What an AI consultant actually does for a 20–100 person business

When a business owner first calls us, the picture in their head is usually the same one: an AI consultant turns up, rolls out ChatGPT to the team, runs a training session, and everyone becomes twenty per cent more productive. Licences, logins, a lunch-and-learn.

That is a perfectly reasonable thing to buy. It is not what we do, and it is not where the money is for a business of your size. If you run a firm of twenty to a couple of hundred people, here is what an AI consultant should actually be doing for you — and how to tell whether the one you're talking to is doing it.

It starts with your processes, not with AI

The businesses we work with all have the same shape. They have grown to the point where they run on a handful of systems — a CRM, an accounts package, a job or case management tool, shared inboxes, and a great many spreadsheets. None of those systems talk to each other properly, so people do the talking for them. Someone re-keys the order from the email into the system. Someone copies figures from the PDF into the template. Someone chases the document that should have arrived on Tuesday.

That manual work between systems is where a 20–100 person business loses its week. It is rarely written down anywhere, it never appears on an org chart, and the people doing it have usually stopped noticing it. Finding it is the first job, and it is mostly a job of asking questions and watching people work.

So a good AI consultant's first engagement is not a build. Ours is a two-hour workshop with the people who do the work, followed by a written map of every place in the business where time is being lost, with an estimate of what each one costs and what it would take to fix. We charge £495 for it, and you keep the map whether or not you go any further. If nothing on that map is worth fixing, we say so — it happens, and it is better to hear it for £495 than after a project.

The AI is the last twenty per cent

Once you know where the effort goes, most of the fix is plumbing: connecting the systems you already pay for so that information moves between them without a person in the middle. That is automation, and much of it needs no AI at all.

The AI comes in at the points where a human used to have to read, judge or write. Reading an incoming document and pulling out the figures. Deciding which of three templates a request needs. Drafting the report, the reply, or the summary that a person then checks and sends. Those are the steps that used to make automation impossible for most businesses — the "someone has to look at it" steps — and they are exactly what modern AI models are good at, provided they are wrapped in enough structure to stop them making things up.

That last clause matters more than anything else in this article. A language model on its own will, sooner or later, produce something confident and wrong. The consultant's job is to build the system around it — the checks, the rules, the "if it isn't sure, ask a human" — so that the business gets the speed without the risk. Anyone selling you AI without talking about that is selling you a demo.

What it looks like in practice

A regulated business came to us with a reporting problem. Every reporting cycle, their compliance team spent three to four days searching through more than a thousand pages of guidance, cross-referencing several frameworks, and hand-building the reports. It was slow, it was inconsistent between people, and a single missed clause could mean a breach. They needed to do more of it without hiring more people.

The instinct is to hand the team ChatGPT and let them ask it questions. We did not do that, because a general-purpose model does not know their guidance, will happily invent a clause, and leaves no audit trail.

Instead we built a system around their specific document library. It indexes the full set of guidance and keeps it current as regulations change. When a report is needed, it retrieves the relevant sections, drafts the report in the firm's own templates, and — this is the part that made it usable — passes every statement through a validation layer written in plain code that checks the output against the source before anything reaches a person. If a claim cannot be traced back to the guidance, it does not go in.

The results: report preparation time down 93%, from three or four days to under two hours. Zero compliance errors since it went live. Around £45,000 a year of skilled time freed up — the equivalent of one and a half people, who now do advisory work instead of searching PDFs. And because the system runs at the same cost whether it produces ten reports or a hundred, the team can respond to new guidance the day it is published. Read the full case study

None of that came from a chatbot. It came from understanding one process in detail and building one system to remove it.

What it costs, and how it starts

Owners often assume this is a six-figure, twelve-month, "get the data team in" exercise. For a business your size it is not, and a consultant who tells you otherwise is describing a different kind of client.

The way we structure it, which is fairly typical of consultancies working with growing firms:

1. Automation Audit — £495, fixed. The two-hour workshop and the written opportunity map described above.
2. Automation Sprint — from £2,500, one month. One system, chosen from the map, built and live within the month, with the hours it saves reported weekly so you can see whether it is paying back.
3. Automation Partnership — from £3,500 a month. For businesses that want two or three systems built and looked after over a minimum of three months, with monthly reporting on hours and cost saved. Most clients end up here once the first system proves itself, because the second and third opportunities are usually obvious by then.

The audit is deliberately cheap and the sprint is deliberately short. You should be able to see a return inside sixty days, and if you can't, you should stop.

When you don't need one

Honesty is cheap and it saves everyone time, so: there are businesses that should not hire an AI consultant yet.

If you have fewer than ten staff, there is usually not enough repetitive work to justify a custom build; off-the-shelf tools and a few hours of setup will serve you better. If your problem is that people don't know how to use the tools they already have, you need training, not systems. And if your processes change every month because the business is still finding its shape, wait until they settle — automating a process that is about to change is money spent twice.

The audit is designed to catch all three. It is a bad outcome for us to sell a sprint to a business that doesn't need one, because it is the second project, not the first, that makes a client worth having.

Five questions to ask any AI consultant

If you are talking to more than one consultancy — and you should — these will separate the builders from the demo-givers.

1. "What will you do before you write any code?" The answer should involve your people, your processes and a written map. If it involves a platform demo, keep looking.
2. "How does your system stop the AI from making things up?" Listen for validation, source-checking, human sign-off on anything consequential, and audit trails. "The model is very accurate" is not an answer.
3. "What happens to our data?" You want to hear where it is hosted, that public models are not trained on it, and who can see it. Vagueness here is disqualifying.
4. "Show me a result with numbers." Hours saved, errors removed, money recovered — for a business of roughly your size. Logos without numbers tell you nothing.
5. "What does month two look like?" A good consultant has a view on what comes after the first system and how you would run it without them. A bad one has a view on your renewal.

Where to start

If you recognise your business in the second paragraph of this piece — the spreadsheets, the re-keying, the chasing — the useful next step is small: two hours with your team and a map of where the time goes. Whether you do anything with it is up to you. Book an Automation Audit