Why CRO stalled, and what comes after it

Picture a shop on a busy street. The door is propped open and a thousand people walk in every day. The shelves are full, the lighting is good, the signage makes sense, and the prices are competitive.

There is nobody working there.

People come in, pick things up, turn them over, look around for someone to ask, and put them down again. Roughly twenty of them work it out on their own and leave money on the counter. The other nine hundred and eighty walk out.

Now hire a consultant to fix it. They repaint the walls. They move the display closer to the door. They shorten the queue at the till. They test two versions of the window sign and keep the one that pulls three percent better. Every one of those changes is real work, done well, and not one of them touches the reason the shop is empty.

What CRO can and cannot do

Conversion rate optimization has had twenty years and an enormous amount of money. The tooling is mature, the practitioners are good, and the method is sound. Yet in Econsultancy's last survey on it, in 2017, only 28% of companies said they were satisfied with their conversion rates.

CRO works on the presentation of a store. Layout, imagery, copy, page speed, checkout flow. Presentation matters, and presentation is still not what converts a hesitant shopper standing in a physical store. A person does that.

A/B testing cannot ask a shopper what they are actually trying to solve. It cannot hear "this is more than I wanted to spend" and answer it. It cannot notice that someone has looked at the same two products four times and say here is the difference that matters for what you described. It cannot build trust either, because trust gets made in an exchange, and a webpage only ever makes one statement, over and over, to everybody.

So results stall, and they stall for reasons that have nothing to do with how well the tests were run.

The gap is architectural

E-commerce solved distribution. It solved marketing, more or less. It built payments, logistics, inventory, and attribution for ads. It never replicated the most effective part of physical retail, which is the person on the floor who sells.

By most industry estimates, physical stores convert somewhere between a fifth and two fifths of the people who walk in. The average online store converts two to three percent. You do not close a gap of seven to eighteen times by testing button colours. The gap is there because one of those two stores has salespeople and the other has none.

Nobody chose this. When stores moved online, the salesperson was the one part that could not come along, and the industry spent two decades building substitutes for them: product recommendations instead of a recommendation, reviews instead of reassurance, an FAQ instead of an answer, a retargeting email instead of a follow-up question. Each of those is a workaround for a missing person.

What we found when we started listening

At one bedding store running KameleonAI, fewer than two in every hundred shoppers messaged it between 4 July and 3 August 2026. Everyone else browsed and left without asking anything. Almost nobody asks.

The shoppers who did ask behaved nothing like the ones who didn't. Of the shoppers who messaged it, 22% added something to their cart, against 3% of those who didn't. They placed an order at 6.06%, against 0.93% for those who didn't, somewhere between five and six and a half times as often depending on how you count buyers we could not link to a visit. People who choose to talk are usually closer to buying anyway, so none of this says the conversation caused the difference.

The part you can act on is how few people ask. On almost every store on the internet, nothing is trying to earn the first question. No mechanism anywhere on the page has the job of turning a browsing shopper into a talking one. The entire apparatus of e-commerce optimization is pointed at making the room nicer.

What comes after

The layer that is missing engages a visitor who is clearly hesitating, asks what they are actually trying to do, answers the real question with a real recommendation rather than a grid of alternatives, handles the objection when it comes, and then takes the action: adds the right variant to the cart, applies the discount the merchant approved, walks them to checkout. That is not an improvement on CRO. It is the thing CRO has been standing in for since the beginning.

If you already run a CRO program, keep it. The room should look nice, the checkout should be short, the page should load fast. None of that stops being true. Just stop expecting the room to do the job of the person who is not in it.

The store with the best-tested landing page in your category still has nobody on the floor. So does yours. That has been sitting in plain sight for twenty years, because everyone agreed to call it a conversion rate instead of an empty shop.