Why AI cannot answer questions about your business yet

Most business owners have tried it by now. Draft an email. Summarise a long document. Tidy up a job spec before it goes out. It works, it is quick, and it is often better than what you would have written yourself at half four on a Thursday.

Then you ask it something that actually matters. Which customers are costing us money to serve. Whether we can take this order without letting another one slip. Why margin dropped last month when nothing obvious changed. And it has nothing useful to say.

It is easy to read that as a limit of the technology, and to quietly file AI under things that are not really for companies like ours. The tool is not failing to think, though. It has never seen your business, and in most small and medium businesses there is no single place where your business can be seen. This is a fixable problem.

What AI for small business is actually doing when you ask it a question

An AI tool answers from two sources: what it learned while it was being built, and whatever you type into the box. The first is worth understanding, because it is rarely explained. These tools were trained by reading an enormous quantity of text, most of it from the open internet, and some will search the web live while answering you.

For general knowledge it works well enough. The trouble starts where the internet is out of date or simply wrong, because the tool has no dependable way of flagging which is which. It will answer a question about VAT treatment or a labelling rule in the same confident tone whether it is repeating something current or something that stopped being true three years ago. And when it does not know, it rarely says so. It fills the gap with an answer that reads correctly. Free tools are the most exposed to this.

That matters before you rely on one for anything serious. But it is still the smaller of the two problems, because even a tool with perfect general knowledge would not have your numbers.

This AI cannot see your bank feed, your stock counts, your open purchase orders or the customer who has been paying 20 days late for eight months. So when it gives you a general answer to a specific question, it is doing exactly what it was designed to do with the only information it has. We pulled this apart in more detail in the first episode of our podcast, if you want the longer version.

Why it cannot see your business

Ask most owners where their business information lives and you get a list rather than an answer. The accounts package holds the financials. Stock is on a spreadsheet one person maintains. Supplier prices are in an email thread. The till or the web shop holds sales. And a fair amount of it is not written down at all, it sits with the operations manager who has been there 11 years.

There is a common assumption that smaller businesses do not have enough data for AI to be useful. The opposite is usually true. Australia's national AI adoption programme puts it plainly: smaller businesses tend to have too much information, spread across spreadsheets, legacy systems, cloud platforms and paper records, and without context that volume is just noise.

None of this happened through carelessness. It accumulates. You bought a system, it half fitted, so somebody built a spreadsheet beside it to cover the gap. Then another one. As our founder and CEO Morgan Browne wrote recently in Business Age, nobody says the hand me down does not fit, you just cannot run in it.

The workarounds become the process, and after a few years nobody remembers they were workarounds. That used to cost you a slower month end and a few arguments about whose number was right. Now it costs you something else, because the thing everyone wants to use next can only work from a record that does not exist yet.

A connected system is what makes the question answerable

If the work happens in one place, the record builds itself. Nobody sits down to do data entry for the benefit of a future AI project. The sales order is raised, the stock moves, the purchase invoice is matched, the payment is allocated, and the byproduct of the ordinary work is a complete, current picture that nobody had to assemble.

That is really all a connected business system is for. Finance, stock, purchasing, sales and production stop being separate accounts of what happened and become one account. Ask what your true stock position is and there is one answer, not three that need reconciling.

Once that is true, AI has something to work with. Not before. A tool sitting beside your systems can only tell you what you have already told it.

Enterpryze is built that way on purpose. It is a cloud native ERP for SMEs that brings finance, inventory, sales, purchasing, service and production into one system, connected from the first day rather than assembled over three years.

That connection is what the AI sits on. Enterpryze Intelligence is built into the middle of the platform rather than bolted onto the side of it, and the first part of it is an assistant called Finn.

Finn is not a separate tool you have to keep in step with everything else. It reads the same live record your team is working in, which means when it tells you something about your stock or your margin, you can click through and see the transactions it used to get there. We are running sessions on how this works in practice, and you can see what is coming and sign up on our AI page.

If you are trying to work out whether your current setup could ever support this, that is a short conversation rather than a project. Talk to us and we will tell you straight.

Where to start

Pick one process where you already know the answer is unreliable and the unreliability costs you money. Stock accuracy is the usual candidate. Suppliers spend is another, because most businesses cannot say what they spent with each supplier last quarter without building a spreadsheet to find out. Get that process into a system properly, then do the next one.

Within a couple of months, the reporting gets easier on its own, because you are no longer assembling a picture from parts.

Do that and the AI question answers itself. Everyone arrives here eventually. The businesses that get value out of it first will mostly be the ones that sorted out the boring part early. If you would rather see it than read about it, there is a recording of a recent session on our sessions & events page that walks through what this looks like in a working system.

If you want to know what that would take in your business, have a conversation with us.

Frequently asked questions

Why does AI give me generic answers about my own business?

Because it has not seen your business. AI tools answer from general knowledge plus whatever you paste into them. They are not connected to your accounts, stock or orders, so anything specific to you has to be supplied manually, every time.

Do I need to clean up my data before I use AI?

You need it in one place more than you need it perfect. Data scattered across systems and spreadsheets cannot be read as a whole, no matter how tidy each part is. Connection comes before polish.

Is my business too small for this to matter?

No. Smaller businesses usually have plenty of information, it is just spread across more places than anyone realises. Size is not the barrier. Fragmentation is.

What is the difference between AI in an ERP and using a chat tool?

A chat tool answers from what you tell it in that conversation. AI built into an ERP reads the live record of your business, so it can answer questions about your actual stock, customers and margins, and you can trace how it got there.