At previous events, people from Zalando, HelloFresh, Delivery Hero, Vinted, Bolt, Wolt, Idealo, GetYourGuide and many others often describe almost the same challenges. Sometimes the blocker is technical. Sometimes it is strategic, methodological or simply organisational. The details vary, but the patterns repeat. When practitioners talk openly, the path forward becomes much easier to see.

This is why I am starting this newsletter. The meetup showed me how important a community can be. Sharing what works, comparing approaches, asking honest questions. My goal is to take this feeling and make it available beyond Berlin, so more people can benefit from the insights and the spark that comes from these conversations.

What you can expect in 2026:

Thank you again for joining early. I hope this becomes a useful part of your week and a small boost for your experimentation practice.

For digital experimentation platforms, 2025 was unusually eventful. Datadog announced to acquire Eppo in May. That same month, Statsig raised $100 million at a $1.1 billion valuation, before being being acquired by OpenAI in September. Combined with earlier consolidation (for example Harness acquiring Split.io in 2024, and Everstone acquiring the company behind VWO), this is a clear signal that the market is entering a new phase.

In 2026 and beyond, these moves will likely reshape buyer expectations, pricing, and product roadmaps. The acquired platforms will keep serving customers and continue to innovate, but consolidation also creates whitespace. Integration work takes time, and that “integration tax” often creates opportunities for others to move faster. It will be interesting to watch whether established players like GrowthBook (open source), ABsmartly, or ABlyft gain share, and whether new contenders emerge.

Let’s zoom in on Eppo and Statsig. Both reached an exit within a few years of building their platforms. That pace is a strong indicator that experimentation is no longer a niche capability. It is becoming a standard layer in the product development stack.

A few things stand out.

First, experimentation is being pulled into broader platforms. Datadog is embedding experimentation into observability and analytics, treating testing and feature delivery as first-class citizens in the engineering stack. OpenAI is betting on strong experimentation discipline to scale AI products safely and effectively.

Second, we are clearly seeing consolidation of existing solutions. The market is shifting away from standalone experimentation vendors toward platforms integrated into broader toolchains. In that world, shipping fast is not enough. Shipping fast with control, safety, and measurable impact becomes the real competitive advantage.

Third, experimentation will become more intertwined with AI-supported software development, whatever label we use today—vibe coding, AI-assisted development, or agentic coding. The common thread is the same: faster iteration cycles increase the need for strong guardrails and trustworthy measurement.

The practical takeaway for leaders remains unchanged: if your organisation cannot test, measure, evaluate, and stop initiatives quickly, you are operating with unnecessary blind spots. The next question is how these acquisitions get integrated, and whether this is the beginning of a larger consolidation wave.

For many teams, this is a good moment to revisit their experimentation platform strategy. And as experimentation becomes part of larger ecosystems, one risk increases: vendor lock-in.

In this environment, vendor lock-in and switching costs become even more important. When experimentation vendors get acquired by companies anchored in a specific vertical—observability, CI/CD, or a broader AI organisation—you have to ask a simple question: will their future roadmap still align with your needs, or will your requirements become secondary?

Historically, “warehouse-native” integration has often been used as an argument to reduce lock-in risk. But after some recent hands-on work with data modelling—and a few conversations with teams in e-commerce—I’ve come to see this argument as more nuanced in practice.

Warehouse-native integration can be a powerful model, especially when your data foundation is mature.

Benefits you often get:

Downsides that are easy to underestimate:

One approach that has proven valuable in this setup is to externalise metric definitions into a central, self-owned place, for example a Git repository. Done well, this helps with vendor portability, but the bigger win is organisational: metric certification, governance (who can edit what), reviewability, and reusability. The same experiment metrics can then power A/B testing, product analytics, and BI use cases, while reducing lock-in and future switching costs.

A pattern you see in many companies for similar reasons is to combine a warehouse-native experiment analysis layer with a separate (or in-house built) feature flagging and assignment component. The upside is control over engineering quality and operational excellence in a critical part of the stack. The cost is predictable too: less “fully integrated” convenience, potential feature gaps, and the ongoing burden of maintaining core infrastructure.

As with most things in experimentation (and in life), the real answer is trade-offs.

2025 was a very successful year for the Berlin Experimentation Meetup, which I co-organise with Patrick Gunia. Starting with the February event, we saw a noticeable increase in interest that kept building throughout the year. Co-organising the meetup has become one of my favourite side quests. I genuinely enjoy it. Here is a short summary of 2025 for the BEM.

I opened with “Why In-House Experimentation Platform Teams Should Focus on Domain-Specific Business Needs.” The core argument: you often need both a strong generic platform capability and a “special ops” layer that focuses on what differentiates your business. That is where you can amplify an existing program or build targeted tooling that unlocks new value.
Ryan Lucht (Eppo) travelled from the US for the event and spoke about building self-serve experimentation by decreasing cost-per-experiment. That framing matters, especially at scale. Cost is not just compute. It is setup friction, organisational overhead, analysis time, and the operational cost of running experiments reliably. Lowering that cost is one of the few levers that truly unlocks high-volume experimentation.
Juan Orduz (PyMC Labs / Wolt) gave a highly technical talk on calibrating media mix models with experimental data. It was dense and full of formulas, and I was slightly worried about how it would land. But in a room of roughly 120 people at Zalando BHW, we still had around 20 people actively asking questions. That moment reminded me how strong and diverse this community is.

Meetup #8 was special because we brought the community back to where it all started: the HelloFresh office in Berlin. The theme spanned essential experiments, scaling strategies, and the emerging role of AI in optimisation.

Craig Sexauer (Statsig) delivered “10 Experiments to Run Before You Die.” The title is LinkedIn-friendly, but the content held up: a fun progression of increasingly hard experiments, with a few sharp insights many teams can apply. Write me a DM if you are interested in the full list, number 9 will shock you.
Aurora Moreno Herrera (HelloFresh) presented “The Metric-Driven Playbook for Scaling Experimentation.” Aurora is consistently excellent on stage, and the talk is one I referenced repeatedly afterwards. Mature organisations, in particular, benefit from structured approaches like an experimentation quality score.
Marcel Fritz (Levered) spoke about AI-powered growth experimentation and the journey from testing to continuous optimisation. This is where many teams are headed. It will not replace A/B testing everywhere in 2026, but we will see a meaningful shift toward optimisation-style use cases in specific domains.

At Bolt we covered cost reduction through experimentation, test quality via MDEs, and a healthy challenge to common misconceptions.

Christoph Rottler-Lavoie (Kameleoon) shared why experimentation lowers costs and saves resources, focusing on removing unnecessary work through evidence.
Garret O’Connell (Bolt) made the deliberately provocative case that many experiments are a waste of time, and used that to highlight why MDE discipline is central to test quality. The meme level was 12/10, in the best way.
Jonny Longden (Speero), visiting from the UK, gave “The Experimentation Delusion: Why We Are Focused on the Wrong Things.” The message landed: experimentation is broader than A/B testing, and teams should expand what they treat as testable hypotheses.

I still remember the rooftop setting: Berlin summer, Bolt’s office near Alexanderplatz and the Fernsehturm, and genuinely meaningful conversations.

This meetup was especially memorable for me. Anna and her colleagues put a huge amount of thought and care into the venue setup, and it showed. Everything came together: great preparation, great talks, and a genuinely warm community atmosphere. It sounds almost too good to be true, but I have it on video.
Olivier Jeunen (Aampe) challenged the status quo in marketing experimentation: are we optimising to estimate effects after the fact, or to continuously drive incrementality through 1:1 personalisation?
Anna Barnett (Taxfix) presented “Chaos to Structure: How We Grew Experimentation at Taxfix.” What I appreciated most were the practical tactics teams can directly copy into their own context.
Dr. Else van der Berg made a strong case for combining methods, showing how quantitative data can push teams into an “optimisation box,” while layered qualitative research can reveal latent user needs that unlock real innovation.

We closed the year with a special end-of-year edition at Delivery Hero. Matthias and the team hosted us in their beautiful office complex.

Matthias Vossberg (Delivery Hero) spoke about Delivery Hero’s experimentation transformation, sharing an inside view of how experimentation moved from an afterthought to a strategic capability. Over two years, the team made a step change in velocity and reliability, enabled by an SDK-driven approach and consistent leadership support. What stood out to me was the balance: strong technology and process, but also a clear focus on the people side, including the character traits needed to scale experimentation without relying on luck.

We wrapped up with a very improvised (a flight got cancelled) but, I hope, inspiring fireside session with friends from Iterable and fellow experimentation practitioners. We then ended the evening with an experimentation practitioner pub crawl—respectfully sized, with genuinely lovely conversations.

This newsletter is an additional side quest for me. My goal is to produce something genuinely useful for the experimentation community, and that includes you as a reader. Thank you for making it this far.

It is also a vehicle for structuring my own thinking. I recently read Shreyas Doshi’s piece “On the Confusion Between Writing and Thinking”—a hilarious Streitgespräch between Shreyas and his AI counterpart (Claude). It is well worth your time.

For me, the value goes beyond the question of whether thinking requires writing. Publishing a newsletter forces a full workflow: collecting ideas, selecting links and stories, connecting them to recent anecdotes and practitioner problems, and then turning all of that into something coherent. It also requires discipline: being structured, editing hard, and building a repeatable process that scales beyond motivation and into habit. I expect that to be a meaningful learning experience in itself.

And, candidly, I am looking forward to the accountability that comes with having you as my reader. The peer pressure is welcome. It raises the bar and keeps me honest about delivering something insightful on a regular cadence.

I am still setting up the workflow, and the early issues will inevitably include a bit more reflection than structure. My intention is to move quickly toward the format I described above: more case studies, more actionable tactics, and more practical takeaways—less rambling. Bear with me. We are in this together.

Thank you again for being one of the first readers. I wish you a Guten Rutsch and a happy New Year, 2026.