We launched a redesign that was genuinely better in most respects. It was faster, it looked considerably more modern, and user testing had been positive.
Organic traffic halved within six weeks. Recovery took about five months and was not complete.
Every cause was known, documented and preventable, which is the frustrating part.
The redirects
The largest single cause and the most common in redesigns generally.
URLs changed. A mapping was produced. It covered the main sections and missed several hundred older pages, which returned errors.
Some redirects were implemented as temporary rather than permanent, which signals something different to search engines and does not transfer authority the same way.
And a portion pointed to the homepage rather than to the equivalent page, which is treated as a soft error and is close to useless for the user.
The fix is unglamorous. A full crawl of the old site before launch, a complete mapping of every URL with traffic or external links, verification after launch, and monitoring of error reports daily for a month.
Content lost in the migration
The second cause and the one nobody noticed at the time.
The new templates were cleaner, which in practice meant shorter. Substantial explanatory content was cut in the interest of visual simplicity.
Several pages that had ranked for a wide range of queries lost the sections that addressed most of those queries.
Nobody made a decision to remove that content. It disappeared through a series of small design choices, each defensible, and the aggregate effect was not visible until traffic fell.
What we do now is a content inventory before and after, comparing word count and covered topics per URL, which would have caught it in an afternoon.
The rendering change
The third cause and the most technical.
The new site rendered much of its content in the browser rather than delivering it in the initial response.
Search engines handle this and they handle it less reliably and with delay, and some parts of our content were not being indexed at all initially.
This is a well-documented issue and it was not raised because the decision was made by a development team on grounds of architecture, and nobody in that conversation was responsible for organic traffic.
The organisational cause underneath
Which is the actual lesson.
The project was owned by design and development. The people responsible for organic traffic were consulted late, when the architecture was fixed and the templates were built.
By that point the objections were expensive to act on, so they were noted and deferred.
Nobody was obstructive. The sequence of the project simply put the relevant expertise after the decisions it should have informed.
What we do differently
Search and analytics involvement from the architecture stage, with a formal sign-off before build.
A pre-launch checklist that is genuinely blocking: full URL mapping verified, content parity checked, rendering tested as a crawler sees it, structured data preserved, analytics verified on a staging environment.
A phased launch where possible, moving a section at a time, so that problems affect a fraction of the site and are detectable before they are total.
And a rollback plan that somebody has actually tested, which we did not have and which would have limited the damage considerably.
What we should have measured beforehand
The baseline we lacked.
Traffic, rankings and conversions per URL for the top few hundred pages, captured immediately before launch.
Without that, diagnosing what had broken took weeks, because we were comparing against aggregate figures rather than being able to see exactly which pages had lost what.
Capturing it takes an hour. It is the single highest-value hour in any migration and it is routinely skipped because it produces nothing visible.
The uncomfortable summary
The redesign was a good redesign. The site is better now than what it replaced.
The five months of lost traffic and the permanent portion that never came back were entirely a project management failure, not a design failure, and they cost more than the redesign did.
The recovery
Since people ask what actually worked to get it back.
Fixing the redirects properly, which recovered the largest share and took about three weeks of unglamorous mapping work.
Restoring the removed content, which was contentious with the design team and which recovered most of the rest.
Server-rendering the content that had been client-rendered.
The portion that never came back was, as far as we could tell, positions lost to competitors during the months we were absent, which do not automatically return when you do.
The case for not redesigning
Worth stating since it is rarely the option on the table.
Most of the benefits attributed to redesigns — speed, mobile behaviour, clearer navigation, updated visual identity — can be delivered incrementally on the existing site, at lower risk, without a migration event.
Incremental change is less satisfying, produces no launch moment, and is considerably harder to get approved because it does not look like a project.
It also does not carry the risk of losing half your traffic in six weeks, which on reflection is a substantial argument.