The failure is older than the model
When an AI initiative disappoints, the review usually lands on the model, the vendor, or the team that built it. Look a little earlier and the cause is often none of those. It was there before anyone opened a notebook, in data that was never inventoried, never governed, and never really ready. An AI strategy resting on that is a plan to reach the wall faster. This is the unfashionable part of the argument: the order of operations is not a formality you can skip to save time. A data strategy is what decides whether your AI strategy has solid ground under it or is standing on a guess.
The questions a data strategy has to settle
Before picking AI use cases, you need clean answers to a handful of questions. What data do we hold, and where does it actually live. Who owns it, and who is allowed to use it. How good is it, and good enough for what. What are we permitted to do with it, legally and contractually. Most organisations cannot answer these without a lot of hedging, and that hedging is precisely why their pilots stall. Getting to real answers is the work: an inventory, clear ownership, a view of quality, and the rules of use. None of it makes a good slide. All of it carries the weight.
Sequence beats parallel
The tempting move is to run the AI strategy and the data strategy side by side and trust they meet in the middle. They usually do not. The AI ambitions sprint ahead while the data reality trails behind, and the gap between them is where the budget disappears. Letting the data picture arrive first changes the AI choices for the better. Once you know which data is trustworthy and usable, the list of viable use cases almost writes itself, and the ones leaning on data you do not have, or cannot use, get parked before they burn through a team. Sequencing is not the slow option. It is how you avoid paying twice for the same lesson.
Readiness is a governance question too
It is easy to shrink data readiness down to cleanliness and completeness. Those matter, but the harder half is governance. You cannot responsibly feed data into an AI system when you have not inventoried it, cannot name its owner, and do not know what you are allowed to do with it. Regulated organisations hit this first because the questions come from auditors, but the logic is the same everywhere. There is a practical way to honour this without freezing AI for a year: run a focused data discovery on the domains your most promising use cases would touch, get ownership, quality, and permitted use clear for those, and let that evidence decide what to green-light. The order of operations turns from a constraint into an edge.
Frequently asked questions
Why should a data strategy come before an AI strategy?
Because most enterprise AI failures start upstream in the data, not in the model. An AI strategy built on data that was never inventoried, governed, or made ready just reaches the wall faster. Getting the data picture clear first gives the AI strategy solid ground to stand on.
What does a data strategy need to answer first?
What data you hold and where it lives, who owns it and who may use it, how good it is and for what, and what you are legally and contractually permitted to do with it. Most organisations cannot answer these cleanly, which is exactly why their pilots stall.
How do I put data first without freezing AI work?
Run a focused data discovery only on the domains your most promising use cases would touch, get ownership, quality, and permitted use clear for those, and let that evidence decide what to green-light. You avoid boiling the ocean while still grounding AI in data you understand.