From the Lab: How We Are Using AI to Rethink CRO

Not in a flashy, “AI-first” way. Not through off-the-shelf automation. But by embedding AI into the way we approach optimisation: how we analyse behaviour, generate ideas, test hypotheses, and move faster without sacrificing intent.

If you’ve worked in digital optimisation, you’ll know that most conversion work still starts with a combination of best practices, gut instinct, and slow experimentation. And while that approach isn’t broken, it’s rarely fast. It’s rarely scalable. And it often misses patterns that are right in front of us, just too complex or nuanced to spot without help.

This is where AI gets interesting. Not as a replacement for CRO, but as a quiet layer behind it, helping us spot what matters, design better hypotheses, and act on insight before it expires.

Take user behaviour analysis. Traditionally, understanding what’s causing drop-off means heatmaps, session recordings, and days spent scanning reports. Recently, we’ve been working with platforms that use machine learning to surface high-friction patterns across thousands of sessions, things like cursor hesitation, repeated back-and-forth behaviour, or micro-interactions that correlate with abandonment. Instead of starting with a vague idea like “people might be missing the button,” we’re now working with precise behavioural signals that point to where attention is breaking down, and why.

Where things have really opened up is in experimentation. We’ve moved beyond traditional A/B testing toward dynamic, AI-driven variant selection. Instead of running tests in sequence, we’re working with platforms that allow dozens of versions to be tested simultaneously across copy, layout and tone, all dynamically weighted based on performance. It’s a different way of thinking. Less “winner vs. loser” and more real-time evolution. In some cases, the winning combination wasn’t something anyone in the room would have predicted, but the data made the case.

Alongside this, one of the most meaningful shifts in our process has come from incorporating AI into wireframe creation. Typically, once an insight is validated, there’s a lag translating it into layout, working up prototypes, and aligning internally. We’ve started using generative tools to go straight from behavioral insight to wireframe concepts. It’s not about handing design over to the machine and more about removing the blank page.

Say we identify that users are stalling on comparison points between two products. Instead of logging that insight and queuing it for design, we can prompt a model to generate wireframe options that emphasise comparison UX patterns. This gives our design team a head start, but more importantly, it anchors the creative process in live behavioural context. We can bring multiple variations into a workshop or test and move from idea to validation in days, not weeks.

The same logic applies to predictive UX. Instead of showing urgency messages or social proof to everyone, we’re working with models that identify when a user might benefit from a nudge, hesitation, re-navigation, or inactivity. It’s not about manipulation. It’s about timing. We’ve learned that well-placed interventions, used sparingly and in context, can perform better than blanket tactics applied universally.

All of this points to a larger shift. CRO is evolving from a tactical layer to a more adaptive, intelligence-driven process and AI is becoming part of the operating system that powers it. This doesn’t mean giving up on rigour or strategy. It means removing friction from how we work, just as much as we aim to remove friction from how users interact.

We’ll be sharing more of this thinking over the coming months, including a deeper dive into how we’re using LLMs in wireframe and UX prototyping, and what happens when you let AI suggest page structure based on user behaviour and business goals. Spoiler: it’s not perfect. But it’s fast. And sometimes fast is the difference between capturing opportunity and missing it entirely.

We’re still early in this work. But what’s clear is that AI isn’t a bolt-on. It’s becoming part of the operating system for modern CRO. Not because it’s trendy. But because it works.

If you’re exploring similar workflows, we’d love to swap notes.

All the best,

LDN AI Lab