Agentic Commerce: From AI Experiments to an AI-Commerce Operating Model

Three agents, one operating model: How AI in e-Commerce actually starts to add up.

Over the past weeks in this newsletter, we looked at three pieces of the next wave in digital commerce.

First, the Catalog & Product Agent: the quiet workhorse that turns product data into living infrastructure instead of a perpetual clean-up project.

Then Agentic Commerce: the shift to a world where software agents search, compare and even buy on behalf of customers.

Most recently, a CRO & Experimentation Agent: the operating system that keeps your stack learning in the background instead of running a handful of A/B tests per quarter.

Each of these can live as a standalone initiative. Many organisations are already doing some version of them: a catalog clean-up here, a GenAI content pilot there, a testing roadmap on the side.

The problem is that, for most teams, these are isolated projects. They do not yet form an operating model.

This article is about that operating model and why it matters specifically for founders, CEOs and partners who are making platform and investment decisions right now.

From Projects to an Operating Model

If you look at most "AI in commerce" roadmaps today, they read like a list of experiments:

Improve product descriptions with GenAI.

Try conversational search on a category or two.

Run more experiments on the checkout.

Maybe add a chatbot for support.

There is nothing wrong with any of these. The issue is that they do not compound.

When the team is busy or budgets tighten, experiments stop, catalog quality regresses to the mean, and AI becomes "that thing we tried last year."

The alternative is to treat AI not as a collection of use cases, but as an operating model made up of three always-on agents:

1. Foundation: What do we sell, in what structure?

2. Journeys: How do humans and AI buy it?

3. Learning: How does this system improve by itself?

Once you see the stack this way, the question stops being "Which AI experiment should we run?" and becomes "Which of these three layers is currently the bottleneck to growth?"

Layer 1: Catalog & Product – Your Critical Asset

In agentic commerce, your product catalogue is no longer just "content"; it is the primary interface that humans, marketplaces and software agents use to understand what you sell. A Catalog & Product Agent makes that interface trustworthy by:

Ingesting feeds from PIM/ERP/marketplaces continuously.

Enriching titles, descriptions, attributes, translations and SEO so every SKU meets a minimum standard.

Prioritising the products that matter most commercially instead of spreading effort evenly across 50,000 SKUs.

This is the most "unsexy" part of the story – and it is where your AI story lives or dies. Every agent you deploy, internal or external, is only as good as the data you feed it.

From a founder/CEO perspective, this is also the most defensible asset: a high-quality, structured catalog makes you easier to integrate, easier to acquire and easier to value.

Layer 2: Agentic Commerce – Designing for Software Buyers

The second layer asks a simple question: if a software agent tried to do business with you today, how far would it get?

In an agentic commerce model, software agents:

Interpret a goal expressed in natural language ("find me a compliant B2B supplier in this category, at this price point").

Traverse multiple storefronts and marketplaces through APIs.

Apply constraints (budget, delivery time, sustainability).

Either present options back to a human or complete transactions within defined rules.

Whether those agents are yours, your customers' or third-party intermediaries, they all depend on a few practical and critical things:

Clean, structured product data and search.

Consistent pricing, stock and fulfilment information.

Robust, well-documented APIs for search, pricing, baskets and checkout.

Clear authentication, consent and audit trails.

This is where your platform choice shows up on the P&L. The same decision that used to be "just" about re-platforming now determines how discoverable and "transact-able" you are in an agent-driven ecosystem.

Layer 3: CRO & Experimentation – The Learning Loop

The third layer is where most organisations are furthest behind.

CRO has traditionally meant a project plan of A/B tests, long analysis cycles and manual roll-out of winners. In a world of volatile traffic, multi-device journeys and AI agents negotiating on behalf of customers, that approach becomes technical debt.

A CRO & Experimentation Agent turns optimisation into an always-on system by:

Generating A/B/X variants for pages, offers and content.

Targeting meaningful audiences (high-intent return visitors, discount-sensitive first-timers, renewal-window customers).

Auto-pausing losing experiences and routing more traffic to winners as evidence accumulates.

Feeding all results into a living knowledge base so new ideas build on past learnings instead of repeating them.

Connected to the catalog and to agentic journeys, this becomes the adaptation layer of your stack. It is what allows you to say, credibly, "this commerce system becomes better the longer we run it."

Where to Start – A Practical Sequence

If you are already running on an enterprise or open-SaaS platform (Shopify Plus, VTEX, SFCC, Adobe, commercetools, BigCommerce, custom), with GA4 and some form of data warehouse (BigQuery or similar), you likely have 80% of the ingredients.

A practical sequence that works:

1. Make the catalog an explicit program.

2. Pick one agentic journey.

3. Wrap that journey in a CRO loop.

4. Codify the operating model, not just the tech.

From there, you are no longer "doing some AI in e-commerce." You are running an AI commerce operating model that can be scaled, measured and explained to boards, investors and partners.

Why This Matters Now

The underlying platforms, data tools and AI capabilities are here today. What is missing in most organisations is a clear, simple operating model that ties them together.

The three agents described here are one way to frame that model. If you would like to see what this could look like in your own stack – with your catalog, your journeys and your data – we are happy to exchange ideas and examples. Find more on Commerce Partners.

Originally published on LinkedIn – AI in e-Commerce newsletter, March 10, 2026.