When Commerce Stops Belonging to the Storefront
Issue 18 examined the measurement problem created when AI shapes a purchase before the customer reaches a merchant. The immediate question was whether businesses could still identify what influenced demand when the click no longer represented the beginning of the buying journey. This week adds the next structural layer. The purchasing journey is not merely becoming harder to measure. It is being relocated into interfaces that merchants do not own, including AI assistants, payment applications, search results and customer-service systems.
This matters because digital commerce was built around a stable division of roles. Search engines generated discovery. Social platforms generated attention. Merchant websites explained products. Payment providers completed transactions. Support teams handled problems after purchase. That division is now collapsing. The same interface can identify a need, recommend a product, apply an offer, complete checkout and resolve the resulting order. Commerce is becoming embedded inside the systems people already use rather than remaining a separate destination they intentionally visit.
The important shift is not simply that customers can buy without opening a retailer’s website. It is that transaction control is moving towards whoever owns the active interface at the moment intent becomes actionable. Merchants may continue supplying the product, fulfilling the order and carrying much of the commercial risk while another system controls the recommendation, interaction, checkout environment and customer context. The next competition in AI commerce will therefore not be limited to product selection or transaction execution. It will be a competition to control the surface where commercial intent is converted into action.
Cross-Border Commerce Is Becoming an AI-Native Service Layer
On 21 July, global ecommerce infrastructure company ESW launched ESW Agentic Commerce, initially integrating with Microsoft Copilot. The service allows brands to optimise their catalogues for AI platforms and support product discovery, checkout and payments inside AI-powered shopping experiences. ESW says it works alongside a retailer’s existing ecommerce infrastructure and plans to extend the capability to additional agents.
The obvious interpretation is that another company has connected products to an AI assistant. The more important interpretation is that complex international commerce is becoming available as an invisible service beneath an external interface. Cross-border transactions require much more than showing a product and collecting payment. They involve local currency, duties, taxes, fraud controls, payment acceptance, regulatory obligations, delivery and returns. Historically, much of this complexity lived inside a merchant’s international storefront or dedicated market infrastructure. ESW is now positioning that operating layer beneath Copilot, allowing the assistant to become the customer-facing environment while ESW handles the commercial machinery behind it.
This changes the role of the storefront. A brand may no longer need to reproduce its complete international buying experience wherever demand appears. Instead, it can expose a catalogue and rely on an infrastructure provider to translate the product into a locally executable transaction inside an AI interface. That lowers the cost of entering new machine-mediated channels, but it also makes the merchant increasingly dependent on external systems to represent availability, total cost, delivery confidence and purchasing rules correctly.
The second-order implication is that cross-border providers may evolve from checkout vendors into distribution gateways. The provider that can make a merchant transactable across multiple AI environments gains influence over which markets can be served, which commercial rules are presented and which assistants can complete the purchase. Merchants will still own inventory and fulfilment, but the infrastructure provider may control whether that inventory can travel into new demand environments at all.
Mental Model Update: A storefront stops being the centre of international expansion when the entire cross-border transaction can be delivered as infrastructure inside someone else’s assistant.
Source: ESW Launch
The Payment Application Is Becoming a Demand Surface
Rezolve AI announced a partnership with Zilch on 21 July to embed its agentic commerce infrastructure into Zilch’s payment platform, which serves nearly six million customers. The partnership is designed to bring personalised retailer engagement and offers into a financial application that already directs more than $3.3 billion in annual spending to partner merchants.
This is structurally important because payment companies traditionally entered the customer journey near its end. They authorised a purchase after the customer had already chosen a merchant and product. Zilch’s position allows that sequence to be reversed. A payment application can now use purchasing context, affordability signals, retailer relationships and AI-led recommendations to influence what the customer buys before payment begins.
Consider the difference in a simple household purchase. A conventional payment provider sees that a customer bought a washing machine for £600. A payment platform with an active commerce layer may know the customer’s available spending capacity, preferred repayment structure, previous purchases and current merchant offers before making a recommendation. It does not merely approve the chosen washing machine. It can shape which washing machine becomes financially practical and which retailer receives the order.
That creates a new form of distribution power. Search platforms historically controlled commercial discovery because they knew what customers were looking for. Social platforms influenced demand because they knew what customers were interested in. Payment platforms may now become equally powerful because they understand what customers can realistically purchase and how they prefer to pay. Their position combines intent with financial feasibility.
The second-order implication is that merchant acquisition budgets may begin moving towards financial interfaces. Retailers could pay not only for visibility or traffic, but for inclusion in personalised purchasing opportunities generated inside wallets, credit platforms and banking applications. This would blur the line between payments, advertising, loyalty and retail media. A payment company would no longer earn only when a transaction occurs. It could influence which transaction occurs.
Mental Model Update: Payment providers stop being neutral rails when they can recommend the merchant, shape the offer and determine which purchase fits the customer’s financial context.
Source: Zilch Partnership
Product Search Is Compressing the Store Into a Recommendation
Kuaishou introduced an AI shopping assistant directly inside its ecommerce search results. Users can describe what they need in natural language, receive a horizontally arranged selection of qualifying products and add an item to their cart without working through long product descriptions or manually comparing specifications. The company’s broader search upgrade uses an AI agent to support product discovery and shopping decisions.
The interface appears simple, but the commercial change is substantial. Traditional ecommerce search presented merchants with a list of opportunities to persuade. Customers could open multiple product pages, inspect images, read claims, compare reviews and reconsider their priorities. The merchant controlled at least part of the evaluation environment. Kuaishou’s assistant compresses that process. The customer expresses the need once, the system interprets it and the product list arrives already filtered.
This reduces the value of information that does not survive machine comparison. A beautifully designed product page may still matter after a product is selected, but it has less influence when the assistant eliminates the product before the customer sees it. Product attributes, price, stock, delivery, reviews and compatibility become selection inputs rather than supporting information. The system decides which facts deserve attention and which products enter the customer’s consideration set.
The effect is particularly important for mid-market and lesser-known brands. Traditional digital shelves gave merchants several ways to earn attention through creative imagery, marketplace advertising, promotional placement and customer reviews. An AI-mediated result may narrow the field before those mechanisms can operate. Merchants will need to know not only whether their products appear in search, but why the system judged them suitable for a particular request.
The second-order implication is that product-page optimisation may separate into two disciplines. One will continue serving humans after selection. The other will serve machines before selection by supplying clear constraints, evidence, availability and comparison-ready claims. The machine-facing layer will increasingly determine whether the human-facing layer is ever reached.
Mental Model Update: The product page loses its persuasive power when the assistant decides which products deserve to be opened before the customer sees the shelf.
Source: Kuaishou Assistant
Customer Service Is Becoming an Execution Environment
OpenAI introduced Presence on 22 July as an enterprise platform for deploying voice and chat agents across customer and internal workflows. Presence agents can answer questions, use company systems, take approved actions and escalate to people under defined policies and guardrails.
This is not presented as a commerce product, but its commercial significance is considerable. Customer service has historically been treated as a post-purchase function. It answered questions, solved delivery problems and processed returns after revenue had already been created. An agent that can access company systems and take authorised actions changes that boundary. The same system that answers a product question can check stock, modify an order, apply an approved remedy, arrange a replacement, preserve a subscription or move a customer towards another purchase.
That means support interactions can become live commercial environments. A customer might contact a business because an item is unavailable, a subscription is too expensive or a delivery date no longer works. A conventional chatbot provides information or transfers the case. An action-capable agent can search alternatives, apply policy, change the order and complete the resolution without leaving the conversation. The support channel no longer merely protects revenue after checkout. It can redirect and recreate revenue during the interaction.
This changes how businesses should think about customer-service data. Service conversations contain unusually strong signals: urgency, dissatisfaction, replacement needs, budget concerns, product failure and changing intent. Once agents can act, those signals become operational inputs. The agent can connect the reason for contact with inventory, pricing, fulfilment and account permissions in real time.
The second-order implication is organisational. Commerce teams, support teams and operations teams have traditionally used separate systems, goals and ownership models. Action-capable customer agents make those separations harder to maintain. A support agent may affect retention, conversion, inventory movement, refunds and lifetime value within the same interaction. Businesses will need governance that follows the action rather than the department that historically owned the channel.
Mental Model Update: Customer service stops being a cost centre when the same conversation can resolve a problem, alter an order and create the next transaction.
Source: OpenAI Presence
The New Commerce Surface Inherits Protocol Risk
Research published on 23 July examined protocol-level attacks across agentic commerce platforms and identified 33 vulnerabilities in the interaction layer between agents and commerce services. The study argues that these weaknesses are structural rather than model-dependent, meaning a more capable AI model does not automatically remove them. The researchers demonstrated recurring failure patterns across independently built platforms and described attack chains capable of redirecting payments.
The distinction matters. Much of the security discussion around shopping agents focuses on whether a model can be manipulated through malicious instructions. That is important, but it is only one risk layer. A commerce agent also relies on protocols that pass credentials, instructions, merchant information, approvals and payment details between systems. Weaknesses in those connections can remain exploitable even when the model interprets the customer’s request correctly.
This becomes more serious as transactions move into external interfaces. When a customer buys inside an assistant, payment app or support conversation, several systems may participate without becoming visible to the customer. One system interprets intent. Another retrieves the product. Another manages identity. Another executes payment. Another fulfils the order. Every connection becomes part of the transaction boundary.
The old storefront concentrated many commercial controls inside one environment. Embedded commerce distributes those controls across multiple companies and protocols. That improves reach and convenience, but it creates more places where instructions can be altered, credentials can be exposed or responsibility can become unclear. The interface may appear unified while the transaction underneath becomes more fragmented.
The second-order implication is that interface expansion will eventually require certification at the protocol level. Merchants and payment providers will need evidence that an agentic channel protects transaction integrity across the full chain, not merely assurances that the underlying model is safe. Security reviews will have to examine the connection between systems as closely as the intelligence operating above them.
Mental Model Update: A trusted AI interface does not create a trusted transaction when the protocols beneath it can be exploited independently of the model.
Source: Protocol Research
The System That Is Emerging
This week’s signals reveal an embedded commerce surface layer. It sits above merchant infrastructure but below the customer’s visible experience. Its role is to take intent from whichever interface currently holds the customer’s attention and convert it into a commercially executable outcome.
The significance of this layer is that it does not belong naturally to merchants. It can be owned by an AI assistant, payment platform, marketplace, bank, customer-service provider or cross-border infrastructure company. Each can become the environment in which products are interpreted, compared, financed and purchased. The merchant contributes inventory, commercial rules and fulfilment capacity, but the surface owner controls how that supply is presented and acted upon.
Three old assumptions are becoming unreliable:
- The merchant website is where product consideration takes place.
- The payment provider enters only after the customer decides.
- Customer service begins only after commerce has ended.
The emerging model replaces these boundaries with continuous commercial interfaces. Discovery can lead directly to checkout. Payment context can generate discovery. Support can alter or produce transactions. International infrastructure can make products executable inside third-party assistants. The customer experiences one interaction while multiple systems coordinate beneath it.
Control therefore moves towards the company that combines three capabilities: persistent customer context, permission to act and access to executable merchant supply. An AI assistant may understand intent but lack payment authority. A payment application may understand affordability but lack rich product knowledge. A merchant may own the product but lack access to the active interface. The strongest platforms will connect all three.
For operators, the strategic question is no longer limited to whether their company has an AI shopping capability. It is whether their products, policies and transaction systems can participate safely across commercial surfaces they do not control. This requires decisions about catalogue exposure, pricing consistency, checkout portability, customer ownership, service permissions, data sharing and protocol security.
It also creates a new dependency risk. Every external surface that can generate a transaction can become a gatekeeper. A platform may control ranking, apply its own eligibility rules, change the commercial presentation or impose new fees for access to demand. Merchants that treat these channels as simple traffic sources may discover too late that they have become infrastructure suppliers to someone else’s customer relationship.
Core Truth: Commerce power moves to the interface that can understand intent, access supply and complete the transaction without sending the customer elsewhere.
