Ex-Booking.com Head of SEO: 1,000 experiments, 3 rules | Julian Redlich, VP @ Permira

Julian Redlich spent around six years at Booking.com, ending as Head of SEO, while the discipline there grew to roughly forty people across development, design, copy, analytics and SEO. He now sits on the other side of the table as an investor at Permira, looking at organic growth across a portfolio that includes mobile.de, Kleinanzeigen.de, Squarespace, Zendesk and Flixbus.

That combination is rare. Dozens of companies cross his desk every year, which means he sees patterns across business models most operators only ever see one of.

My 8 biggest takeaways from our conversation:

1. The number one thing that worked at Booking.com was being able to change things on the website.

Not analysis. Most teams he sees spend the overwhelming majority of their time analysing, building internal decks and fighting internal battles, and only a few percent actually making a change. Adding content, creating pages, fixing the technical foundation: if those three are unblocked, you grow. Watching Search Console go up and reporting that it went up is not the work.

2. Three conditions decide whether SEO works at all: management support, peer trust, and access to the site.

If you lack top-management backing you spend your life re-proving the channel. If your development or data peers do not trust you, you stay in defensive mode. And if you cannot change the site, strategy is the only thing left to produce. When SEO stalls it is almost always one of these three, not a missing tactic.

3. Scaled landing pages only work when there is enrichment data behind them.

The Booking template was a database crossed with a pattern: hotels in {city} across roughly 230,000 cities, then the same again for around 600 regions. But a page for a village with no available hotels has no right to exist. That constraint is the whole difference between a large site and a thin one, and it is why "we need one perfect article" was never the strategy. A hundred thousand good pages bringing one click each beats one article.

4. Test on as much traffic as you can, not as little.

The instinct is to isolate an experiment in one small market so nothing breaks. Booking did the opposite: expose a big hypothesis to large traffic, get the signal back fast, keep it if it turns positive and kill it if it does not. A single blog article is usually not worth experimenting on at all. Experimentation earns its keep when the website is the product.

5. Failed experiments are the point, not an embarrassment.

Across roughly a thousand experiments plenty did not work, including a widely believed pagination theory the team tested at scale, which went nowhere because Google had quietly stopped using the signal at all. Every experiment that fails removes something from the list, so the list gets shorter and the remaining bets get better. Every experiment at Booking was peer reviewed, and product managers had to review other people's experiments as part of their own review cycle.

6. Delete the laundry list. Ship the one thing.

The complaint he hears at every company and every conference is that IT will not implement the backlog. His answer is blunt: delete the list, work out the single change most likely to move the site, and get that one shipped. Half the problem is self-inflicted. Stories are over- or under-documented, and SEOs routinely ask engineering for the wrong fix because they have already assumed the right one is impossible. Give the development team the full context of the real problem and they will usually solve it.

7. He hires on business acumen, communication and craft, and looks for people who build rather than opine.

Business acumen means understanding how the site makes money and how much your channel actually moves. Communication means being able to explain your three priorities to a C-level and to a junior. Craft is tested against the specific site type, because marketplace SEO and content-site SEO are different jobs. The disqualifier: arriving with a Lighthouse score to raise from 55 to 60. He wants to know whether you understand the demand in the market and can show relative share against competitors over time. He also still rates candidates with their own side project, because breaking your own site teaches you what does not move the needle.

8. The generative AI risk is ultimately a Google risk.

An opinion he formed back in 2022 and still holds. Google has the data, the decade of AI investment, and distribution across a stack of apps with billions of users: everything needed to do what ChatGPT, Claude and Perplexity are doing, better. On the super-app question his read is that customer behaviour changes slowly. The same fear was attached to Facebook becoming the place people shop and book, and to mobile-first destroying conversion rates. If it happens it happens gradually, with leading indicators in faster-moving markets, and that is when you react.

Also in this episode: how to size whether SEO deserves a place in your mix at all (if paid search works for you, organic probably matters too), and the keyword-universe method he uses to judge a market. Pull your competitors' keywords, pull their competitors' keywords, bucket the whole universe with an LLM, and compare share by commercial-intent bucket. Under a couple of hundred dollars in credits.

Let's connect

Niklas on LinkedIn: https://www.linkedin.com/in/niklas-buschner/
Radyant on LinkedIn: https://www.linkedin.com/company/radyant/
Julian on LinkedIn: https://www.linkedin.com/in/julianredlich/
Check out Permira: https://www.permira.com/