- How Marks & Spencer's £150 million website redesign caused an 8.1% sales drop — and why nobody saw it coming
- Why the Dunning-Kruger effect means internal teams are the worst judges of their own websites
- What CRO actually fixes (it's not just the conversion rate)
In February 2014, Marks & Spencer launched a new website. The project had taken years and cost £150 million. They'd migrated away from Amazon's infrastructure, built their own platform, and redesigned the whole experience from the ground up.
Online sales dropped 8.1% in the first quarter after launch. Items vanished from shopping carts. Checkout failed for entire UK cities. Six million existing customers were forced to register again from scratch. The problems were everywhere.
The M&S team wasn't reckless. They had a clear vision, experienced leaders, and a serious budget. What they didn't have was a way to find out how real customers would respond to the changes before those changes were already live and irreversible.
That's the problem CRO was built to solve.
What CRO actually is
Conversion Rate Optimization is the practice of using data collected from real visitors to understand how website changes affect behaviour — and using that understanding to increase the percentage of visitors who do what you want them to do. If you want the full picture of how it works from the ground up, the CRO learning path walks through it step by step.
The "conversion" is whatever your website exists to drive: a product purchase, a subscription, a lead form, a booking. The "rate" is what share of your visitors complete it. And "optimization" is the ongoing, methodical process of raising that share — not through guessing, not through preference, but through observation and controlled testing.
This is a meaningful distinction. A hunch isn't a hypothesis. A confident redesign isn't an experiment. And an agency's track record somewhere else is not evidence that something will work for your specific visitors, on your specific site, in your specific context.
CRO replaces the confident guess with the disciplined question: what do our visitors actually respond to, and how do we know?
The problem lives between your ears
In 1999, psychologists David Dunning and Justin Kruger published a finding that's now one of the most cited in psychology: people who perform poorly at something dramatically overestimate how well they're doing. Participants in the bottom quartile of a test believed they'd scored around 68%. The actual figure was 23%. The gap wasn't dishonesty — it was a genuine inability to see the shortfall. They didn't have enough knowledge to know what they were getting wrong.
This is not abstract. It describes what happens in almost every website team that has never studied visitor behaviour.
The M&S team had deep retail expertise. What they didn't have was a trained eye for visitor friction — the moment a confused user abandons a checkout, the hesitation before a form that asks for too much, the instant a page fails to answer the question a visitor arrived with. They couldn't see those problems because they didn't know how to look for them. And the less equipped you are to spot a problem, the more certain you tend to feel that everything is fine.
Your brain is not a reliable instrument for evaluating your own website. Not because you're not smart, but because you've spent too long inside it. You know what every element means, what it was designed to do, why it looks the way it does. That familiarity makes it genuinely impossible to experience the site the way a stranger does — arriving cold, scanning quickly, deciding within seconds whether to stay or leave.
Two cognitive biases show up in almost every website team:
Confirmation bias. You notice the data that supports what you already believe and quietly discount the data that doesn't. When the team is excited about a redesign, every early positive signal feels like proof. Every warning sign feels like a statistical blip. This isn't dishonesty — it's just how brains filter information under conditions of enthusiasm.
Novelty bias. New and creative things feel like they'll work. A fresh design, an unexpected layout, a bold creative direction — the excitement they generate internally gets confused with evidence that visitors will respond well. Visitors weren't in the brainstorm. They don't share the excitement. They just see a page, and decide.
You are, in every A/B test sense, a sample size of one. Your opinion of your own website is one data point, from a person who is deeply unrepresentative of your actual visitor base.
What goes wrong without it
The $3 million loss is an extreme case, but the underlying pattern plays out quietly at every scale. Here's what accumulates when website decisions are made by conviction rather than evidence:
Wasted time. Months go into work that doesn't move any metric, because no one defined what improvement would look like before starting. Success and failure become impossible to distinguish.
Helicopter management. After a campaign underperforms, senior people become involved who weren't before. They bring their opinions — which are also a sample size of one — and now there are more guesses in the room, not fewer. The team that could have shown data has nothing to show.
Borrowed bad ideas. Without a basis for evaluating your own changes, the next move is often to copy what competitors appear to be doing. But you can't see their data either. You're copying the visible output of their decisions, with no idea whether those decisions are working for them.
Playing it safe. Eventually, teams that have been burned enough times stop taking creative risks. The writing becomes corporate. The images become stock photos. The layouts become forgettable. Playing it safe is the quiet end state of a team that never built a way to validate whether risks are worth taking.
The compounding effect is the dangerous part. Each missed quarter makes the next decision harder — there's more pressure, more opinions, less appetite for testing, and less patience for the time it takes to learn something real.
What CRO gives you instead
A structured CRO practice doesn't just improve a number on a dashboard. It changes the dynamics of how decisions get made.
| Without CRO | With CRO |
|---|---|
| "I think we should change the layout" | "Here's what the data says — here's my hypothesis" |
| Large, irreversible redesigns | Small, reversible experiments with clear success criteria |
| Debate settled by seniority | Debate settled by the next test |
| Copying what looks successful elsewhere | Testing what actually works for your specific visitors |
| Launches that might work | Launches you understand — win or lose |
The shift isn't just strategic. It's cultural. When a team has a shared, evidence-based way of evaluating ideas, the arguments get shorter and the learning compounds. A 5% lift this quarter, another 4% next quarter — that's not linear, it's compounding. Teams that never test never compound.
Who can do this
CRO is often associated with data scientists, specialized analysts, expensive tooling. In practice, the skills that actually matter are learnable by any marketer or product manager willing to approach their own ideas with a degree of healthy scepticism:
- Framing a question as a testable hypothesis before acting on it
- Reading analytics well enough to know where visitors are dropping off
- Prioritizing which problems are worth solving first
- Running a simple experiment and interpreting the result honestly — even when it doesn't confirm what you hoped
The barrier to CRO isn't technical. It's the willingness to replace we think with let's find out — and to genuinely accept either answer. If your site has real traffic and you want this done properly, the CRO audit is where to start.
You don't need a sophisticated setup to start. You need one question worth asking, one change worth testing, and one metric that will tell you which version won.