What AI can — and cannot — automate in a small business
Published on
There is a lot of noise around AI for business, and it makes decisions hard. This is our view from what we actually implement: what works today in a small company, what does not, and how to choose where to start.
Where AI clearly adds value today
Answering common questions using your business information. An assistant that knows your hours, services, prices and cancellation policy resolves most first-line enquiries without anyone stepping in, at any hour.
Classifying and prioritising. Reading an incoming enquiry and deciding whether it is urgent, which service it concerns and who should handle it. Repetitive work that still needs judgement is exactly what these models are good at.
Drafting. Replies to routine enquiries, summaries of a long conversation, first versions of a quote. The key word is draft: someone reviews it before it goes out.
Extracting data from documents. Pulling the fields from an invoice, delivery note or PDF form and putting them into your system, without typing.
Where it does not reach yet
Anything where being wrong is expensive and nobody will check. Sending binding quotes unsupervised, making decisions about payments, promising deadlines. AI gets things wrong with great confidence, and that confidence is precisely the problem.
Conversations with emotional weight. A serious complaint, an angry client, a delicate negotiation. People can tell, and the cost of them telling is high.
Judgement about your business. A model does not know that this client has been with you eight years, or that this supplier let you down in January. That context is written nowhere, and it is exactly what good decisions rest on.
How to choose where to start
Look for tasks that meet three conditions: they repeat often, they follow reasonably clear rules, and a mistake is not catastrophic. That intersection is where AI automation pays back quickly with low risk.
First-line enquiry response is usually the best first project in most businesses: high volume, clear patterns, and a mistake is corrected in the next message.
Avoid starting with the most impressive-looking project. The one that demos well is rarely the one that saves you the most hours.
A warning about cost
Automating a bad process only makes it go wrong faster. Before automating anything, it is worth asking whether that step should exist at all. Sometimes the best automation is removing the task.
And budget for maintenance: an AI system is not fire-and-forget. Models change, your business changes, and answers need reviewing. Anyone selling you "set it up and forget it" will not be there when it stops working.