What does an AI consultant do?

An AI consultant helps a business decide where AI should change how the work gets done, and then makes that change happen with the team. That’s the whole job. Everything else is detail.

The detail matters though, because “AI consultant” now covers everything from a person selling prompt courses to a firm that builds custom software. So here is what the job looks like when it’s done properly, from someone who does it inside agencies.

The short answer

A good AI consultant does five things:

1. Finds the work where AI will earn its place first.
2. Chooses the tools and sets the rules for using them.
3. Builds the first new ways of working with the team.
4. Brings the rest of the team with it.
5. Hands it over, so the business runs it without them.

If someone only does the first one, they are a strategist. If they only do the third, they are a developer. The value is in doing all five, in that order.

What an AI consultant does, week to week

Finds where AI earns its place

Most businesses already use AI. A few people have ChatGPT open all day, a few refuse to touch it, and nobody has written down what it’s for. The first job is to look at how the business really works: where the time goes, which jobs repeat, where mistakes are expensive, and what clients are paying for.

From that you get a short list. Not fifty ideas, three or four, with the first one chosen because it’s useful, visible and low risk.

Chooses the tools and sets the rules

There are more AI tools than anyone can keep up with, and most businesses are paying for several that overlap. A consultant decides which ones the business will standardise on, and writes down the rules: which tools can touch client data, what an agent is allowed to do on its own, and which decisions always stay with a person.

This is the part people skip, and it’s the part that keeps you out of trouble.

Builds the first new ways of working

Advice without building doesn’t change anything. The consultant should sit with the team and build the first workflow on real work, to the business’s own standards, and check that what comes out is good enough to put in front of a client.

In an agency that might be an agent that writes a first pass to your house style, or a QA step that tests a site the way a real user would before it goes live.

Brings the team with it

Tools are the easy part. People are the work. The pattern I use is to find one person who is already curious, start them on the job they’re sick of doing, and let everyone else see it working. It sticks when the team owns it, and it fails when it’s handed down as an order.

Hands it over

The aim is a business that runs this without the consultant. That means documentation, standards your team can maintain, and a point where the consultant steps back. If someone is setting things up so you’ll need them forever, that tells you something.

A good AI consultant should be trying to make themselves unnecessary.

What an AI consultant is not

• Not a software vendor. They shouldn’t be paid by the tools they recommend.
• Not a prompt course. Training matters, but a workshop on its own changes very little by the following month.
• Not a strategy deck. If the output is slides and no working change, you’ve bought a report.
• Not a replacement for your team’s judgement. AI takes on more of the production. Your people still decide what goes out and own the result.

Consultant, consultancy, agency or fractional lead

These get used as if they mean the same thing. They don’t.

I work as the last one: an independent consultant who acts as an agency’s AI lead, part-time. There’s more on that in what a fractional chief AI officer is.

When you need one, and when you don’t

You don’t need a consultant to start. Give your team access to a good model, let them try it on their own work, and see what happens.

You probably do need one when:

• AI use is scattered, and nobody can say what the business gets from it.
• Client or customer data is going into tools nobody has checked.
• You’ve run experiments, and none of them became the normal way of working.
• Your competitors are quoting faster or cheaper, and you can’t see how.
• Your team is split between people who love it and people who are worried about their jobs.

If none of those sound familiar, carry on as you are.

How to tell a good one

Ask these before you hire anyone, including me:

• What have you changed in a business like mine? You want specifics, not a client logo wall.
• What will you build yourself? If the answer is nothing, you’re buying advice.
• What would you leave alone? Someone who thinks AI fits everything hasn’t done much of it.
• How do you handle client data? There should be a clear answer before anything is switched on.
• How do we run this when you’ve gone? The answer should include your team’s names, not theirs.

What it looks like in an agency

Agencies are where I work, because I’ve run one for 20 years and rebuilt four of my own around AI. The shape is the same each time.

It starts with a fixed Blueprint that takes about two weeks: a half-day with the owners and leads, short conversations with the team, a look at the tools and where the time goes, and a written plan for where AI should change the agency first and what to leave alone. Then a monthly plan, where I’m inside the team for a set number of days, building alongside them. And at the end, a handover.

If you want to talk through where your agency is, send me a message. If I’m not the right fit, I’ll tell you who is.