Is your data ready for AI? Five questions to answer before choosing a model

We open every AI engagement with a conversation about data, and it tends to end the same way: the ambition is real, the data is not yet ready, and the fastest route to the ambition runs through the data regardless. The five questions below are the ones we ask. The list is short because the answers are usually no.

The five questions

• Can you name the owner of each dataset the system will rely on? Not the team—the individual who is accountable when the data is wrong.
• Do the numbers agree across systems? Inventory in the ERP and inventory on the shelf; customers in the CRM and customers who paid. Where they disagree, the model will learn the disagreement.
• Is there a tested pipeline, or a person? A dataset that exists because someone runs a script on Mondays ceases to exist when that person is on leave.
• Do you know what you are permitted to do with it? Consent, contracts, and residency rules determine which fields a model may see. Discovering this after the build is the expensive way to learn it.
• Can you reproduce last month? If the same query returns a different answer next week and no one can explain why, no evaluation of a model built on top of it will be meaningful.

What to do with the answers

None of these is a reason to stop; they are the work. A governed platform with tested pipelines, lineage, and clear ownership is a smaller undertaking than most organizations expect, and it pays back on the first analytics question, long before a model is involved. Build it in the order the AI use case requires rather than all at once.