile. Ninety-five percent of retailers see retail media as a major source of future profit growth.
While each capability works on its own, the flywheel delivers compounding benefits. (See Exhibit 1.) Amazon and Alibaba’s Tmall established this model at scale, and major retailers—including Walmart, Target, Lowe’s, Kingfisher, and Tesco—are building versions of their own.
Joe Cano, senior vice president of digital commerce and retail media at Lowe’s, describes how the parts depend on one another. “A marketplace expands your assortment, which drives more traffic, and more traffic means better return on ad spend and engagement for media, because you’re matching the right customer to the right product. You can’t scale a marketplace without a media network, and you can’t scale a media network without a marketplace.”
The Flywheel Now Serves Two Types of Customers
Retailers designed the flywheel for people who are browsing, clicking, and choosing. As shoppers delegate those activities, external agents increasingly behave like a second shopper. The agents shape discovery and comparison, and they evaluate retailers on product data, prices, availability, delivery, returns, and offers that they can interpret.
Agents also create a new feedback loop for trust. A shopper may tell an agent to shop first with a trusted retailer. If the retailer performs well on the criteria the agent evaluates, that preference holds. If the retailer falls short, the agent may recommend alternatives, influencing both the shopper’s trust and future instructions.
Agentic commerce remains at an early stage. Even so, 81% of consumers in the US and 53% in Europe expect to use AI agents when shopping. By 2030, more than half of e-commerce spending could be agent-assisted.
BCG’s analysis of leading European brands and retailers found that referrals from external agents accounted for less than 1% of website visits in late 2025, but with recent growth rates, that share could reach 5% by the end of 2026. Referral counts understate how much traffic agents drive. A shopper who browses and chooses inside an external AI platform and then comes to the website on their own is not counted. With those visits included, agents’ effect on website traffic can be up to four times the referral figure. In the US, where the shift is further advanced, the same analysis estimates that AI-driven traffic could overtake organic search traffic by as early as 2028.
Retailers are preparing. Sixty-eight percent of retailers surveyed expect agentic commerce to have a significant or transformational effect within two years, and most see an opportunity to reach more customers and generate more transactions. More than one-third are running agentic commerce pilots, and 18% are deploying these at scale, while 24% have yet to move beyond assessment. Nordstrom offers an example of a staged approach, building capabilities early and expanding them when the demand is clear and the customer experience meets its standards. (See “Nordstrom Tests, Learns, and Stays Ready.”)
Nordstrom Tests, Learns, and Stays Ready
That discipline shaped the retailer’s approach to agentic commerce. Of the four opportunities it identified, Nordstrom focused on two in the short term: greater visibility on external AI platforms and AI-assisted discovery on its own website. The third, agentic checkout, is largely built and will be switched on when customer demand warrants. The fourth opportunity, a dedicated presence inside a platform such as ChatGPT, isn’t a current priority, as other industry attempts have yet to show compelling results.
To become visible on external platforms, Nordstrom first had to fix its data. The company cleaned and enriched its product information, resolved technical problems that had kept the catalog out of agent results, and integrated its product feeds with OpenAI, Google, and others. Nordstrom reports strong early results on agentic visibility, and it holds third-party sellers in its Mirakl-powered marketplace to a high standard for product data to help them get discovered by agents as well.
The company also took a differentiated stance on its loyalty benefits based on the platform. “If we are the merchant of record, we’d be willing to expose all of our loyalty benefits and let the customer complete a frictionless experience outside of our own website,” says Almeida. On its own website, Nordstrom has steadily advanced personalization, extending it to anonymous visitors. That helps turn agent-referred traffic into longer-term customers.
To support agent-assisted discovery on its website, the retailer has launched its own shopping agent, built on its proprietary data. Early testing shows the agent is driving meaningfully higher conversion among shoppers who use it, though Nordstrom cautions the results may partly reflect selection bias. For now, the agent stays in the background while the experience improves.
Retailers Play Two Games, Destination and Evaluation
Shoppers are already handing discovery and evaluation to agents. Completing the purchase is another matter. Just 4% of consumers would let an agent finish a purchase on its own, even though shopping-related use of generative AI rose 35% in nine months and more than 60% of consumers say they trust what it tells them. The agent is becoming the most influential voice in the shopping journey, while the checkout stays with the shopper. The battleground today is the recommendation, not the transaction.
Retailers need to compete for both the shopper’s visit and the agent’s recommendation. Competing for the visit is a destination game, in which retailers give shoppers reasons to shop with them directly. Competing for the recommendation is an evaluation game, in which retailers must win the agent’s selection. The same flywheel supports both games, and each puts different demands on marketplaces, loyalty and personalization programs, and retail media. (See Exhibit 2.)
The destination game matters because agent-led shopping can reduce direct traffic, customer-data capture, and the value of onsite retail media (advertising on a retailer’s own website or app). Losing the customer relationship is retailers’ most common concern, cited by 91% of respondents. The risk is greatest in low-consideration categories such as household consumables and repeat orders, which are easiest to delegate. Fashion, beauty, and luxury remain less exposed because discovery plays a larger role.
Agent-routed demand is also unlikely to stay free. Every intermediary that has come between retailers and shoppers online has eventually charged for the position. Comparison websites and paid search did so, and online travel agencies now take roughly 15% to 25% of each booking. Retailers should prepare for agents to follow and include that cost in their margin models.
Hotels provide a useful model. Operators use travel websites to reach potential customers they may not otherwise find while also investing in direct booking, member rates, and loyalty programs to keep their highest-value repeat guests in their own channels.
Retailers need the same balance. To remain a destination, a retailer needs a differentiated onsite proposition. Unique products, best-in-class prices, loyalty benefits, and a tailored onsite experience all give shoppers a reason to come direct. Many retailers have built their own shopping agents to make that experience more engaging and more personal. Amazon’s Alexa for Shopping, Walmart’s Sparky, Lowe’s Mylow, and Instacart’s Clementine help shoppers discover products or build carts by drawing on the respective retailer’s data and expertise. Bol, a leading marketplace in the Netherlands and Belgium, has embedded its agent throughout the shopping journey rather than treating it as a parallel shopping gateway. (See “Bol Builds Its Own Agent into the Customer Journey.”)
Bol Builds Its Own Agent into the Customer Journey
“We didn’t want two separate gateways to the customer,” says Joris Scheepens, chief advertising and new business officer at bol. “The agent needed to be fully embedded in the traditional shopping journey, not a parallel path. A customer comparing products should be able to move into the agent naturally and move back out just as easily.”
Bol also built retail media into the agent from launch. That decision allows the company to begin testing how advertising can work in a conversational environment while customer behavior is still developing. “The whole industry is trying to figure out how advertising works in a conversational environment. We want to be at the forefront of that,” says Scheepens.
The agent is part of a broader effort to strengthen bol as a destination in its core categories. The company is adding expert advice, inspiration, and contextual product comparisons that go beyond price, while expanding its loyalty proposition. A better understanding of what customers value in a particular shopping moment can also make advertising more relevant and impactful.
Bol has drawn firmer boundaries around external agents. It sees them as an increasingly important source of product discovery but intends to retain the transaction. “We’re keeping agents to the discovery part of it,” Scheepens says. “We don’t plan to take the transaction offsite.”
The company will also be selective about where it participates, focusing on platforms its customers use and whose approach to brand trust aligns with its own. On those platforms, bol plans to compete on the strengths it already possesses, including its brand, assortment, seller base, and product data.
Together, these choices establish a clear division of roles. Bol’s own agent strengthens the direct customer experience, while external agents extend discovery without taking control of the transaction. Both depend on bol continuing to give customers a reason to choose it. Scheepens concludes, “With the Shopper Agent, we’re not building a new destination. We’re making bol more helpful in every shopping moment, while continuing to earn the customer’s attention and trust every single day.”
Winning the evaluation game means consistently appearing at the top of an agent’s recommendations across likely search prompts. A strong proposition is the starting point, but an agent can only recommend what it can find, so retailers need to make themselves discoverable. Sharing and surfacing data through feeds does the most to achieve that. It also brings risk, because the proprietary first-party data behind a retailer’s own site is what gives it an edge as a destination over a third-party agent.
Retailers Decide What Data Agents Can Access
Retailers need a consistent strategy across the flywheel that makes their products visible to agents without surrendering the customer data on which their loyalty, personalization, and retail media businesses depend. Product, inventory, pricing, offer, and fulfillment data should be accessible to agents, while customer identities, preferences, loyalty histories, and behavioral data remain under the retailer’s control.
Most retailers surveyed are already taking a position on their approach to external agents. Forty-two percent are collaborative, structuring their data so that any agent can discover and retrieve their products. Another 32% are selective, giving only chosen agents access through direct data feeds. Just 6% are wholly defensive and block agents from their data, while 20% are still deciding.
How much retailers share will likely depend on their ability to attract shoppers directly. Retailers with a strong destination appeal can restrict access and steer consumers toward their own shopping agents, as Amazon has done. Those that depend more heavily on agents for reach may need to share more. The right balance will differ by retailer and should allow marketplace and retail media businesses to compete for agent-led demand while preserving the customer insights that support loyalty and personalization. (See Exhibit 3.)
Marketplaces, Loyalty, and Retail Media Support Both Games
Each part of the flywheel must help retailers keep shoppers coming to them directly and win recommendations from agents.
Marketplaces must adapt. The marketplace channel remains a top-three priority for more than two-thirds of the retailers and brands surveyed, and respondents see it as central to winning in an agentic world. To win the evaluation game, retailers need a broad assortment and clean, current data on products, prices, inventory, and fulfillment. To win the destination game, marketplaces must serve a shopper’s whole mission and fill enough of the basket to encourage consolidated purchasing. Marketplaces are essential to meeting both needs, but few are built to do so today.
Only 18% of retailers consider their marketplaces ready for agents. Fragmented legacy systems, poorly structured metadata, and the lack of an integration layer make it difficult for agents to retrieve and act on marketplace information.
The standard for agent readiness is also rising. Retailers initially focused on making product pages easy for AI platforms to crawl, as they did for search engines. Major AI platforms are pursuing direct integration with retailers so that agents can retrieve live information on price, availability, delivery, and returns. Retailers need to make their catalogs available through data feeds and application programming interfaces and adopt emerging standards. Sixty-seven percent of those surveyed plan to enrich their product data, and 58% plan to provide real-time availability within two years. (See Exhibit 4.)
Agents weigh assortment, price, and execution as well as data. Referrals from external agents therefore depend on a differentiated assortment, with unique products from exclusive sellers both local and international; on sharp pricing, through the buy box and incentives such as lower take rates on key value items; and on an owned fulfillment capability that competes on delivery. A retailer that can onboard a long tail of third-party sellers quickly, enrich their data, integrate the data with agents, and hold sellers to delivery and returns standards can present a better answer to an agent.
Technology partners see the same requirements. “Winning in an agentic world requires more than intelligent interfaces,” says Anne-Claire Baschet, chief data and AI officer at Mirakl. “It requires trusted data, relevant choice, seamless fulfillment and after-sales experiences, and the infrastructure to connect the right AI capabilities as the market evolves.”
Loyalty and personalization must become richer for shoppers and quantifiable for agents. Nearly half of retailers are open to honoring loyalty benefits, offers, and points when an agent handles the purchase. At evaluation, member value needs to be expressed in terms agents can calculate and compare, such as member-only prices, a guaranteed delivery window, or the cash value of points earned. Early access and other experiential rewards are harder for an agent to value. At the destination, retailers need to offer benefits that agents cannot easily reproduce, including surprise rewards, games, communities, exclusive experiences, and real-time offers. Only 14% plan to give external agents the same benefits that are available through their own channels, while 58% would offer a lower tier.
“In an agentic economy, a retailer’s greatest advantage is the direct, trusted relationship it has built with its shoppers,” says Paul Tepfenhart, director of global retail strategy and solutions at Google Cloud. “Autonomous agents must do more than simply transact. They must understand and honor earned loyalty, member benefits, and individual preferences.”
Personalization gives retailers another advantage in the destination game. Years of data on purchase frequency, basket composition, and brand affinity allow retailers to tailor offers and experiences in ways that an external agent cannot replicate. That advantage can erode if agent-referred purchases remain anonymous. Retailers need to use registration, loyalty enrollment, and preference capture to turn those purchases into identified, consented relationships, and then act on the data quickly enough to improve the next offer or recommendation. Over the next two years, retailers plan to deepen both their personalized propositions and the capabilities behind them. (See Exhibit 5.)
Retail media should monetize the engagement and fund the flywheel. Retailers look to retail media for much of their future profit growth, and 65% of brands surveyed expect to spend more as shopping agents become common, while only 12% expect to spend less. As discovery moves beyond retailers’ websites, growth from traditional banner ads and sponsored search is expected to slow, while agentic, social media, video, and other offsite formats are likely to gain importance.
To win at the destination game, retailers must bring sponsored prompts and recommendations into their own shopping agents as Amazon, Walmart, and bol have begun to do. Onsite formats also need to change. Shoppers referred by an agent come with higher intent and browse less, so retailers need formats that engage them quickly, such as video and user-generated content. Retailers should also extend retail media to other channels and stores, so that advertisers can reach shoppers through one program.
Lowe’s is already moving in that direction. “We’re future-proofing retail media by exploring opportunities beyond onsite formats,” says Cano. “That includes testing new experiences such as in-store activations as we learn where retail media can create the most value for both customers and advertisers.”
To win at the evaluation game, retailers must secure sponsored visibility on external AI platforms for the brands that advertise with them. In both games, they should close the loop by connecting advertising exposure to purchases and proving that the advertising generated incremental sales. That ability, built on proprietary customer and transaction data, remains retail media’s principal advantage.
Most retailers are strengthening their data, targeting, and measurement capabilities. Among those that have already launched these capabilities, 65% report better campaign performance, 55% report additional third-party-seller demand, and 40% report higher retail media pricing. (See Exhibit 6.) Sponsored recommendations and visibility on external AI platforms, however, remain largely on retailers’ two-year roadmaps.
Brands are ready to move as well. Thirty-two percent of brands are very willing to join agentic pilots, and another 38% would participate given a clear business case. But brands want proof of incremental rather than attributed sales before investing materially. Retail media is therefore the part of the flywheel most exposed if that proof doesn’t materialize.
Together, the three parts reinforce one another in an agentic world as they do today. Marketplaces protect direct traffic and create first-party data that grows more valuable as agents handle discovery. Loyalty and personalization use that data to give shoppers an onsite experience that no third-party agent can reproduce. Retail media monetizes the traffic and funds the investment that keeps the proposition competitive. A retailer that already runs the flywheel well therefore starts agentic commerce ahead.
Where Retailers Should Focus Now
Retailers should prioritize moves that strengthen their direct appeal to shoppers and improve their chances of being recommended by agents. These are no-regret actions, because they pay off regardless of how quickly agentic discovery and commerce develop. (See Exhibit 7.)
- Build the data foundation by managing the flywheel as one system. Map the opportunities and risks by category, shopping mission, and customer group. Align marketplace, loyalty, personalization, and retail media on one data foundation and one policy for what data agents can access and what is protected.
- Make the marketplace easy for agents to use and trust. Provide clean, structured, real-time information on products, prices, inventory, and fulfillment through data feeds and application programming interfaces. Strengthen assortment through unique or exclusive seller relationships, use incentives to encourage competitive seller pricing, and invest in owned fulfillment so that delivery is competitive on speed, reliability, and cost.
- Use loyalty and personalization to serve both customers. Express member benefits in terms external agents can recognize and compare, preserve exclusive rewards and experiences for shoppers who come to the retailer directly, and turn agent-referred traffic into consented relationships.
- Extend retail media beyond traditional onsite formats. Test sponsored visibility on external AI platforms, embed sponsored content within retailers’ own shopping agents, adapt onsite retail media to higher-engagement formats, expand advertising to other channels and stores, build full-funnel omnichannel experiences that third parties cannot replicate, and prove incremental sales in each channel.
- Keep building a differentiated onsite experience alongside all of this. Give shoppers a reason to come direct, with a tailored journey powered by the retailer’s own shopping agent, own-brand products they cannot buy elsewhere, and sharp, trusted pricing.
Retailers that got the flywheel right over the past ten years earned higher growth, profitability, and valuation multiples. Those that now adapt the flywheel to serve external agents can compound those gains and build a lead that competitors find harder to close.
The result could lead to a clear split between retailers that build one flywheel for shoppers and agents and those that keep building for shoppers alone. Retailers in the first group can turn their data, assortment, and brand relationships into a lead that widens with every sale, with many likely to run shopping agents of their own.
Retailers in the second group could find agents standing between them and their customers and, over time, become interchangeable, margin-compressed fulfillment nodes.
Seventy-nine percent of the retailers and brands surveyed expect those that prepare the flywheel for agentic commerce to pull further ahead. For retailers that have not started, the cost of delay rises every quarter.
