AI systems inherit whatever data quality your business has. That’s the sentence I open more client engagements with than any other, because most SMBs I meet believe their data is in better shape than it is.
Here’s the diagnostic test I use. Pick any business-critical process — customer onboarding, order fulfillment, invoicing, whatever’s core to your business. Then ask, for that process: where does the source-of-truth data live, who owns it, and how do we know it’s current?
If you can answer that in an hour with confidence, your data is more ready than most. If you can’t — if the answer requires four meetings, ends with “well, we think it’s in the CRM, but Sales keeps a separate spreadsheet” — you have work to do before AI can help.
The problem isn’t abstract. AI systems that consume enterprise data will confidently produce answers based on whatever data they find. If your customer records are duplicated across three systems with different values in each, the AI won’t tell you it saw conflicts; it’ll pick one and answer as if it were true. That confidence is the danger. You won’t notice the AI is wrong until it matters — usually in front of a customer.
The good news: fixing this doesn’t require a data warehouse project. For most SMBs, the first-order fix is naming a data owner for the five or six most-referenced entities in your business — customer, product, order, employee, supplier. Just naming who owns each. Not fixing everything they own. Just having a person accountable for the quality of that data, empowered to make definitional decisions.
Once ownership exists, quality improves without anyone launching a formal initiative. The named owner starts noticing when their data goes wrong, because they’re now accountable for it. Six months of owned data beats two years of a data governance program that nobody enforces.
The order matters. Data ownership before AI. If your organization can’t name who owns “customer” — really name a person, not a department — you’re going to have a bad time trying to ground AI in your enterprise data.
Diagnostic question 4 tests this. The answer is often less encouraging than leaders expect, which is why it’s on the list.