Before you add AI, check your data. A 10-point readiness checklist
Most stalled AI projects have a data problem, not a model problem. Ten questions to answer before you build, and what the EU AI Act timeline means now.
Every leadership team is being asked what its AI plan is. The pilots that impress in a demo and then stall usually fail for the same reason: the data underneath isn't ready. The model gives confident answers from stale, duplicated or wrong data, and people stop trusting it within a week.
Answer these ten questions before you build.
The checklist
- Which decision or task will this improve? Name it, and the number that should move: hours saved, response time, error rate. "Use AI" is not a goal.
- Where does the data live? List every source the AI will need. If it's spread across a dozen systems and spreadsheets, consolidate first.
- Do the numbers agree? If two reports give different revenue or stock figures today, an AI will happily give a third.
- Who owns each dataset? Every important source needs a named owner who fixes it when it's wrong.
- How fresh is it? An assistant answering from last month's export will be confidently out of date.
- Are the definitions written down? Customer, active, margin, churn. Shared definitions are what make answers consistent.
- Who is allowed to see what? The AI must respect the same access rules as the people asking, or it becomes a very efficient data leak.
- Is personal or confidential data involved? Know where it is, whether it can leave your environment, and under which agreements.
- How will you measure answer quality? Write down test questions with expected answers before you start, and run them after every change.
- Who reviews the output? Anything that changes data or reaches a customer should have a person approving it until it has earned trust.
If you can't answer most of these, that's not a reason to give up on AI. It tells you where to start.
What the EU AI Act means right now
If you serve customers in the EU, the timeline changed this year. The transparency obligations, such as telling people when they're talking to an AI system, have applied since August 2, 2026. The obligations for high-risk systems listed in Annex III, which include uses in employment, credit scoring and education, were moved by the Digital Omnibus to December 2, 2027. (overview of the changes)
The requirements themselves didn't get lighter, only later. Good data governance, documentation and human oversight are exactly what they ask for, so the checklist above is also your head start on compliance. This isn't legal advice; check how the Act applies to your specific use cases.
Where to start
Pick one use case with a clear payoff, fix the data it depends on, and ship it with guardrails and measurement. That's the approach behind our Data strategy & AI service, built on the foundations described in Data platforms & integration.
Dealing with this in your own system? We do this work for a living. The first call is free.
Book a call