AI in e-Commerce: CRO & Experimentation

Over the past weeks I wrote about two foundations of the next wave in digital commerce:

The Catalog & Product Agent – continuously curating, enriching and governing product data as living infrastructure.

Agentic Commerce – where autonomous agents execute what used to consume entire e-commerce teams, turning static journeys into adaptive, intent-driven flows.

The logical next layer is CRO & Experimentation.

Most organisations still run experimentation as a project: a few A/B tests per quarter, long analysis cycles, and manual roll-out of winners. Meanwhile, acquisition costs rise and AI agents raise the bar for relevance and performance at every step of the journey.

The CRO & Experimentation Agent

We are exploring a CRO & Experimentation Agent that embeds continuous optimisation into the fabric of the stack:

A/B/X variant generation

Audience targeting and segmentation

Auto-pausing losing experiences, amplifying winners

Journey analytics across GA4 and the data warehouse

From Projects to Operating System

CRO has always been about running experiments to improve conversion – but the environment around it has changed. Journeys span channels and devices, traffic is volatile, and AI agents are starting to search, negotiate and buy on behalf of customers. In that world, a quarterly roadmap of a few A/B tests and a dashboard review at the end of a campaign is no longer a strategy; it is technical debt.

What is needed is an experimentation operating system: something that runs continuously in the background, watches every journey and adapts experiences in near real time. A CRO & Experimentation Agent does exactly that. Instead of treating tests as one-off projects, it treats the entire funnel as a living system that is constantly generating hypotheses, running controlled changes and learning from the outcomes.

Data-Driven Hypothesis Generation

The starting point is data. The agent ingests behaviour from GA4, joins it with warehouse data (e.g. BigQuery) and reconstructs real user paths, not just page-level metrics. From there, it surfaces and ranks opportunities: where high-value users stall, which sequences drive repeat purchase, where shipping, pricing or UX issues create friction. Each opportunity becomes a hypothesis ready for testing.

It then turns those hypotheses into concrete experiments. The agent can propose A/B/X setups, generate copy and layout variants, and suggest alternative flows for different segments, within governance rules defined by your team. Targeting becomes smarter too: instead of "one test for everyone," it matches experiences to meaningful audiences such as high-intent return visitors, discount-sensitive first-timers or customers in renewal windows.

Continuous Execution and Learning

Execution is continuous. Rather than waiting for a fixed end date, the agent monitors performance, automatically pauses clearly losing variants and shifts more traffic to winners as evidence accumulates. This captures upside faster, reduces the cost of bad ideas and makes it safer to test bolder changes. Over time, all experiments feed a living knowledge base – what was tested, for whom, with what result – so new ideas build on history instead of repeating it.

The Adaptation Layer of Agentic Commerce

Connected to product and catalog agents that continuously enrich and govern product data, and to AI shopping agents acting on buyer intent, the CRO & Experimentation Agent becomes the adaptation layer of agentic commerce. For organisations already using GA4, BigQuery and modern commerce platforms, the opportunity is clear: let experimentation stop being an isolated project and turn it into the system that keeps the entire stack learning, optimising and aligned to outcomes by default.

In combination with the Product & Catalog Agent and an Agentic Commerce architecture, a CRO & Experimentation Agent turns optimisation into an always-on system: experiences are generated, targeted, evaluated and scaled continuously – for human visitors and AI shopping agents alike.

If you are working with GA4, BigQuery and an enterprise commerce platform and want to explore what this could look like in your landscape, we are happy to exchange ideas and examples. Find more on Commerce Partners. If you would like to start testing, do let us know – we have partner offerings available.

Originally published on LinkedIn – AI in e-Commerce newsletter, March 5, 2026.