What an AI Consultant Actually Does — and When You Need One

“AI consultant” has become one of those job titles that means everything and nothing. It covers the person who’ll build you a working customer-service assistant, and the person who’ll charge you five figures for a slide deck titled “Your AI Journey.” Same words; wildly different value.

So before you hire one — or decide you don’t need to — it’s worth being clear about what a good AI consultant actually does, what a session or engagement should feel like, and how to spot the ones who are mostly selling the word “AI.”

The four things a good one actually does

Strip away the buzzwords and useful AI consulting comes down to four jobs:

1. Find the value. Look at how your business actually runs and identify where AI would genuinely save time or make money — and, just as importantly, where it wouldn’t.
2. Prioritise. Turn a long list of “we could…” into a short list of “we should, first” — ranked by impact and how easy each one is to do.
3. Implement. Actually build the first pieces with you — the automation, the assistant, the workflow — not just recommend them.
4. Transfer the capability. Leave your team able to run and extend what’s been built, so you’re not paying forever for something you could own.

The first two are strategy. The second two are where most of the value is — and where a surprising number of “consultants” quietly stop.

What a good engagement looks like

You shouldn’t need a six-month contract to get value. A sensible engagement usually moves through three stages:

Notice what’s not here: a generic “AI landscape” briefing, a tour of forty tools, or a maturity model. Those make a consultant look knowledgeable. They rarely change anything in your business.

The red flags

Because “AI” sells, the field has attracted people long on vocabulary and short on results. A few warning signs:

The opposite is a good sign: someone who shows you working software they’ve built, is honest about what AI can’t do, and scopes a small first piece of work rather than a grand programme.

Consultant, training, or both?

Here’s the part most consultants won’t tell you: you might not need one. For a lot of small and mid-sized businesses, the better value is learning to do it yourself — because then the capability stays in your team instead of leaving with the invoice.

The honest answer is often both, in that order: a little hands-on training to build confidence and cover the everyday wins, plus focused implementation help for the one bigger thing that’s worth building properly. What you rarely need is a long advisory retainer that never quite gets to the building.

What a useful proposal should contain

Do not evaluate proposals by the number of AI products named. A credible first engagement should make seven things concrete:

1. The problem: the current workflow, its users and the evidence that it is worth changing.
2. The outcome: an observable result, not “AI transformation”.
3. Scope: the first use case plus an explicit list of what is not included.
4. Data boundary: what data enters the system, where it goes, who can access it and how it is deleted.
5. Acceptance tests: examples the system must handle, including failure and edge cases.
6. Human control: who reviews which outputs and which actions the system cannot take alone.
7. Handover: documentation, ownership, supplier access, maintenance and the route out.

Ask to see the acceptance tests before the polished roadmap. They reveal whether the consultant understands your real work. Also ask what would make them recommend not using AI. Someone unable to answer is testing a sales proposition, not your problem.

The 2026 UK context: capability is part of implementation

The June 2026 Skills for AI upskilling research found that AI is becoming embedded in everyday UK work while workforce capability often struggles to keep pace. Its PRIMES framework emphasises learning that is practical, reachable, integrated, modular, expandable and sustainable.

That matters to consulting because a technically working system can still fail if the team does not understand:

• what the tool is allowed to do;
• how to recognise a weak or unsafe output;
• when to escalate to a person;
• how performance and errors are recorded;
• who owns the workflow after the consultant leaves.

Training is therefore not an optional presentation at the end. For any system people will use weekly, knowledge transfer is part of the implementation itself.

What it should cost

Prices are all over the place, so anchor on value instead of headline figures. A focused session to identify and set up one or two high-value automations can pay for itself almost immediately — and it’s a far safer first step than a big open-ended engagement. Start small, prove it works on your real business, and only scale up the commitment once something specific has earned it.

The short version

A good AI consultant is really a translator and a builder: they turn “we should use AI somehow” into a couple of specific, working things that save you time this month — and they leave you more capable, not more dependent. Anyone who can’t show you something they’ve actually built, or who wants a large cheque before understanding your business, is selling the word, not the work.

If you’d like an honest look at where AI genuinely fits in your business — and a plan you could act on this month — that’s exactly what AI consulting in London is for. Prefer to build the skill in-house? Book a private lecture and we’ll work on your real business together.