Why CRO stalled, and what comes after it
Twenty years of testing headlines and moving buttons hit a ceiling. Not because the execution was bad, but because it was optimizing the wrong layer.
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. Satisfaction with conversion rates across the industry still sits stubbornly low, and most programs describe the same arc: strong early wins, then diminishing returns, then a plateau nobody can explain.
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 the curve flattens, and it flattens 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.
Physical stores convert somewhere between a quarter and half of the people who walk in. The average online store converts under two percent. You do not close a gap of eight 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
Across the stores we measure, only about one and a half to two percent of visitors ever send a single message. Everyone else arrives with a question and leaves with it. That number holds steady across very different catalogs, and on its own it is a fairly bleak statistic. Almost nobody asks.
The shoppers who do ask behave nothing like the ones who don't. They add to cart at many times the rate, and the ones who talked to it converted at multiples of the rate of the ones who didn't. The effect scales with the conversation: someone who types one to three messages converts better than someone who only taps a suggested reply, and someone who types four or more converts far better than either.
[SLOT: replace the three sentences above with the measured figures from the current window once GO.sh has run: chat vs non-chat add-to-cart rate, the conversion-by-depth curve, and the lift range with its floor and dates. Attach the self-selection caveat in the same paragraph.]
Read that curve carefully, because it says something more useful than "chat is good." Depth is what moves the number. A shopper who asks four questions is a different shopper from one who asks none, more engaged, closer to deciding, more likely to buy whether or not anyone helps them. That self-selection is real and we are not going to pretend otherwise. Talking to it did not cause the difference on its own.
It also points at something you can act on. On almost every store on the internet, nothing is trying to earn the fourth 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.