Why AI needs good data first

There's an old rule in computing that has never been more relevant: garbage in, garbage out. AI doesn't think the way a person does. It finds patterns in the data it's given and makes predictions or decisions based on those patterns. If the underlying data is incomplete, out of date or inconsistent, the AI will confidently produce answers that are wrong — and it won't tell you they're wrong.

Imagine a Brisbane wholesaler asks an AI tool, "Which customers are at risk of leaving?" The AI scans the sales records. But half the customers were entered twice under slightly different names, recent orders sit in a separate spreadsheet the system never sees, and a big client's details were never updated after they switched contacts. The AI gives an answer, but it's built on a broken picture. The business acts on it, and the advice quietly leads them astray.

This is the part most people skip when they get excited about AI. The clever model is the easy bit now — there are dozens you can plug into. The hard, valuable work is making sure the data foundation underneath is accurate, complete and consistent. Good data is what separates an AI that earns its keep from one that creates expensive confusion.

Most Australian mid-market businesses don't have one tidy database. They have a CRM for sales, accounting software like Xero or MYOB for finances, a rostering tool for staff, a support inbox or helpdesk, maybe a point-of-sale system, and a handful of spreadsheets someone built years ago. Each tool holds a slice of the truth, and none of them talk to each other properly.

This fragmentation is the single biggest barrier to useful AI. A question as simple as "What's our most profitable type of customer?" requires joining sales data, cost data and service data — three systems that have never met. So the answer either takes a staff member two days of manual exporting and reconciling, or it never gets answered at all.

The technical term for the muddle is data silos: useful information trapped inside individual tools. When data is siloed, you can't see the full story, and neither can any AI you point at it. Worse, the same thing gets recorded differently in each system — "NSW" in one, "New South Wales" in another, "N.S.W." in a third — and those small inconsistencies pile up into unreliable results. Before AI can help, this data usually needs to be brought together and cleaned up.

Good data isn't about having more of it — it's about having data you can trust. In practice, that comes down to a few qualities worth knowing. Accuracy: the data reflects reality (the customer's email actually works). Completeness: the important fields aren't blank (you know which sales rep closed each deal). Consistency: the same thing is recorded the same way everywhere (one format for dates, one spelling for each state).

It also means data is connected. A customer record should link to their orders, their invoices and their support tickets, so you can see one whole picture rather than four partial ones. And it should be reasonably current — a forecast built on last quarter's numbers is less useful than one built on this week's.

You don't need perfection to start. A useful rule of thumb: data needs to be good enough for the decision you want the AI to support. Predicting next month's stock levels can tolerate a few gaps; automatically emailing customers cannot, because every error goes straight to a real person. Match the quality of your data to the stakes of the task.

Resist the urge to buy an AI tool first. The smarter sequence is: understand your data, connect it, clean it, then automate or apply AI on top. Skipping to the last step is how businesses end up with impressive-looking tools producing answers no one trusts.

A practical takeaway you can act on this week: pick one business question you wish you could answer instantly — say, "Which products give us the best margin?" or "Which customers haven't ordered in 90 days?" Then map out every system that holds a piece of that answer and how the data flows (or doesn't) between them. That single exercise will show you, in concrete terms, where your silos and quality gaps are.

For many mid-market businesses, the recurring obstacle is that the data never sits in one place. This is exactly why a unified-data foundation — bringing your CRM, accounting, rostering and support data together into one consistent database — tends to be the groundwork that makes AI and automation genuinely useful later. Get the foundation right, and everything you build on top of it gets easier, faster and far more trustworthy. In the next lesson, we'll look at what automation can realistically do once that foundation is in place.