Using AI in a Small Business: What Actually Works

We wrote the first version of this post in 2023. It was about chatbots answering customer questions, because that is what everybody thought AI was for at the time.

That turned out to be one of the less useful applications. Here is what has actually held up.

Where it earns its keep

First drafts

AI is good at getting you from a blank page to a rough draft. Proposals, policies, position descriptions, the awkward email you have been putting off since Tuesday.

The draft will not be good enough to send, and it does not need to be. Editing something mediocre is much faster than starting from nothing, and that is where the time saving actually comes from.

Summarising

Hand it a long document, a meeting transcript or a thread of 40 emails and ask what it says. This works well, and the risk is low because you can check the answer against the source sitting right in front of you.

Meeting notes are the standout. Teams and most conferencing tools will transcribe a meeting and pull out the actions. For a business that never wrote minutes, that is a genuine change rather than a marginal one.

Starting technical work

Scripts, spreadsheet formulas, database queries, configuration files. AI will get you most of the way and you fix the rest.

The catch is that you have to know enough to spot what is wrong. It will write a confident PowerShell script with a subtle bug in it and never mention the bug. In the hands of someone who can read the output it saves hours, and in the hands of someone who cannot it quietly creates a problem for later.

Searching your own information

This is the area that has improved most since 2023. Tools that index your own documents and answer questions from them, such as Microsoft 365 Copilot, are genuinely useful in a business with years of files nobody can find.

Know what it exposes, though. If your permissions are untidy, a search tool will happily surface the salary spreadsheet somebody saved in the wrong folder. Tidy the permissions before you switch it on rather than after.

Where it still disappoints

Customer service chatbots are the obvious one. They handle narrow, well-defined questions and they frustrate people with everything else. Most small businesses never had the volume to justify building one.

Then there is any job where being wrong matters and nobody checks. AI is confidently wrong in a way that reads exactly like being confidently right, and it has no reliable way to tell you it is unsure.

Anything depending on current facts needs verifying too. Prices, dates, legislation and the specifics of your own industry are all worth checking yourself.

The data question

Be clear about where your information ends up.

A free consumer tool may use what you type to improve the product. Business tiers usually commit not to. Read the terms for the tier you are actually on, because they differ more than you would expect.

Then set one rule your staff can remember. Ours is that client data, personal information and anything under an NDA does not go into a tool we have not approved. A single sentence gets followed. A four-page policy does not.

If you already run Microsoft 365, Copilot keeps data inside your tenant boundary, which is usually why it clears approval faster with regulated clients.

What it costs

Per-seat AI licensing is not trivial. Across a team it becomes a real line item, and it renews every year whether anyone is using it or not.

So do not licence everybody on day one. Give it to five people who want it, wait a month, then ask what they actually use it for. Some will have built it into their working day. Others will have opened it twice.

Expand from there, based on what you learned rather than what the vendor’s slide deck promised.

How to start

Pick two tasks that are repetitive, text heavy and low risk. Meeting notes and first-draft documents are the usual pair.

Do those properly for a month and get a rough sense of how much time came back. Then decide whether to go further.

That is a dull recommendation and it is the one that works. Most failed AI projects we have seen failed because somebody bought the platform before picking the problem.

Element Digital offers IT consulting services in Hobart. If you want a hand working out where AI fits, and where it does not, get in touch.