— From traffic operations to agent coordination: the repricing of product data, supply orchestration and fulfilment

Figure 1 – Product discovery and cross-border fulfilment with AI agents
The easiest reading of the platform announcements that landed in November 2026 is “peak season again”. That reading is wrong. Promotions, rate cards, memberships and quarterly results are the surface. What changed is the division of labour along the transaction chain: who discovers a product, who decides the price band in which it appears, who presses the pay button on the shopper’s behalf, and who absorbs fulfilment cost and risk. All four are changing hands at once.
Two moves landed in the same week. Amazon folded listings from its low-price storefront Haul into standard search results, covering roughly 184,000 sellers — about 40 percent of active sellers on its US marketplace — most of whom never opted in; their inventory was synced automatically across channels. In the same week, OpenAI let businesses claim a profile inside ChatGPT, upload product catalogues carrying price and availability, and launch conversational Business Agents. One is a platform deciding where a seller’s product appears. The other is a model deciding which business a shopper sees. Sellers yielded ground at both ends.
The argument of this article is that the centre of competition in cross-border e-commerce is moving from winning traffic entry points to qualifying to be read and executed by agents. Platforms are consolidating the right to orchestrate supply while handing execution of the transaction to agents. Whether a seller stays in the game now depends on whether product data, fulfilment routes and operating rules are precise enough for a machine to act on. Four shifts below decompose that claim.
1. This is not a feature release — four roles have moved along the chain
Reading these developments platform by platform assumes the platform is the unit of analysis. It is not. The change sits in the chain itself. Discovery is moving from the search box into conversational interfaces. Orchestration is moving from sellers to platforms. Execution is moving from users to agents. And the bearing of cost — media risk in advertising, volatility during peak fulfilment — is moving from merchants to platforms and service providers.
Shopify’s Campaign Autopilot now charges on a per-sale basis: the merchant sets an acceptable cost per conversion, and the platform carries the media risk. AliExpress hands detail-page optimisation, campaign enrolment, route matching and customer service to an AI operations agent for its growth markets. Amazon lets Alexa for Shopping add items to a cart and complete payment on the strength of a single text message. Three different links, one direction of travel: the human retreats from operator to rule-setter, while the agent advances from adviser to executor.
Two common misreadings follow. If AI were merely a productivity tool, the 800,000 new eBay listings created by AI each day would be a quantitative change in automation, nothing more. But an agent that can pay on a user’s behalf changes the nature of the relationship from efficiency to authorisation. And if growth came from campaign traffic, US retail sales of $561.7 billion in October — up 3.57 percent year on year but only 0.15 percent month on month, the weakest monthly reading since November 2025 — would offer a wider window than it does. The aggregate market is not supplying one.
| Structural shift | Key facts this cycle | What it means for brands |
|---|---|---|
| Agents take over execution | Text-to-order shopping, business agents, AI-managed operations and per-sale ad billing all landed at once | Product data must be machine-readable and executable; rules replace manual operation |
| Platforms orchestrate supply | Haul folded into main search, roughly 184,000 sellers synced automatically; commission relief tilted to local sellers | Channel and price-band choices move to platforms; multi-market combinations become mandatory |
| Fulfilment repriced both ways | Southeast Asian direct mail down up to about 70 percent; US rates up 5.9 percent on average in 2027 | Localised fulfilment becomes an entry requirement; routes must be dynamically combinable |
| Growth shifts to the long tail | Growth-market transaction value up over 50 percent; livestreams and memberships bring new buyers and repeat purchase | Increments come from overlooked markets and relationship mechanisms, not mature-market traffic |
2. Who decides for the shopper: discovery and execution are coming apart
Three routes are now visible, and they are not compatible with one another.
Closed loop inside the platform. Amazon has opened text-based shopping to a subset of US users: Alexa for Shopping adds items and completes payment, with a rollout to millions of users planned over the coming weeks. Mercado Libre is testing its own AI shopping agent with a small group of users in Brazil, Mexico and Argentina, and has stated explicitly that it will not open its inventory to third-party agents for now. Its intention is to deliver a complete agentic shopping experience inside its own app.
An open discovery layer. OpenAI’s Business in ChatGPT runs the other way. Businesses claim a profile in Business Hub, upload feeds with price and availability, and — if they are eligible US e-commerce or online-service companies — publish a Business Agent. The back office reports mentions, profile views and website clicks; Shopify merchants can connect their store directly. The decisive trade-off is that OpenAI retired its direct-checkout attempt, including the shopping cart, in March this year. It gave up control of the transaction to compete for something else: who gets to be described accurately by the model. With roughly 1.2 billion weekly users, that discovery layer is an entry point in its own right. Claiming a profile does not buy ranking, however; mention frequency still depends on whether the information fits the question being asked.
Agents on the merchant side. In the first seven days of AliExpress’s overseas 11.11 campaign, daily transaction value in growth markets rose more than 50 percent year on year, and nearly 70 percent among merchants using the AI operations agent. More than 800,000 new eBay listings are now created by AI each day. Around 56,000 Mercado Libre sellers use its AI adviser to manage advertising, almost four times the figure three months earlier. Cainiao has brought AI and robotics into peak-season operations.
All three routes raise the same bar: product information must be readable, verifiable and executable by a machine. AI search connected to a structured product catalogue converts at roughly twice the rate of search relying on scraped or stale data. About half of AI-referred sessions land directly on a product detail page — 2.5 times the rate seen from conventional search — which removes the browse-and-compare funnel entirely. In a benchmark of AI shopping agents, close to half of realistic purchase tasks ended in failure, yet 54 percent of those failures still completed an order. An agent does not tolerate ambiguity the way a person does. It either drops you, or places the order with the wrong specification and the wrong price.
Generative engine optimisation gets a brand mentioned. Agent-readiness gets it selected and bought correctly. The first is won with content; the second with product master data — specifications, price, inventory, availability, delivery promises and return policy, all structured, current and consistent across channels. When a sync rule places the same item in two different price contexts, and when a business profile is quoted directly by a model, a single mismatch in price or stock stops being an operational blemish and becomes a failure of trust on the agent side.
3. Who decides where a product appears: supply orchestration is concentrating
Reading Haul’s integration into main search as “a low-price channel getting more traffic” understates it considerably. In two years, its selection has grown more than 25-fold to over 10 million items, and the median price of a seller’s best-selling Haul item is $9.98. Around 95 percent of Haul sellers are active, against roughly 13 percent across the US marketplace as a whole — a gap explained by the system syncing live fulfilment inventory rather than by recruiting new merchants. The sync terms are equally explicit: commission rates stay unchanged, promotional subsidies inside Haul are funded by Amazon, and a seller cannot delist an individual synced item — the only exit is to withdraw the entire account.
Taken together, these conditions mean Haul is no longer a separate store one can apply to join. It is a fixed price band inside the marketplace. Folding it into main search completes the transition: low-price supply now sits beside mainstream listings on the same results page, pulling the shopper’s reference price down systemically, while the brand owner never made a decision about entering that band at all.
The pattern repeats. Shopee Thailand cut sales commissions by up to 33.3 percent, but restricted the benefit to Thai local sellers, locally produced goods and locally sourced materials; electronics and cross-border stores are excluded, and the two benefit windows run to 31 May 2027. That is not a broad-based fee cut but a rule with an industrial-policy intent. In the US, TikTok Shop is piloting a paid membership, TikTok Shop Plus, offering more than a million selected users a 30-day trial before a $9.99 monthly fee — bundling free shipping, exclusive discounts and instant refunds into a subscription. Platforms are beginning to own the customer relationship rather than merely distribute traffic.
As platforms move from recruiting merchants to orchestrating supply, the merchant’s job shifts from choosing a lane to being assigned one. When a channel can sync your inventory automatically, when a market can exclude cross-border stores from a benefit, and when a membership programme can decide who sees free shipping first, the ability to combine multiple platforms, markets and fulfilment models stops being a strategy option and becomes a condition of survival.
4. Why the same item earns different margins in two places: fulfilment is being repriced in both directions
Cost is no longer a fixed line to be averaged across the month. It is a variable that sets the boundary of categories and markets — and it is moving in opposite directions.
Downward, in Southeast Asia. Since September, TikTok Shop has cut cross-border direct-mail rate cards across Thailand, Vietnam, the Philippines and Malaysia, with merchants saving up to about 70 percent on logistics. The steepest reductions are in the sub-200g band: up to 88.2 percent in Vietnam and 86.7 percent in Malaysia, 60 percent across all weight tiers in Thailand and up to 28.2 percent in the Philippines. On 9 November, sea-freight direct mail went live, initially for Singapore and Malaysia and later covering all categories in the Philippines; at 30 to 80 percent below air freight, it brings bulky and heavy goods into the feasible range for cross-border sellers. During Shopee’s 11.11 campaign, localised fulfilment drove an eightfold surge in orders across multiple categories, while express, same-day and cross-border Mall sales grew five to sixfold. Mercado Libre handled 258 million same- and next-day orders, up 49 percent, with 79 percent of parcels in its fast-shipping network delivered within 48 hours. Cainiao, meanwhile, activated peak-season coverage spanning both direct shipping and overseas warehouses.
Upward, in the United States. On 12 November, the US Postal Service filed a proposal to raise rates on four domestic product categories by an average of 5.9 percent from 17 January 2027, converting an expiring 8 percent transport surcharge into a permanent 11 percent fuel surcharge, to be reviewed in July 2027 against first-half fuel prices. International rates would rise 7.9 percent on average; the new surcharge does not currently apply to international products, and the proposal remains subject to regulatory approval. The same week brought a $6.5 billion loss for fiscal 2026.
Put the two sides together and the feasibility calculus changes for both lightweight parcels and heavy freight: last-mile costs harden in mature markets just as emerging ones open up. More importantly, localised fulfilment has moved from advantage to entry requirement. Direct mail, sea freight, overseas warehouses and local warehouses are no longer four alternatives to choose from — they are four routes to be combined in real time by market and category.

Figure 2 – From conversational discovery to product, warehouse and delivery
5. Where growth actually comes from: modest aggregates, long-tail and relational increments
Nine of eleven major US retail categories grew year on year in October. Electronics and appliances led at 9.3 percent, while furniture and home goods fell 2.62 percent. The aggregate is rising while categories diverge; the conditions for a rising tide to lift every category are gone. Australia shows the same rotation: in the twelve months to June 2026, Amazon’s consumer base grew to 9.5 million, Temu and SHEIN reached 5.1 million and 3.0 million, while eBay shed 389,000 users.
The real increment comes from two places. The first is long-tail markets that were never worth staffing. Across AliExpress, growth markets in South America, Europe, Southeast Asia and Africa all expanded: buyers in the Falkland Islands placed 5.5 orders per head during the campaign, Svalbard recorded its first 11.11 order, and GMV doubled in Mali and Norway. Small volumes, fragmented languages and rules, and prohibitive marginal operating cost are precisely what AI-managed operations compress towards zero. The gap between 50 percent and 70 percent growth measures not the capability of the model but the scale of demand released in the long tail.
The second is the mechanism that turns a transaction into a relationship. TikTok Shop’s UK live-commerce report finds that 61 percent of local small and mid-sized merchants earn more from a single livestream than from a full week of their other channels, and 24 percent more than a full month; 92 percent reach consumers they could not previously access, and 75 percent report higher repeat-purchase probability. The UK marketplace now hosts more than 6,000 livestreams a day, with live sales up 55 percent year on year. On 11.11, cross-border livestream hours on Shopee exceeded 36 million and short-video views passed 150 million, while VIP registrations grew sixfold and member spend eightfold. Mercado Libre posted third-quarter GMV of $23 billion, up 38 percent, with a 31st consecutive quarter of growth above 30 percent, 867 million items sold and more than 93.9 million unique buyers.
Set side by side, three conclusions hold. Content is not decoration on top of traffic; it carries both acquisition and repeat purchase. Membership is not a discount mechanism; it locks frequency into the system. And the fastest-growing markets are frequently neither the largest nor those previously targeted with the heaviest spend.
6. What brands should prepare: make the product something an agent can execute
If the above holds, the most certain investment over the next twelve months is not media budget but data precision. Four preparations, in priority order.
Unify product master data. Specifications, price, available inventory, delivery commitments and return policy must exist in machine-readable form and stay consistent across the brand site, marketplace stores, product feeds and agent interfaces. Generative engine optimisation gets a brand mentioned; agent-readiness gets it bought correctly. There is no shortcut here — it starts with complete fields on every SKU.
Make fulfilment composable. Break the cost and lead time of direct mail, sea freight, overseas warehouses and local warehouses into configurable combinations rather than binding them to a single provider. The scissors between falling Southeast Asian rates and rising US rates will persist; switching dynamically by category and destination matters more than negotiating one annual rate.
Structure content as an asset. Livestreams, short video and detail pages no longer serve human viewing alone; they must be quotable, parseable and reusable in product comparison. Explanatory content — use conditions, specification differences, selection criteria, return rules — is one of the few variables in the discovery layer that brands still control themselves.
Make operating rules delegable. Write down the boundaries of AI-managed operations: acceptable cost per conversion, price floors, brand presentation rules, escalation paths for exceptions. An agent executes rules, not intentions. Ambiguous rules produce divergent execution.
All four point to the same shift: the unit of competition is moving from how well a store is run to whether its product information can be executed by a system without ambiguity. Cleaner data and clearer rules make a brand easier for an agent to select. It is a migration from operational labour efficiency to data executability — the least visible part of this cycle, and the hardest to recover with a short burst of ad spend.
The discovery layer is opening, the transaction layer is closing, fulfilment costs are moving in opposite directions in two regions, and supply orchestration is concentrating in the hands of platforms. These four forces do not converge on a single end state: open protocols, in-platform closed loops and proprietary walled gardens will coexist for some time, and what merchants need is the ability not to be locked into any one entry point. One thing is certain. Once adding to cart and paying are done by agents, product data stops being a back-office field and becomes the only part of the transaction a brand still writes for itself.
