We’ve heard this sentence from clients more times in the last year than in the previous decade combined: “We want AI to sort out our data.” We understand the hope. But AI doesn’t sort out data. AI consumes it. If your data is a mess, AI doesn’t fix the mess, it institutionalizes the mess, at machine speed.
The Three Checks We Run Before Any AI Project
Before we’ll scope an AI feature, we run three checks against the data it will touch. They’re simple, and they kill more projects than any model choice ever will.
1. Can you find it? Where does the data live? We’ve walked into companies where customer data sat in a CRM, three spreadsheets, a legacy database from 2009, and the inbox of one long-serving employee. If you can’t enumerate your data sources, you can’t feed an AI system.
2. Can you trust it? Is the data complete, current, and consistent? We ran a check for a client once and found their “customer” records were 23% duplicates with conflicting addresses. An AI trained or prompted on that produces confident nonsense. It can’t know what’s wrong, because nothing is labeled wrong.
3. Can you connect it? AI features don’t run on one table. They run on all of your systems at once: CRM plus ERP plus support tickets plus invoices. If those systems have never been integrated, the “AI project” you’re about to start is actually an integration project wearing an AI costume.
Why the Gap Keeps Growing
The data readiness gap is getting worse, not better, for one reason: the tools people use to capture data keep getting easier to adopt and harder to govern. Someone’s team sets up a new SaaS tool in an afternoon, and the data lands in a silo that nothing else can see. By the time AI enters the conversation, the sprawl is the project.
What Readiness Costs
The uncomfortable part is that for most businesses, the data work is 60-70% of the effort of an AI project. The model is the easy part. If a vendor quotes you an AI project with no data assessment attached, they’re either guessing or they’re planning to bill you for the discovery later, when it’s framed as a crisis instead of a plan.
The same work that makes you AI-ready makes you ready for everything else. Clean, connected, governed data pays for itself in reporting, automation, and integration, whether or not a single model ever ships.
The Honest First Step
You don’t need a generative AI strategy to start. You need a data inventory: a list of where your data lives, who owns it, and whether it can be trusted. If you’d like a hand running those three checks, get in touch. It’s a short engagement, and it tells you exactly what your AI ambitions are worth right now.