Ask ten businesses why their AI project stalled before it started, and a good number will say some version of "our data wasn't ready yet." Sometimes that's true. Often, though, what actually happened is different: someone raised the idea of "getting the data ready" as a prerequisite, the scope of that cleanup quietly expanded to cover the entire organization, and the AI project got shelved indefinitely behind it.
This is one of the quieter ways good AI initiatives die. Not because the team ignored data quality, but because they treated it as one giant, undifferentiated problem to solve before doing anything else — instead of a specific question tied to a specific use case.
Readiness isn't a company-wide state
There's no single moment when "our data is ready for AI." Readiness is always relative to what you're trying to do with it. A support chatbot that answers questions from your help documentation needs those documents to be accurate and current — it doesn't need your entire CRM cleaned up first. An internal tool that summarizes sales calls needs decent call transcripts — it has nothing to do with your product catalog.
When data readiness gets treated as a single, company-wide checkbox, the natural response is to try to fix everything at once. That's a much bigger project than the AI initiative it was supposed to unblock, and it rarely has a clear finish line. Months later, the AI project still hasn't started, and the "data cleanup" has become its own open-ended commitment.
The narrower question that actually matters
The more useful question isn't "is our data ready?" It's "what does this specific use case actually touch, and is that particular slice good enough to start with?"
That reframing does a lot of work. It usually shrinks the scope from "audit everything" to "check the twenty documents this chatbot will actually draw from" or "confirm this one system has accurate records for the fields this workflow needs." That's a task a team can finish in days, not a mandate that never ends.
It also surfaces a more honest picture of risk. Some gaps matter enormously for a given use case — stale pricing information feeding a customer-facing assistant is a real problem. Other gaps are irrelevant to it — a messy internal wiki that the project never touches doesn't need to hold anything up. Scoping readiness to the use case is what lets you tell those two apart instead of treating every imperfection as equally disqualifying.
Good enough, with a plan to notice when it isn't
None of this means quality doesn't matter — it clearly does, and building on data that's actively wrong is worse than building on nothing. But "good enough for a bounded first version" is a very different bar than "perfect across the organization," and it's the bar that actually gets projects started.
The practical version of this looks like: pick a narrow, well-defined first use case. Check the specific data it depends on — is it accurate, is it current, does someone actually own keeping it that way. Fix what's broken in that slice. Then launch something small enough that you can watch its output closely and catch problems early, rather than trying to guarantee perfection before anything goes live.
That last part matters as much as the upfront check. A pilot scoped this way gives you a fast, low-stakes way to learn where your data actually has gaps you didn't anticipate — which is information no amount of pre-launch auditing reliably surfaces on its own. You learn more from watching a small, well-monitored system in real use for a few weeks than from another round of manual review.
The order that works
Don't try to make all your data pristine before deciding what to build. Decide what you're building first, however small. Then check the specific data that decision depends on. Fix that slice. Launch narrow. Expand once it's working.
That order gets real AI projects off the ground instead of leaving them permanently "almost ready." It also tends to produce better outcomes, because the data work stays tied to something concrete rather than becoming an abstract, unbounded cleanup effort.
If you're not sure how to scope that first slice — or want a second opinion on whether a specific use case's data is genuinely ready to build on — that's a quick conversation, not a big commitment. Book a free 30-minute call with Kaidon Labs, or reach out through our contact form.