AI for business
Everyone is telling you AI will transform your business. Nobody is telling you where to start.
I start with the admin that eats your team's week, not with the technology. Where automating it genuinely pays, I build it. Where it doesn't, I say so.
The wrong first question
"How do we use AI?" is the question that wastes the money.
It starts from the tool and goes looking for a problem, which is how businesses end up with a subscription nobody opens and a pilot that quietly stops being mentioned. The useful version of the question is duller and much more productive: what does this business spend hours on that nobody would miss doing?
That is usually admin. Moving information from one place to another. Reading something to work out what it is and where it should go. Retyping what a supplier already sent you. It is rarely the exciting part of the business, and it is almost always where the hours actually are.
A worked example
The invoice that files itself.
Every business receives supplier invoices and receipts by email. Somebody opens each one, works out what it is, and puts it into the accounts package. It takes a few minutes. It happens a few dozen times a week. Nobody enjoys it.
For one client, that now happens on its own. The automation watches the inbound mailbox and, for each message, works out whether it is an invoice or a receipt at all. If there is already a PDF attached, it takes it. If there isn't — because plenty of suppliers put the invoice in the body of the email — it converts the email itself into a PDF. Then it uploads the result straight into Xero.
What that buys is not just the minutes. It is that the filing happens the day the invoice arrives rather than the weekend before the VAT return, and that nothing sits unnoticed in an inbox because one person was on holiday.
This is what practical AI in a small business usually looks like. Not a strategy. One irritating job, done properly, that never needs doing again.
What I'll actually tell you
Plenty of it isn't worth doing.
A task is a good candidate when it is repetitive, follows a rule you could write down, happens often enough to be worth automating, and does no damage if it occasionally needs a human to check it. The invoice job qualifies on all four.
It is a bad candidate when it needs judgement you would not delegate to a new starter, when being wrong is expensive, or when it happens six times a year. Automating those costs more to build and babysit than the hours it saves, and I will tell you that rather than quote for it.
- Worth automating: sorting, filing, extracting, routing, chasing, formatting, and the retyping of things you were already sent
- Usually not: decisions with money or safety attached, anything a customer would be upset to learn was automated, and anything that happens rarely
- Needs care: anything touching personal data, which is most things once you look closely
The question almost nobody asks
Where does your data go when your team uses AI?
Staff are already pasting things into AI tools. Customer emails, quotes, contracts, spreadsheets with names in them. Mostly they are trying to do their jobs faster, and mostly nobody has told them what is and is not acceptable. That is not a disciplinary problem, it is a missing policy.
This is where AI stops being a productivity topic and becomes a data protection one, which is the part most people selling AI to small businesses do not raise. I also run the data protection and outsourced DPO side of this business, so the two conversations happen together rather than one of them happening two years late.
In practice that means a short, readable policy your staff will actually follow, a decision about which tools are approved, and knowing which of them train on what you put in.
How to start
An afternoon, a list, and no commitment.
The first step is a conversation with the people who do the admin, not with the management team. They know exactly which jobs are tedious, and they are usually right about which ones a computer should be doing.
What comes out of it is a short list of candidates with a rough sense of the hours each one costs you now. Some will be worth building. Some will turn out to be a setting in software you already pay for. A few will be neither, and knowing that is worth the afternoon on its own.
What does your team waste time on?
Start there rather than with the technology. No charge for the first conversation, and if the answer is that nothing here is worth automating yet, that is a perfectly good outcome.