When the algorithm chooses (and maybe, in the future, buys) for us

 

Agentic commerce is rewriting the rules

For relationship brands, the risk is not invisibility. It is losing control of the conversation before the sale; and sometimes, of the sale itself.

There is a moment in the daily routine of millions of consumers that, until a couple of years ago, simply did not exist

A young woman picks up her phone. She does not type keywords. She does not open any website. She does not ask anyone on social media for advice.

She opens ChatGPT and types:

I need a serum for sensitive skin, sixty dollars max, something that won’t make me look shiny under makeup

Three seconds. Three options. One tap. Bought.

Your product was perfect for her. But she will never know. Because she did not exclude you. She simply chose not to choose.

She delegated the decision to an algorithm. And the algorithm did not know you, your brand, your products well enough.

The question that should keep a CEO up at night is not:

Why didn’t the AI pick us?

It is more radical:

Why do more and more people prefer not to choose at all? And why does delegation actually work?

The answer is simple, and precisely for that reason, uncomfortable: it works because it reduces cognitive effort: comparisons, compatibility, alternatives, reviews, all condensed into a few seconds.

It works because it reduces error: more precise questions generate more targeted purchases, fewer returns, fewer disappointments. At least in theory, because the principle holds only if you know exactly what you need and can communicate it clearly to the AI. And it works because it enables things that did not exist before: smart reordering, automatic substitution, end-to-end assistance.

But here is a distinction that matters, and that the market is only now beginning to understand. Consumers delegate discovery with enthusiasm: fifty million shopping queries a day on ChatGPT alone, according to OpenAI’s Economic Research team.

Checkout, however, is a different story. Conversion rates for in-chat purchases remain below 1%. According to Forrester, completing a transaction within an answer engine is the least adopted use case among regular users.

People trust the AI to help them decide. They do not yet trust it to spend their money.

This gap between delegated discovery and delegated transaction is the most important fault line in agentic commerce today. And for brands, it is both a warning and a window.

An infrastructure shift, not a channel

This is the point that many in our industry underestimate. Agentic commerce is not a “new channel”. It is not an evolution of search. It is an infrastructure shift, closer to what the web did to physical retail in the early 2000s than to what mobile did to the web a decade later.

The numbers are making this concrete. During the 2025 holiday season, traffic to retail sites from AI assistants grew roughly 700% year over year, according to Adobe Analytics. But this is no longer just a traffic story: conversions from AI sources are running 31% higher than from traditional channels, with revenue per visit up 254%. The signal has moved from volume to value.

Meanwhile, new infrastructure is being built. Google launched the Universal Commerce Protocol (UCP) in January 2026, an open standard co-developed with Shopify, Walmart, Target, Visa, Mastercard and over twenty partners, designed to standardize the entire commerce journey specifically for AI agents.

The most relevant competitor is already live. The Agentic Commerce Protocol (ACP), co-developed by OpenAI and Stripe and released as open source in September 2025, arrived four months before UCP. Where UCP covers the full commerce journey from discovery to post-purchase across the open web, ACP defines how an AI agent authenticates, tokenizes payment, and closes a transaction inside a conversational interface. Salesforce, Commercetools, and Etsy signed on as early partners.

The key update: ACP did not disappear when OpenAI walked back in-chat checkout in March 2026. OpenAI and Stripe confirmed they will continue developing it to power the new app-based purchase model: ChatGPT Apps from Instacart, Target, Expedia, Booking.com. The protocol is alive; its original use case has simply shifted.

For brands, the practical implication is straightforward: UCP is where you need to be to surface in Google AI Mode and Gemini; ACP is where you need to be to transact inside ChatGPT and its partner apps. Both require the same foundation (clean, structured, machine-readable product data) and the strategic posture is the same for both: build that foundation now, before a winner is declared.

What announcements can and cannot tell you

The protocol layer is new and it is real. And yet, what is announced as revolutionary today can collapse tomorrow. 

In late September 2025, OpenAI launched ChatGPT’s in-chat checkout with enormous fanfare, promising millions of shoppable products inside the chat. 

By March 2026, it had quietly pulled back: a dozen merchants live out of millions, no infrastructure to collect state sales taxes, transactions negligible. 

The lesson for C-level executives is not to ignore these signals, nor to chase them. It is to observe with discipline. The landscape shifts fast enough that last quarter’s breakthrough can become this quarter’s case study in overreach. 

Invest in foundations, not in the protocol of the moment.

At the same time, and this is the other side of the same coin, Amazon is quietly training hundreds of millions of consumers to delegate. Rufus, its conversational shopping assistant, has an estimated incremental sales impact exceeding ten billion dollars per year. Alexa+, free to all U.S. Prime members since early 2026, has tripled shopping activity compared to its predecessor. 

The habits being formed there do not stay there. A consumer who discovers how effortless AI-assisted shopping feels on Amazon will expect the same everywhere, including on your DTC site. Amazon’s influence, however, extends well beyond habit formation.

The double bind for relationship brands

For brands that have built their value on the relationship, particularly in luxury, premium beauty, fashion, design, hospitality, this scenario creates a double bind that must be understood clearly.

• The first is a narrative bind. When a generative AI model tells your brand story, it does not use your words. It reconstructs your identity from dozens of sources: reviews, communities, media, comparison sites, YouTube videos. If you control only your own website, you control only a fraction of your representation. At JAKALA we see this every day with J-Horizon, the tool we use to monitor how brands and products are narrated in AI responses: owned sites represent a minority share of the sources cited by the models. The AI draws from elsewhere. And elsewhere, your story might be told very differently from how you intended.

• The second is a transactional bind. If the purchase closes in chat, the buyer may never see your store — less brand experience, fewer behavioral micro-signals, less first-party data. And here a paradox emerges: the brand often remains the merchant of record, handling shipping, returns, and customer service, while the AI platform retains most of the pre-purchase intelligence. You win the sale. You lose the understanding of demand.
  

But the transactional bind is not monolithic. Three distinct models are emerging, and they carry very different implications.

• The first is fully delegated checkout, where the AI platform handles the entire transaction inside the chat. This is the model that was attempted and, so far, has failed. Operational complexity proved overwhelming, and consumer willingness to pay through a chatbot remained near zero.

• The second is what I would call "the bridge model": the AI handles discovery and the frontend transaction, while the merchant retains the backend checkout infrastructure. Shopify’s Agentic Storefronts, launched at the beginning of 2026, are the clearest example. Products are surfaced automatically to AI platforms through Shopify’s catalog infrastructure; the customer completes the purchase seamlessly within the chat, while the transaction is powered entirely by the merchant’s Shopify checkout working behind the scenes. The AI works as the storefront, but the brand keeps the register.

• The third is the protected proprietary experience, where the brand deliberately keeps its most valuable interactions outside the AI ecosystem entirely: membership, exclusive services, personalization, rituals.

The strategic question is not which model to choose. It is which model applies to which part of your portfolio. A luxury hero product and an entry-level replenishment item do not belong in the same conversation with an AI agent.

Faced with this scenario, the temptation is to react with a tactical reflex: “let’s optimize for AI”. But if the strategy stops there, it is cosmetic.

The real question for a C-level is different:

If the AI wins because it reduces effort and risk for the consumer, where are we generating effort and risk in our own journey?

Because agentic commerce does not steal customers from brands that work well. It steals customers from brands that work worse than the algorithmic alternative. If your DTC e-commerce is slow, confusing, lacking honest comparison and real consultation, the AI agent is simply better. And the consumer knows it.

This is the point that turns a threat into an opportunity: redesigning the owned experience to win the comparison with delegation. Configurators, conversational consulting, guided comparison, end-to-end post-sale support.

Not because the AI demands it. Because the customer deserves it.

For entry-level products and replenishment, consider opening up: enable the bridge model, focus on efficiency and volume. For hero products, protect the owned experience and push AI to generate qualified traffic, not to close the sale in your place. For the hybrid territory, use the AI as concierge and filter, but anchor brand identity to touchpoints you control. Make the algorithm work for you, not instead of you.

The metrics that matter

If there is one thing to bring to the board, it is this: “AI conversions” is not enough as a metric. It is like measuring the success of a retail store by counting only who enters, without asking who buys, who comes back, and how much it costs to make them return.

The metrics that align CFO and CMO are three:

• The first is the contributive margin per customer, inclusive of discounts, returns, and cost of service.

• The second is customer quality: repurchase rate, return rate, NPS, churn.

• The third, and probably the least obvious, is the loss of pre-purchase data, first-party data, and in some cases, of relationship itself: how much intelligence on consumer behavior stays outside the CRM when the decision happens in a chat.

The infrastructure that pays regardless of what wins

Being “citable” by AI is not just a CMO problem. It is a CTO problem too. When the AI’s answer becomes the shelf, the brand’s informational packaging changes radically: rich data, structured metadata, machine-readable content, accessible APIs.

In fashion, semantically rich data has already become the prerequisite for visibility and preference among the models.

But structure alone is not enough. Conversational queries do not look like keywords: a person no longer types “red jacket brand X,” they ask:

Make me look chic for an outdoor dinner in New York

In many categories, the majority of searches include no brand names at all. The very language through which a product is found is changing.

This is the only investment that pays regardless of which protocol wins. UCP, ACP, or something that does not yet exist. They all require the same foundation: clean, structured, intent-oriented product data. The brand that gets its catalog right is ready for every scenario. The brand that waits for a winner is ready for none.

One consideration for C-level executives operating in Europe: the EU AI Act becomes fully applicable in August 2026, yet no jurisdiction has enacted regulation specifically addressing agentic commerce. Who is liable when an AI agent makes an erroneous purchase remains in a regulatory void. Moving early on compliance is not just risk mitigation. It is an opportunity to establish trust where trust is becoming the scarcest currency.

The distribution of value

Agentic commerce is not a trend. It is not a channel. It is a structural redistribution of power toward whoever controls intent: who gets cited, who makes the shortlist, who closes without friction.

For relationship brands, the defense is not to resist. It is to become legible to machines without losing soul for humans. To govern the narrative across the ecosystem.

To decide where to make yourself purchasable and where to protect the experience.

To measure the impact with the right metrics: margin, quality, data.

The issue is not the AI. It is the distribution of value.

And whoever does not decide where to stand will find that someone else has already decided for them.