How to use AI in a small business: a practical guide

If you run a business with somewhere between three and a hundred people, you have probably reached the same conclusion most owners have: AI is clearly useful for something, and nobody has explained what.

This guide is the practical version. No predictions, no transformation language. Just where the money actually is for a business your size, what it costs, and what to do in the first month.

Start with the constraint, not the technology

The single most common mistake is starting with a tool. Somebody reads about a product, buys it, and then hunts for a problem it solves. That order almost never works.

Start instead with a simple question: what is the thing that stops your business handling more work without hiring?

For most small businesses the honest answer is one of four things. The phone goes unanswered. Quotes and enquiries take too long to turn around. Someone spends hours a week moving information between systems by hand. Or the person who knows how everything works is the bottleneck for everything.

Each of those has a well-understood answer now, and none of them require you to become a technology company.

The four areas that pay back fastest

1. Customer conversations that happen without you

A talking virtual agent answers the phone, handles the routine questions, books the appointment and writes the details into your system. It does this at seven in the evening and on a Saturday, which is when a meaningful share of enquiries actually arrive.

For a trades business, a clinic, a lettings agency or a garage, this is usually the highest-return thing available. The maths is not subtle: if you miss six enquiries a week and convert one in three, you are losing two jobs a week to a ringing phone.

2. The admin nobody should be doing

Copying details from an email into a spreadsheet. Re-keying an order. Chasing a supplier for a delivery date. Producing the same report every Monday.

Workflow automation handles the shape of the task, and a model handles the judgement inside it, such as reading an unstructured email and pulling out the right fields. This is unglamorous work and it is where most small businesses recover the first ten hours a week.

3. A customer record that stays honest

Most small businesses do not have a CRM problem. They have a data-entry problem that makes the CRM useless. Notes never get written up, follow-ups get forgotten, and after a year nobody trusts what is in there.

An AI-assisted CRM writes the notes from the call, flags the customers who have gone quiet, and drafts the follow-up. The value is not the software. It is that the record finally reflects reality.

4. The tool that does not exist yet

Every established business has one process that is genuinely theirs, run on a spreadsheet that only one person understands. Historically, having software built for that was a six-figure conversation. It is not any more. A focused internal tool is now a matter of weeks, which changes which problems are worth solving.

What it actually costs

Be sceptical of anyone who answers this without asking questions first, but some honest ranges help.

A single, well-scoped automation, such as enquiry handling or document processing, is typically a low four-figure piece of work plus modest running costs. A voice agent handling real customer calls is a larger project, though usually comparable to a month or two of the salary it offsets. A bespoke internal tool depends entirely on scope, but the useful benchmark is that a first working version should be in your hands within weeks, not quarters.

The running costs surprise people in the other direction. Model usage for a typical small business workload costs tens of pounds a month, not thousands. The cost is in getting it built properly, not in the electricity.

Do you need technical staff?

No, and this is the part that has genuinely changed. What you need is someone internally who understands the process well enough to say when the output is wrong. That is a domain expert, not a developer.

What you should not do is hand the whole thing to a member of staff as a side project because they are good with computers. It will get to eighty per cent, stall there, and quietly become another system nobody trusts.

Can it be customised to how we actually work?

This is the most common question we get, and the answer is that customisation is the entire job. A general-purpose assistant that knows nothing about your pricing, your suppliers or your booking rules is a novelty. The work in any real implementation is connecting it to your actual data and constraining it to your actual rules.

If a supplier cannot explain how their system will learn your specifics, they are selling you a demo.

A 30-day plan

Week one: measure one thing. Pick the constraint you identified above and count it for a week. Missed calls. Hours spent on the admin task. Days from enquiry to quote. You need a number, because without one you cannot tell whether anything improved.

Week two: scope narrowly. Choose one task that happens at least twenty times a week, follows a recognisable pattern and carries low risk if it occasionally gets something wrong. Resist the urge to solve everything. The businesses that get value do one thing properly first.

Week three: build or buy the smallest useful version. Not the full vision. The version that handles the most common case and hands anything unusual to a person.

Week four: run it alongside the human process. Compare the output. Fix what is wrong. Only then remove the safety net.

Then measure the same number you measured in week one. If it did not move, you learned something cheap. If it did, you now have an internal case study and the confidence to do the next one.

The mistakes that waste the most money

Buying a platform before defining a problem. Starting with the most complex process because it is the most annoying. Automating something without first fixing the broken process underneath, which just produces mistakes faster. And treating it as a project with an end date rather than something that needs an owner.

Where to start today

If you can name your constraint, you are most of the way to knowing what to do about it. If you cannot, the fastest way to find out is to describe your business plainly to someone who has seen the same problem in fifty other companies, and let them tell you what is normal and what is not.

That is what our free AI adviser does in about ten seconds, and what a half-hour conversation with us does properly.