Why AI Can't Tell You Much About Your Business (Yet)

You've probably tried it: you ask ChatGPT or Claude something about your own business, and the answer comes back vague, generic, or just wrong. The instinct is to blame the AI. Usually the AI isn't the problem. Your data is. AI can't give you a useful read on your business if the numbers it's working from are scattered, inconsistent, or out of date. Here's what that actually means, and what "AI-ready data" looks like for a normal small business.

Why AI makes things up about your business

A language model's job is to produce plausible text. When you give it clean, complete information, plausible and correct line up. When you give it gaps, it fills them with patterns and guesses, and you get a confident answer that happens to be wrong. People call these hallucinations and treat them as a flaw in the model. More often they're a symptom of the input. The model didn't know your real numbers, so it invented numbers that sounded right.

This is the old "garbage in, garbage out" rule, and AI doesn't repeal it. If anything it makes it worse, because the output is so fluent that a wrong answer reads as authoritative.

Most small business data is not ready for this

The reason isn't carelessness. It's how businesses actually grow. Your data ends up spread across QuickBooks, your point-of-sale system, a few spreadsheets, your email, and somebody's notebook. None of it talks to each other. Spreadsheets quietly drift, too: research on real-world spreadsheets has long found that the overwhelming majority contain errors, not just typos but formula mistakes that bend the numbers downstream (EuSpRIG).

Now point an AI tool at that. It has no single version of the truth to read, so it does its best with fragments. The output looks like insight and isn't.

This is the same reason most AI projects fail

It isn't only a small-business issue. When RAND studied AI projects, it found that more than 80% fail to reach meaningful deployment, roughly twice the failure rate of regular software projects, and data problems are near the top of the list of reasons (RAND, 2024). The companies that do get real value tend to do one boring thing first. McKinsey's 2025 research found that organizations seeing significant returns from AI were about twice as likely to have redesigned their data workflows before picking a model (McKinsey, 2025). First the plumbing, then the analytics.

It's worth knowing that even professional data scientists spend the bulk of their time, commonly estimated at around 80%, just finding and cleaning data before any analysis happens (Pragmatic Institute). If the pros spend that much time on prep, skipping prep with an off-the-shelf AI tool isn't a shortcut. It's a guarantee of bad answers.

What "AI-ready data" means for a small business

You don't need an enterprise data team. At a small-business scale, AI-ready mostly means a few practical things:

One source of truth per thing. One place that holds the real customer list, one for revenue, one for inventory. Not five versions that disagree.
Consistent formatting. Dates are dates, dollars are numbers, and a customer's name is spelled the same way everywhere.
No duplicate records quietly pulling your averages in two directions.
Current data, not a snapshot from six months ago.
Clear labels, so a machine (or a new hire) can tell what each column actually means.

That's most of the work. It isn't glamorous, and it's exactly the part the AI hype skips over.

Where to start

You don't have to fix everything at once. Pick the one question you most wish you could answer ("which products actually make money," "what's my real cash position"), and get the data behind that one question clean and in one place. A small, trustworthy dataset beats a giant messy one every time.

This is the part we help with. If you want AI to give you straight answers about your business, the first job is getting your data in order, and that's work we do before anyone touches a model. If that's where you're stuck, book a free call and we'll look at what it would take to make your data actually usable.