Data governance: the minimum viable setup to start AI

'Our data is not ready' is the most declared cause of AI project postponement — and the most misdiagnosed. The good news: the data governance needed to start AI is much lighter than theoretical frameworks suggest. This guide describes the minimum viable setup, by maturity stage.

The indispensable base holds in three elements: knowing what data exists and where (mapping), knowing who can access it (rights), and knowing its reliability (quality measured on critical fields). For a first AI use case, this base is built on the use case's scope only — not on the company's entire estate.

The classic mistake: launching an eighteen-month 'data programme' before the first AI project. Successful organisations invert the logic: they pick a high-ROI use case, bring data to quality on that restricted scope, deliver results, then capitalise. Each AI project improves the data estate — governance is built while walking.

On the regulatory front, GDPR and the AI Act converge on the same requirements: knowing which data feeds which processing, being able to document it, and controlling access. A well-kept processing register is already 60% of AI compliance work — an often under-exploited asset.

The indicator that sums it all up: the time your teams need to obtain a reliable dataset on a given scope. If it counts in days, you are ready for AI. If it counts in months, that is THE bottleneck to fix — and it is a weeks-long project, not years, when properly targeted.