If you lead e-commerce or marketing technology at a large retailer, you have probably read more about agentic commerce in the past six months than about any other single topic. McKinsey projects $5 trillion in global agentic commerce by 2030. Bain says 15-25% of all e-commerce will flow through agent channels. The projections are real. So is the fact that the channel is early, with ChatGPT referrals accounting for less than 0.2% of e-commerce sessions today.

What most coverage misses: the marketing funnel did not disappear. It relocated. Discovery, consideration, comparison, and decision-making still happen. They happen inside the AI agent, not on your surfaces. And the infrastructure you need to regain visibility into that funnel is the same infrastructure your data team has been requesting for years.

How AI shopping agents change the e-commerce funnel

When a customer asks an agent to find running shoes for flat feet under $150, the agent queries multiple retailers, compares options, weighs reviews and specifications, applies the customer's constraints, and narrows to a recommendation. All of that happens before your systems register that the customer exists. Your first signal is the add-to-cart event. The customer has already decided.

Here is what that shift looks like across each stage:

The Agentic Commerce Protocol defines the handshake between agents and merchants. It secures the transaction. It does not transmit the journey. Merchants receive the checkout payload: product selected, quantity, price paid, shipping selection. Merchants lose everything that happened before: how the customer found the product, what alternatives were evaluated, what made this option win.

That asymmetry is the central challenge of agentic commerce. The transaction arrives complete but contextless, providing revenue without the behavioral intelligence that traditional e-commerce generates as a byproduct of every customer interaction.

Why the agentic commerce performance gap is an infrastructure problem

The consumer demand signal is not ambiguous. Salesforce research shows 39% of consumers already use AI for product discovery, and Adobe found AI traffic to retail sites grew 805% year-over-year during Black Friday 2025. Morgan Stanley puts it more concretely: 23% of Americans purchased something via AI in the past month. Adoption is outpacing most retailers' infrastructure readiness.

The performance gap tells a more useful story than the projections do. McKinsey's research shows AI-generated product recommendations achieve 4.4x higher conversion rates than traditional search. Yet actual agent commerce converts 86% worse than affiliate traffic. If AI recommendations are objectively better at matching products to intent, why does the channel underperform?

Because the bottleneck is not consumer demand or agent capability. It is merchant infrastructure. The analytics, attribution, and personalization systems on the merchant side were all built to observe a customer journey that no longer happens on merchant-controlled surfaces. Agent transactions arrive, but no one in your organization can tell you what happened before checkout, and your optimization stack has nothing to work with.

That is a solvable problem. And the solution looks a lot like what your data team has been asking for regardless of agent commerce.

What agentic commerce means for attribution, personalization, and retail media

You might be thinking: this is a 0.2% channel, I have bigger problems. That is a reasonable reaction, and it is also the exact reasoning that left retailers scrambling to build mobile commerce infrastructure in 2014 after dismissing mobile traffic in 2011.

The question is not whether agent commerce matters today. It is whether the infrastructure you are building (or renewing contracts for) over the next 18 months can handle a channel that strips away the data your entire optimization stack depends on.

Three specific consequences for decisions that are likely on your desk right now:

‍Attribution modeling. Multi-touch attribution assumes observable touchpoints across the customer journey. Agent-mediated purchases have one observable touchpoint: checkout. If your attribution vendor is pitching an upgrade, ask them how the model handles transactions with zero pre-purchase visibility. If the answer involves client-side JavaScript, it does not work for agent traffic. Agents do not load JavaScript.

‍Personalization investment. Your personalization engine learns from browse history, consideration patterns, and purchase behavior. Agent purchases provide the purchase but not the browse or consideration. Every agent transaction that enters your personalization system carries less signal than a traditional transaction. As agent share grows, your personalization models degrade unless you build a way to capture or infer the missing context.

‍Retail media measurement. If you operate or invest in a retail media network, the measurement problem is existential, not incremental. According to eMarketer, 55% of US advertisers already report inconsistent targeting and attribution from retail media networks. Agent commerce makes this worse because brand influence on agent recommendations cannot be tracked with existing RMN measurement infrastructure. When advertisers cannot measure impact, budgets move.

Server-side infrastructure for agent-ready commerce

Here is what makes agent commerce preparation different from most emerging channel investments: it does not require a speculative bet. The infrastructure that supports agent commerce is the same infrastructure that improves data quality, attribution accuracy, and personalization performance across every channel you already operate.

‍Server-side data collection. AI agents make API calls directly to merchant systems. They do not load web pages, render JavaScript, or accept cookies. If your data collection depends on client-side execution, you have a blind spot that grows with agent traffic. But client-side dependency is already a problem. Ad blockers, browser privacy restrictions, and consent management gaps already degrade your data. Server-side collection solves the agent commerce visibility problem and the data quality gaps that already cost you accuracy across every existing channel.

‍Identity resolution across surfaces. Agent-mediated purchases arrive without journey context. But the customer who bought through an agent is often the same customer who browsed your site last week, clicked an email yesterday, and visited a store last month. Connecting agent transactions to existing customer profiles turns a contextless checkout into a data point that enriches your understanding of a known customer. This is also the same identity resolution work that improves match rates, extends tracking beyond browser restrictions, and unifies cross-channel customer views for traditional commerce. One capability, compounding returns across current and emerging channels.

‍A centralized data control plane. The ability to capture signals at the point of collection, normalize them, and route them to downstream systems (analytics, attribution, personalization, retail media) is the architectural pattern that makes everything else work. It turns raw event data into something your optimization stack can act on, whether the event originated from a web browser, a mobile app, or an AI agent. The routing logic, normalization, and downstream destinations do not change. Agent commerce adds a new ingress point, not a new architecture.

‍Product data hygiene. Agents parse structured data, not marketing copy. Schema.org markup, clean product feeds, and consistent attribute formatting all matter for agent discoverability. Research from Mirakl found that 42% of customers abandon purchases due to insufficient product information, and that number likely understates the problem for agent surfaces, where data quality determines whether products appear at all. This does not require new infrastructure. It requires better discipline on infrastructure you already have.

Building a data strategy that works today and scales for agent commerce

Agentic commerce is a major trend in its early stages. How fast it scales, which protocols win, and how the agent ecosystem evolves are open questions that no one can answer with confidence today. That is fine. You do not need to predict the future to prepare for it.

The infrastructure that agent commerce requires is the same infrastructure that makes your current data stack more accurate, more complete, and more actionable. Server-side collection improves data quality now and captures agent signals as the channel scales. Identity resolution increases match rates across existing channels while giving you a way to connect agent transactions to customer profiles. A centralized control plane strengthens every downstream system and extends to new ingress points without requiring new architecture.

None of this requires building agent-specific infrastructure or betting on a particular protocol. First-mile data infrastructure strengthens every existing channel while extending naturally to new ones. That is the best preparation for an emerging channel: infrastructure that does not depend on predicting exactly how that channel evolves.

‍