Data governance: the checklist that avoids the endless project

Data platforms rarely fail for technical reasons. They fail when, a year in, nobody knows who answers for a given field, when a producer changes a layout without telling anyone, and when data quality is an opinion rather than a measurement.

The first practice is explicit ownership: every domain has an owner with a name, who decides what its data means and how long it lives. It is not a formal role — it is the person you write to when a number does not add up.

The second is a data contract between producer and consumer: schema, frequency, validity rules, what happens when something is late. Written down, versioned and checked by the pipelines, so a break shows up when it happens and not three reports later.

The third is quality measured on a few stable metrics — completeness, freshness, uniqueness, referential consistency — published alongside the data itself. When the service level is visible, arguments about the data take minutes instead of meetings.