Here is a sentence I never expected to say, but it keeps proving itself true: way too many of the companies that submit data to us for analysis provide information with significant errors. Not rounding errors. Not minor inconsistencies. Errors that change the conclusion. We’ve analyzed thousands of data points for hundreds of companies.
That number should stop you cold, especially if you are about to hand data to AI. Beware of “garbage in, garbage out.”
AI’s greatest strength is also its greatest vulnerability: it processes whatever you give it, fast and without judgment. It cannot tell you that a sales metric was calculated in two different ways by two different departments. It cannot flag that the same customer appears in your database under three slightly different names. It cannot notice that last year’s definition of “on-time delivery” measured departure from your dock, while this year’s measures arrival at theirs. It will simply calculate, confidently, from whatever it was handed.
I call this feeding the landmine. The data looks fine on the surface. Nobody challenges it because it has always been there. And then you connect it to AI and amplify the problem across every channel, every decision, and every customer conversation you have.
Most companies don’t build bad data on purpose. It accumulates slowly and invisibly. Spreadsheets built by people who left years ago. Definitions that drifted when a new VP arrived and nobody updated the documentation. Employees in two departments measure the same thing two different ways and both of them are certain they are right.
Before you ask AI to generate anything – insights, recommendations, a competitive dashboard, a sales report – ask yourself five questions about the data you are about to hand it:
- Are our metrics clearly defined, in writing, so every department calculates them the same way?
- Is data being entered consistently, or does “close date” mean something different to every salesperson on the team?
- Have outdated and duplicate records been cleaned out, or is the same customer in the system three times?
- Do employees across the company interpret our key numbers the same way?
- And most importantly: are we measuring what our customers actually value, or just what is easy to put on a dashboard?
That last question is the one most companies skip. They clean up the data they have without ever asking whether it is the right data to begin with. Clean information about the wrong things is still the wrong information. AI will analyze it beautifully and lead you in the wrong direction, with complete confidence.
Companies that do this work before using an AI agent do not just get better outputs. They gain a competitive advantage that compounds over time, because they are monitoring the things that actually determine whether a customer stays or leaves.
AI does not create truth. It amplifies whatever is already hiding in your data. The last question worth asking before you flip the AI switch is simple: is the truth in your data something you want amplified?
August, 2026