When Data Mesh become as mess

Last week, I spoke at a trivago meetup about SumUp's Data Mesh journey from the perspective of our Data Platform team.

I highlighted a few non-technical changes we made along the way:
✅ Hiring and upskilling towards software and backend engineering, not just data engineering
✅ Doubling down on developer experience: building UIs, killing off unnecessary JIRA tickets for each request (as much as possible), fixing bugs before a new feature
✅ Borrowing the same incident management practices from our backend + SRE teams

But what surprised me most wasn’t in the slides—it was in the Q&A.Since I was the last speaker, I had more time for questions. And while I expected a flood of “which tool do you use?” or “how much does that cost?” … nearly half of the questions I got was towards data governance.
- How do you handle data governance and data contracts?
- How do you align on metrics definitions (like revenue)?
- Who defines the source of truth?

That moment gave me confirmation of something I’ve long believed: the hardest part of the four pillars of Data Mesh isn’t tech—it’s governance.
Buying a new modern infra or platform can't solve this.

So maybe the lesson is this: if you’re thinking about Data Mesh, don’t just obsess over Snowflake or Databricks, Iceberg or Delta, or finding the best data ingestion tooling. Spend equal (if not more) time on governance—on how teams align, define, and trust their data.

Otherwise, you end up with a mess not mesh.