There is a question being asked in every company right now, usually with a lot of excitement: “where can we add AI?” It is a fair question. But there is a less glamorous one that should come first, and almost nobody asks it: “is our data actually ready?”
Because here is something the demos never mention. Most AI projects do not stumble because the AI is weak. They stumble because the data feeding it is a mess. The cleverest model in the world is still only as good as what you put into it.
What “good enough” data looks like
Nobody has perfect data, and you do not need it. What you need is data you can trust. In practice that comes down to four plain qualities. It should be accurate, so the email actually works and the order total is really correct. It should be complete, with the important fields filled in rather than left blank. It should be consistent, so the same thing is written the same way everywhere, not “UK” in one place, “United Kingdom” in another, and “U.K.” in a third. And it should be fresh, reflecting today rather than a snapshot from two years ago.
Why AI raises the stakes
Messy data has always caused quiet problems. AI makes those problems louder and faster. A person glancing at an odd record might pause and think, “that does not look right.” AI rarely does. It takes the bad data at face value, acts on it with total confidence, and then repeats the same mistake at a scale and speed no human could match. Poor data used to just sit there. With AI on top, it turns into poor decisions, made quickly, again and again.
A check you can run this week
You do not need a six-month audit to begin. Pick one set of data that really matters, and your customer list is a good place to start. Then ask a few honest questions. How many records are missing an email or another key field? How many look like duplicates of the same person wearing different hats? How old is the data, on average? And do different teams record the same thing in different ways? The answers tend to be uncomfortable, and that is exactly the point. They show you the two or three fixes that will matter most.
The unglamorous secret
AI is genuinely exciting, and it is worth doing. But the smart order is data first, AI second. Clean, complete, consistent, fresh data is not the boring chore you do before the real project. It is the thing that quietly decides whether the real project works at all. Fix the foundation, and the clever tools finally have something solid to stand on — and you finally see the results everyone was promised.