"AI consultant" can mean almost anything, from someone who runs a workshop on writing prompts to a firm that rebuilds how a company handles its customers. If you run a business of five to fifty people, most of that range is not for you. This guide covers the part that is.
It is written by one, so read it with that in mind. Where we describe prices they are our own, because those are the only ones we can vouch for.
What the job actually is
Strip away the vocabulary and an AI consultant for a small business does three things, in this order.
- Finds the repeated work. The jobs your team does again and again that follow a pattern: answering the same enquiry, chasing the same quote, copying the same details from an email into a spreadsheet.
- Works out which piece would pay back first. Not the most impressive piece. The one where the hours saved or the work won clearly outweighs what it costs to build and run.
- Builds it, or makes sure it gets built, and measures whether it worked. Advice that never becomes a working system is a report in a drawer.
The third step is where consultants differ most. Some stop at recommendations. We think the person who recommends a system should be the one who has to make it work, because it keeps the recommendations honest.
What that looks like in a small firm
The work that suits AI in a small business is usually unglamorous:
- Enquiries. An assistant replies within minutes, at any hour, asks the questions your office would ask, and books a visit or hands over a complete record. You can try a working example on this site.
- Follow-up. Every quote sent gets a polite chase at intervals you set, until there is a yes or a no.
- Admin between systems. Details moved from inbox to job system to invoice without anyone retyping them.
- Reporting. The numbers you look at each week, assembled for you and checked by a person before they are sent.
What it rarely means, at this size, is an "AI strategy", a company-wide training programme, or a chatbot that tries to answer everything.
How an engagement runs
| Stage | What happens | What you get |
|---|---|---|
| A first call | Half an hour on where your team's time goes and how enquiries reach you | An honest view on whether there is anything worth doing |
| An audit | A short, fixed-fee look at your real numbers: hours, enquiries, response times, what a customer is worth | A ranked list of what would pay back first, with the sums shown |
| One build | The system at the top of that list, inside the tools you already use | Something in real use, with one number it is meant to move |
| Run and adjust | Watching that number and changing what does not work | A plain report, and a system that is either earning its place or cut |
The audit comes before the build for a reason. Until someone has looked at your numbers, any recommendation is a guess about businesses like yours, not a finding about yours.
What it costs
Prices across the market vary widely, and we will not quote figures we cannot stand behind. These are ours:
- The audit is £1,500, credited in full against the build if you go ahead.
- An enquiry assistant is priced at 10% of the yearly salary of the person who handles your enquiries, or would, per year, with the build, running and support included. The minimum is £2,400 a year.
Whatever anyone charges, the test is the same: can they show you, with your own numbers, how the system pays for itself? Our missed enquiry calculator does a rough version of that sum in two minutes, and will tell you when the answer is no.
Six questions to ask before you hire one
- Will you look at our numbers before recommending anything? If the answer comes before the question, it is a sales pitch.
- Do you build it, or only advise? Neither is wrong. You need to know which you are paying for.
- Can I see something you have built, working? A live example tells you more than a slide.
- Who owns it when it is finished? You should, without a licence fee for every member of staff.
- What happens to our data? What the system can read, where it runs and who sees the output should be written down before it goes live.
- What would make you tell us not to do this? A consultant who cannot answer has never said no.
When you don't need one
- The work isn't repeated enough. If something happens twice a month, automating it will not pay back.
- An ordinary tool already does it. Much of what gets sold as AI is a feature in software you may already pay for. Check there first.
- Nobody will own it. A system with no named person responsible for it quietly stops being used.
- The real problem is somewhere else. If too few enquiries arrive in the first place, answering them faster is not the fix.
A good consultant will say any of these to you on the first call.