The job title covers a wide range of people. It includes a former data scientist with a laptop and a partner at a global firm with a team of thirty. The useful work is the same at any scale.
Finding where the hours go
The first thing an AI consultant does is sit with the people doing the work. That means the people who process the invoices, write the quotes or key the orders rather than the leadership team. They map what happens, how long it takes, where it goes wrong and what the person doing it knows that is not written down anywhere.
This is the whole job. Most AI projects that fail were designed by people who never watched the work being done.
Deciding what is worth fixing
The map usually produces a dozen candidates. A good consultant scores them on four things. Hours consumed, cost when they go wrong, how messy the inputs are, and how many systems are involved. Most candidates do not survive.
What comes out is a short list, usually two or three items, with a reason for each thing that was cut. That list, priced, is the roadmap.
Building or supervising the build
Some consultants build. Some hand the roadmap to your team or to a developer. Either is fine as long as it is clear up front.
If they build, the work is putting an automation into production against your real systems. They test it against your own history so you know its error rate before it goes live. They design the path for the exceptions it cannot handle.
If they supervise, the work is making sure whoever builds it does those things.
Handing over the keys
The last part is the one most often skipped. Someone inside your business must be able to run what was built. When the business changes, they change the rules. When it breaks, they know what to do. A consultant who leaves without doing this has left you dependent on them.
What a consultant is not
An AI consultant is not a vendor. If they take referral fees from software they recommend, they are a reseller. Read their advice with that in mind.
They are not a guarantee. Any consultant who promises a specific saving before discovery is guessing.
And they are not necessary for every business. If your problem is common and a product solves it well, buy the product. The consultant earns their fee in three cases. The problem is specific to you. Several systems are involved. Or you cannot tell which vendor is telling the truth.