By: Jay Kt
The launch party is the most dangerous hour in a website’s life. Everyone in the room is proud. The homepage looks current. Someone grabs a screenshot for LinkedIn. And nobody is watching the number that actually matters, because that number won’t reach the dashboard for another three weeks.
When it does arrive, it’s often pointing the wrong way.
Marks & Spencer spent around £150 million rebuilding its site. The project ran two years and went live in February 2014. By the quarter ending late June, online sales were down 8.1% year over year, and the CEO described it at the annual meeting as a “settling in” period. My favorite detail from that whole mess isn’t the money, though. It’s this: the migration forced every customer to re-register. The retailer had roughly 6 million accounts on its old platform. After the relaunch, it had 3.2 million. Half the customer base evaporated at a login screen built by people thinking about brand refresh, not about the friction of recalling a password you set in 2011.
That’s a large company with a serious budget. The pattern scales down to twelve-page sites almost perfectly.
A Redesign Is Often the Answer to a Question Nobody Asked
Here’s a conversation I’ve had more times than I can count. Traffic is flat, or leads are down, or the sales team is grumbling. Someone says the site looks dated. Within a month there’s a redesign budget.
But nobody asked what was actually broken.
A website sits downstream of a lot of things. If the offer is confusing, a new grid system won’t fix it. If pricing is wrong for the market you’re chasing, better typography is an expensive way to avoid that conversation. If organic traffic is sliding because a competitor published forty comparison pages while you published two, no amount of scroll animation closes that gap.
Redesign is an appealing project precisely because it’s visible. You can show it to the board. You can’t show the board a quarter spent rewriting product pages that nobody will compliment.
So you treat the symptom, and the cause keeps running. Six months later, the site looks great, the numbers haven’t moved, and there’s no budget left to find out why.
One blunt question, before anyone opens Figma: if this redesign works perfectly, which specific number goes up, and by how much? If the room can’t answer that in a sentence, the project isn’t ready to start.
What Actually Breaks Under the Hood
A redesign gets discussed like a haircut when it’s much closer to transplant surgery. The visible part is the smallest part.
Search Engine Journal ran the numbers on 892 domain migrations, with data collected in October 2024. Average time for the new domain to match the old one’s organic traffic: 523 days. Call it seventeen months. The fastest recoveries in that set came in at 19, 22, and 23 days, so doing this well is clearly possible. But 17% of those sites never returned to their previous traffic level, even after a thousand days.
Rankings behave like a rent-controlled apartment. Cheap to keep. Nearly impossible to get back once you’ve moved out.
The failure modes are boring and specific.
URLs change, and nobody maps them. Every old address that returns a 404 instead of a 301 drops its rankings and orphans every link pointing at it. This is the largest avoidable loss in most migrations, and it happens because redirect mapping is thankless work that gets scheduled for the final week.
Content gets “cleaned up.” A hundred and forty blog posts become twelve, because the new design only has a template for twelve. Those posts were the pages that ranked. What follows looks like a migration problem. It’s really just deletion.
The staging block ships to production. Somebody blocks the staging site from crawlers, exactly as they should, and the noindex tag or the robots.txt rule rides along to the live server. Traffic doesn’t dip. It goes to zero.
Metadata gets overwritten by defaults. Titles, descriptions, alt text, structured data, years of accumulated work replaced by whatever the CMS generates on its own.
None of that is a design failure. All of it lands on the design project’s tab.
When the Prettier Site Performs Worse
This is the uncomfortable part.
Sonos shipped a rebuilt app in May 2024. Cleaner, more modern, built partly to support the company’s first headphones. It also shipped without sleep timers, alarms, and several accessibility features, while introducing connectivity problems that left some older speakers effectively unusable. Around 100 people were laid off that August. Revenue fell. The CEO was gone by January 13, 2025, eight months after launch.
Nobody at Sonos set out to make the product worse. That’s the whole point. A team staring at an interface all day sees clutter where a customer sees the button they press every single night. How many people have quietly built a routine around a feature your next release is about to remove?
Aesthetic judgment is real, and it’s valuable. It just answers a different question: “Does this still work for someone who’s used it for six years?” The failure mode isn’t ugly design. It’s design that was never asked to defend itself against usage data.
The teams that end up on shortlists of the best web design agencies tend to work in an order that feels backward to clients: audit first, then analytics, then a map of which pages carry traffic and links and revenue, and only after that anything resembling a layout. It’s a slower start. It’s also why their launches don’t produce the three-week surprise. Worth asking for that discovery phase in writing; if a proposal jumps from kickoff straight to wireframes, you’re buying decoration.
What a Redesign That Survives Contact With Reality Looks Like
Measure before you touch anything. Export your top pages by organic traffic, by conversions, by inbound links. That list is now a protected class. Nothing on it changes URL, gets merged, or loses content without a named person signing off.
Ship in pieces. The all-at-once relaunch is popular because it makes a good announcement, and it’s the riskiest available approach. Templates, sections, one funnel at a time.
Keep what works even when you hate it. Somewhere on your current site, an awkward element is driving more conversions than everything around it. You’ll want to delete it. Test first.
Test against the old version, not against opinion. A/B where traffic volume allows. Where it doesn’t, at least stage a genuine before-and-after window with identical tracking on both sides, and keep the redirect map live for six months, not six weeks.
Then watch Search Console daily for the first month. Not weekly. The window in which a stray noindex tag is cheap to fix is measured in days.
Most of the redesigns I’ve watched fail didn’t fail on execution. The redirects were fine. They failed because nobody in the room was willing to say out loud that the site wasn’t the problem.
That conversation costs nothing and takes an afternoon. Which is probably why it almost never happens.
