AI for small business: what's actually useful right now (and what's just hype)

Where AI genuinely saves a small business time, where it causes problems, and a simple way to choose your first task to hand over.

There is an enormous amount of noise about AI at the moment, and very little of it is aimed at a business with fifteen staff and a busy phone. Here is our honest read on what is useful today, what is still hype, and how to pick a sensible first step.

The everyday wins

The real gains right now are not dramatic. They are the twenty minutes here and there that add up across a week, usually on writing and reading tasks that nobody enjoys.

• Drafting: a first version of a customer email, a quote covering letter, a job ad or a policy update. You still edit it, but starting from something beats starting from nothing.
• Summarising: turning a long supplier contract, a council document or a forty-page report into the five points that matter to you.
• Answering the same questions: opening hours, parking, what to bring to an appointment, how to lodge a warranty claim.
• Meeting notes: recording a conversation and producing a summary with the actions listed, so nobody has to write it up afterwards.
• Tidying data: reformatting a messy supplier list, or pulling details out of emails into a spreadsheet.

Notice what these have in common. They are tasks where a good draft is genuinely useful, where a human still checks the result, and where being wrong occasionally is annoying rather than serious.

Where AI goes wrong

The first problem is made-up facts. These tools produce confident, fluent text, and confidence is not the same as accuracy. They will invent a clause, a date, a product code or a piece of legislation and present it exactly like the true parts. Anything that goes to a customer, a regulator or a court needs a human to check it.

The second problem is privacy, and it is the one we get called about most. Pasting patient notes, client financials, staff records or your customer list into a free consumer AI tool means handing that information to another company, under terms most people have not read. For some industries that is a straightforward breach of your obligations.

The third is quiet dependence. If one staff member builds a clever process on a personal account and then leaves, the process leaves with them. Anything that matters should sit on a business account that your business controls.

Using a chat tool versus handing over a whole job

Most businesses start with a chat tool. Someone opens it, asks for help with an email, and copies the answer back into Outlook. That is useful, but it is still a person doing the work with a helper. The time saving is real but modest, and it only happens when someone remembers to use it.

The next step is different in kind. Instead of helping with a task, the AI handles a whole workflow start to finish: watching an inbox, reading enquiries, answering the routine ones in your tone, booking them in, and passing the unusual ones to a person with the background already summarised. It works overnight and on Saturdays, and it does not forget the process.

We work with XoomAI on exactly this. Their XoomAgent product is best described as a managed AI employee: it is set up around one of your workflows, given your rules and your tone, and it keeps running without anyone remembering to prompt it. We are not going to pretend it suits every business, and we would rather talk you out of it than sell you something you will switch off in three months.

How to pick your first AI task

Do not start with the most interesting idea. Start with the most boring one. The best first candidate has three qualities.

1. Repetitive: it happens many times a week, so a small saving multiplies.
2. Rules-based: you could write the steps on one page and a new staff member could follow them.
3. Low risk: if it gets something wrong, someone notices and fixes it before any harm is done.

Answering the same fifteen customer questions fits. Drafting standard replies fits. Approving payments does not. Giving clinical or legal advice does not. Anything where a confident mistake is expensive should stay with a person for now.

Measure before you start, even roughly. If nobody knows how long the job currently takes, nobody will be able to say whether the new way is better, and the conversation turns into opinions.

Ground rules for staff

You do not need a long policy. You need a page that a new casual can read in two minutes. Ours generally covers four points.

• Which tools are approved, and that they are used through business accounts rather than personal ones.
• What must never be pasted in: customer personal details, health information, financial records, passwords, anything commercially sensitive.
• That a person checks anything before it goes to a customer, and that the person who sends it owns it.
• Who to ask when someone is unsure, and that asking is always fine.

Write it, put it where people can find it, and revisit it in six months. The tools will have changed by then and the page should change with them.

The honest summary

AI is not going to run your business, and anyone telling you otherwise is selling something. What it will do, used carefully, is take a chunk of the repetitive reading and writing off your team so they can spend the time on customers.

Start small, keep a human in the loop, and be deliberate about what information you feed it. That is most of the good advice there is.

One last thing worth saying to owners who feel behind. You are not. Most of the businesses we talk to are at exactly the same point: somebody has had a play with a chat tool, nothing has been agreed as a business, and nobody is quite sure whether they should be doing more. Picking one small, repetitive job and doing it properly puts you ahead of almost everyone, and it costs you far less time than reading another twenty articles about it.

• ai
• productivity
• small business
• privacy