When AI makes buying easier, who keeps the saving?

The incentive to build AI marketplaces is strong: a platform that connects buyers, suppliers and transaction services can make each more useful to the others. But making a transaction easier and making its terms better are different achievements. The sensible protection is not to avoid marketplaces. It is to retain the ability to compare them with direct alternatives — and to recognise when platform rules discourage suppliers from offering those alternatives.

A friend who sells replacement parts used to find Amazon an attractive proposition. It brought him customers he would otherwise have had to find, and the cost of reaching them seemed reasonable.

Over time, his calculation changed. Expenses described as optional increasingly felt necessary to sustain sales. He raised prices to recover some of the cost, then moved much of his business towards his own storefronts. With lower selling costs there, he could offer lower prices.

His particular frustration was the Buy Box: the prominent purchasing area associated with Amazon’s Featured Offers and its Add to Cart and Buy Now buttons. In his experience, losing that position made products difficult to sell economically. That is one seller’s account, not a finding about every Amazon business. But the distinction between being listed and being prominently offered is real. ¹

It is also worth remembering before handing purchasing authority to an AI system. A service can make the next transaction exceptionally convenient while gradually making alternative routes less attractive.

The question is not whether automation creates a saving. It is how much of that saving the buyer and seller get to keep.

Why the marketplace is tempting

Suppose you operate a platform that provides automated sales agents. Your customers want buyers. Bringing buyers into the same environment would give the sales agents more opportunities, while the growing supplier base would make the platform more useful to purchasing teams.

Identity checks, payment arrangements, approved contract formats and transaction records could make the proposition stronger still. A buyer would not merely be purchasing access to software. They would be joining a place where much of the awkward commercial preparation had already been done.

This is the network-effect argument: more suitable suppliers attract buyers; more buyers attract suppliers. It gives platform owners a rational reason to build marketplaces rather than remain suppliers of isolated automation tools.

That is an inference from the incentives, not evidence that every provider has a secret marketplace plan. Nor does the benefit require two autonomous agents. A buying agent dealing with a human salesperson could remove substantial work without asking the supplier to automate anything.

A well-run marketplace could genuinely improve the economics. Faster sourcing, fewer administrative exchanges and more reliable settlement may be worth considerably more than its fee. The concern begins when a valuable route becomes a route that is difficult to compare with anything else.

Optional on the rate card

There is evidence behind the concern about accumulating channel costs, although it needs careful handling.

In February 2023, specialist researcher Marketplace Pulse estimated, from a sample of sellers’ profit-and-loss statements, that Amazon charges could consume more than half of sales revenue, compared with roughly 40% five years earlier. The total included fulfilment, storage and advertising as well as referral fees. The article did not disclose a sample size. This is an indicative historical analysis, not a current universal rate — and certainly not a claim that Amazon took half of sellers’ profits. ²

The useful point was the composition. Marketplace Pulse reported that the commonly charged 15% referral fee had stayed unchanged while other selling costs grew. It also argued that paid visibility could be difficult to avoid, while acknowledging that advertising expenditure varied substantially between businesses. ²

eBay also increased selected UK business-seller charges in February 2026, while Amazon cut some European referral fees for lower-priced products. Marketplace economics do not move in only one direction, but sellers operate within charging structures that platforms can change as their markets develop. ^{3 4}

Nor are services free outside a marketplace. Sellers still need customer acquisition, storage, delivery and support. The relevant comparison is the cost of serving a customer through each channel, not platform charges against zero.

Nevertheless, contractually optional is not the same as commercially avoidable. An extra service can become part of the effective cost of sustaining a business even without appearing in the compulsory-fee column. Advertising expenditure and Featured Offer eligibility are separate mechanisms; sellers do not simply buy the Buy Box. ^{1 2}

When the cheaper channel becomes awkward

Amazon’s published explanation in June 2025 makes the external-price mechanism unusually clear:

“When a product is available for less at a reputable competing store, the offer may not become the Featured Offer.” ⁵

Amazon presents this as protection for customer trust: it should not prominently recommend an uncompetitive offer. Its seller guidance advises businesses with ineligible offers to consider adjusting prices to match or fall below the competitive external price. ^{1 5}

That is not simply the old contractual most-favoured-nation requirement under another name. Amazon dropped its UK and EU price-parity policy in 2013. A condition for preferred placement and an explicit contractual restriction are different things. ⁶

But they can create similar commercial pressure.

Consider an illustrative seller whose costs make £30 through a marketplace and £25 directly sensible for otherwise equivalent offers. If the cheaper direct price jeopardises prominent marketplace placement, the seller has an uncomfortable choice: absorb more cost on the marketplace sale, raise the direct price, or become less dependent on the marketplace.

The cheaper channel may then exist without its lower costs fully reaching the customer.

Exclusivity does not necessarily solve this. A sole supplier's own offers can still differ across channels. Amazon's current US guidance says that since September 2026 all sellers' offers can be evaluated for the Featured Offer; an individual offer can still fail to be selected, and a page may have no Featured Offer if none meets the requirements. External competitive pricing remains one factor in that assessment. ¹

This is more than a theoretical concern. On 5 February 2026, Germany's competition authority prohibited Amazon's price-control mechanisms on its German marketplace, subject to exceptions. Germany's Federal Network Agency says those mechanisms could remove offers or keep them out of the prominent Buy Box. Amazon said it would appeal; this dated action is not a claim here about a final appeal outcome or a rule applying everywhere. ⁷

The managerial issue is broader than the legal label. A supplier can be formally free to offer a lower price elsewhere while facing a substantial commercial reason not to.

The next Buy Box may not look like a box

An AI marketplace would not need to reproduce Amazon’s interface to create a similar dependency.

Imagine a purchasing system that considers only suppliers enrolled in its network. Perhaps participants receive easier identity checks, cleaner product data and faster approval. Those could be legitimate reasons to prefer them. But if the agent rarely looks beyond that network, the practical market could become smaller than the market available to the organisation.

The decisive position would no longer necessarily be a prominent button. It could be inclusion in the set of suppliers the agent is permitted, configured or rewarded to consider.

There are several possible routes to monetisation: transaction charges, supplier subscriptions, paid discovery, premium access or payment services. These are possibilities to examine in future contracts, not a claim that today’s providers all impose them.

The sequence is plausible: attractive initial terms encourage participation; participation improves the service; integration makes departure more disruptive; the platform gains room to charge more. But it is not inevitable. Competition, regulation and the ability to use several channels can interrupt it.

We should also resist a misleading contrast. Amazon already combines selling with services and transaction infrastructure. AI would not invent that model. The potential change is deeper delegation: the intermediary might help decide what is considered and accepted, rather than merely providing somewhere to complete a purchase.

Common rules need not mean a common owner

There is important evidence against an entirely closed future.

The Universal Commerce Protocol describes a shared framework through which platforms and businesses can exchange commerce information. Its specification separates participants and capabilities rather than requiring one company to own both ends. Sharing a compatible protocol is not the same as sharing a proprietary platform. The specification’s term capability negotiation concerns matching supported features, not bargaining over commercial terms. ⁸

Commercial implementations need not all follow an enclosed-marketplace model either. Shopify’s current documentation describes its ChatGPT channel as product discovery followed by checkout on the merchant’s own store, with no additional selling fee for that channel beyond standard processing fees. It explicitly excludes B2B-only products. ⁹

Google's May 2026 announcement described a cart spanning retailers and Google services, with options to check out or transfer to a merchant's website. Its September update reported some UCP-powered checkout already operating. These are consumer-commerce developments, not evidence of general-purpose autonomous B2B negotiation. ¹⁰

These examples matter because they show alternatives to requiring both parties to subscribe to the same agent provider. They do not guarantee neutral ranking, unrestricted access or low future charges.

Interoperability means systems can communicate. Independence means your commercial interests can prevail. The first can support the second, but does not establish it.

Compare channels, not just suppliers

The useful alternative is not an anti-marketplace agent. It is a buying system that is not bound to one marketplace’s view of the available deal.

Such a system should be designed to compare approved routes: a marketplace, a supplier’s own portal, a direct quotation and, where appropriate, a conversation with a human salesperson. The supplier need not install a matching agent before your organisation can benefit from automation.

The comparison must cover more than price. For software it might include renewal increases, minimum commitments, support and cancellation rights. For equipment, delivery certainty, warranty and spare-parts availability may outweigh a small discount. These are decision criteria, not claims that every agent can already negotiate them reliably.

Imagine, purely as an example, a £10,000 marketplace purchase against a £9,600 direct offer for equivalent equipment. The direct route is not automatically better. If it creates more than £400 of additional work or materially greater risk, the marketplace may earn its premium. If those costs are small and the terms equivalent, the direct route may preserve more of the saving.

Count charges once: a seller’s platform fee already reflected in its quoted price is not another buyer expense to add on top.

Sellers have the corresponding opportunity. They can retain marketplaces where those channels bring profitable demand while developing direct relationships elsewhere. Lower channel costs can sometimes let them charge customers less and retain more themselves. Neither permanent allegiance nor complete withdrawal is required.

There is a limit, however. An independent agent cannot guarantee a better outside offer if platform incentives discourage the supplier from making one. Freedom to search is valuable; it is not the same as freedom in the market being searched.

The outside-option test

Before delegating purchasing, ask the provider to demonstrate five things on a representative, low-risk requirement:

On a narrow screen, each check is shown as a separate card.

A controlled comparison is more revealing than a claim to be platform-agnostic. Keep supplier qualification, security requirements and spending authority intact. An outside option should be a credible alternative, not an instruction to buy from any website the agent finds.

Keep the market wider than the interface

My friend’s experience was not that a marketplace had never been useful. It was that the balance changed, and his alternative channels became more valuable.

That is the lesson worth carrying into automated buying. Adopt convenience where it earns its keep, but preserve the ability to notice when the economics have changed.

The important question is not only whether your agent is free to shop elsewhere. It is whether your suppliers are commercially free to offer you a better deal there.