What Does an AI Consultant Actually Do?

If you’ve searched “AI consultant” you’ve probably already noticed the problem: everyone’s calling themselves one. Former IT managers. Management consultants who’ve added “AI” to their LinkedIn headline. Freelancers who built a chatbot once. Big Four partners charging Big Four rates.

The title covers a vast spectrum, and that makes it almost meaningless without context. So let’s fix that. What does an AI consultant actually do, what should you expect from one, and how do you tell the good ones from the ones who’ll burn through your budget while delivering PowerPoint?

The Job, in Practice

Forget the marketing. When a competent AI consultant works with a business (particularly an SME) the work falls across six phases. Not all consultants do all six. Some specialise. But the full picture looks like this:

1. Discovery and Assessment

Before anything else, they audit what you’ve actually got. Your processes, your data, your tech stack. They work out where AI can add measurable value, and, just as importantly, where it’s a waste of money. This phase is about understanding your business, not selling you AI.

2. Strategy and Roadmap

Once they know what you’re working with, they prioritise. Which use cases offer the highest impact for the lowest effort? What does success look like? Not in tech terms, but in business terms. Cost saved, time recovered, revenue unlocked. A good consultant will give you a roadmap with sequenced priorities, not a wish list.

3. Vendor and Tool Selection

Build, buy, or partner? That’s the core decision. An AI consultant evaluates the options, shortlists platforms or tools, and runs proof-of-concept tests. They should be vendor-neutral. If they recommend the same platform to every client, ask why (and check whether they’re a reseller).

4. Implementation Support

This is where the work gets real. They oversee or directly manage integration, working with your internal team or dev partners to handle data preparation, model selection, prompt engineering, and getting the thing actually working in your environment.

5. Change Management

This is the phase most businesses don’t budget for and most AI projects fail on. Training staff. Redesigning workflows. Addressing resistance, because there will be resistance. People worry about their jobs. They’re sceptical of new tools. They’ve seen “transformation” projects before. A good consultant addresses this head-on.

6. Ongoing Optimisation

AI isn’t set-and-forget. Models drift. Prompts need refining. New opportunities emerge. The best consultants build in a period of monitoring and iteration, and, crucially, transfer enough knowledge that your team can eventually do this without them.

Where the Time Actually Goes

There’s a common misconception that AI consultants spend their days building AI models. Most don’t. A McKinsey study found that only around 10% of AI project effort goes into the actual model. The rest is data preparation, integration, and organisational change.

That’s worth sitting with for a moment. The technology is the easy part. The hard part is your data, your people, and your processes. That’s what a consultant spends most of their time on.

How to Evaluate an AI Consultant

Five questions that separate the credible from the questionable:

“Can you show me a case study where you measured ROI, not just built something?” If they can only show what they built but not what it achieved, that’s a red flag. Delivering a tool isn’t the same as delivering value.

“What would you tell us NOT to do?” Good consultants push back. If every idea you float gets a “yes, we can do that,” they’re agreeing their way to an invoice, not advising you.

“How do you handle data we don’t have?” Most SMEs have messy, incomplete data. The answer to this question reveals whether they’ve actually worked with real businesses or just enterprises with dedicated data teams.

“What happens after you leave?” Dependency is the business model for bad consultants. Good ones build your internal capability. You should be less reliant on them over time, not more.

“What’s your stance on off-the-shelf vs custom?” The answer should be nuanced. Anyone who always recommends custom solutions (or always recommends off-the-shelf) isn’t thinking about your situation. They’re selling their preference.

Red Flags

Watch for these:

• Leads with technology, not business problems. If the first meeting is about LLMs and neural networks rather than your P&L and operations, they’re showing off rather than listening.
• Can’t explain things without jargon. If they can’t make it clear to a non-technical business owner, they either don’t understand it well enough or don’t respect your time.
• No experience at your scale. Enterprise consultants often struggle with SMEs. Different budgets, different timescales, different constraints.
• Proposes a six-month discovery phase before delivering any value. Discovery matters, but it shouldn’t take half a year. Expect quick wins alongside strategic planning.
• Won’t commit to measurable outcomes. “We’ll explore the AI landscape and identify synergies” is not a deliverable.

Green Flags

And look for these:

• Starts by understanding your business, not pitching AI.
• Has delivered for companies of similar size and complexity.
• Can articulate what AI is bad at, not just what it’s good at.
• Proposes quick wins alongside strategic plays, proving value early while building towards something bigger.
• Talks about people and process as much as technology.
• Leaves documented processes and playbooks behind so your team isn’t stranded when the engagement ends.

What Should You Expect to Pay?

Not a price list. A number that knows nothing about your business tells you nothing about whether a proposal in front of you is fair. What does help is understanding how the number is built, because then you can take it apart yourself.

Every consulting quote is people, multiplied by days, multiplied by a rate. Ask for all three. Multiply them and see whether you arrive at the total on the front page. If you do not, ask what the difference is. There may be a perfectly good answer, and hearing it is the point.

Duration is the part that is reasonably predictable. What sits inside it is what you should be interrogating.

Engagement Type	Typical Duration	What Moves the Number
AI readiness assessment	2–4 weeks	How many systems the data sits across, and how many people have to be interviewed before the answer is honest
Strategy and roadmap	4–8 weeks	How much of the business is in scope, and how many people have to agree on the answer
Proof of concept	4–12 weeks	Whether the data is usable as it stands, and whether it has to talk to a live system
Full implementation	3–12 months	Integration with what you already run, and how much of the process has to be redesigned before it can be automated
Fractional AI leadership	Ongoing	How many days a month, and whether the person is advising or building

Rates move with seniority and with overhead. An independent is charging you for one person’s time. A large firm is carrying a great deal that a small one is not, and that shows up in the rate whether or not it shows up in your project.

For most UK SMEs, independent or boutique work is better value. They are more hands-on and less likely to send a junior to do the work, and the person who sold it to you is usually the person who turns up.

Do You Actually Need One?

Honest answer: not always.

Many AI tools (ChatGPT, Microsoft Copilot, off-the-shelf SaaS) don’t need a consultant. You can trial them yourself at minimal cost. If you’re just looking to automate meeting notes or draft marketing copy, save your money and experiment.

You need a consultant when:

• You’re integrating AI into core business processes where getting it wrong has real consequences.
• Your data is complex, sensitive, or spread across multiple systems.
• You’ve tried AI before and it didn’t stick.
• The investment is significant enough that a structured approach will save you from expensive mistakes.
• You need someone to challenge your assumptions, not just execute your brief.

The Bottom Line

An AI consultant’s value isn’t in the technology they know. It’s in the problems they’ve solved. The best ones will save you from wasting money on the wrong things, get you to value faster, and leave your team stronger than they found them. The worst ones will sell you a project, deliver a deck, and move on.

Know the difference before you hire.

Chat with Cass about what you’re trying to fix