Search "what does an AI consultant do" and most of what comes back is written for people who want to become AI consultants. Skills to learn, certificates to collect, how to find your first client.

Useful if that's you. Less useful if you're the one about to sign the cheque. This guide is for the buyer.

The short answer: an AI consultant helps a business decide where AI is worth using, then gets it working. In practice that means finding the few processes where AI pays back, checking your data and systems can support them, choosing tools or a build approach, running a pilot, and making sure your people actually adopt it. Good ones also set the guardrails: data protection, governance and a way to measure results.

Key points:

  • About 35% of UK businesses with ten or more staff now use at least one AI technology, nearly three times the level of late 2023, but only 10% of those users describe their use as extensive (ONS, July 2026)
  • In McKinsey's global survey, nearly nine in ten respondents reported AI use in their organisation, yet only 37% attribute any profit (EBIT) impact to it (McKinsey, August 2026)
  • The organisations getting value put around 70% of their AI effort into people and processes, not algorithms (BCG)
  • That gap between "using AI" and "getting value from AI" is the job. It's what you're hiring a consultant to close.

If you're thinking of becoming an AI consultant rather than hiring one, my guide to becoming an AI consultant covers the skills and fees side.

The AI consultancy guides

This is the hub for a 20-part series on buying AI help: what to ask, what it costs and what to avoid. Start wherever your problem is.

person holding pencil near laptop computer
Illustrative photograph. Photo: Scott Graham / Unsplash.

What does an AI consultant do day to day?

Less AI than you'd think. A good engagement is mostly listening, mapping and deciding, with a smaller amount of building at the end.

The work usually runs in this order:

  1. Discovery. Interviewing the people who do the work. Where does time go? Which tasks are repetitive, rules-based or buried in documents? Where do errors cost money?
  2. Use-case selection. Turning a long list of "we could use AI for..." into a short list ranked by value, feasibility and risk. Most of the value of a consultant sits in this step, because it's where bad ideas get killed cheaply.
  3. Readiness checks. Is the data where the use case needs it? Is it clean enough? Can the systems talk to each other? What does UK GDPR say about using it this way?
  4. Tool or build decision. Off-the-shelf tool, a configured platform, or something custom. Most small and mid-sized businesses should buy before they build.
  5. Pilot. One use case, real data, a baseline and a pass mark agreed before anyone starts.
  6. Rollout and adoption. Training, new ways of working, someone who owns it internally.
  7. Governance and measurement. A usage policy, who approves what, and a simple way to show whether it's paying back.

Day to day, that's workshops, process maps, spreadsheets of use cases, vendor calls, a fair amount of prompt and workflow testing, and a surprising number of conversations about who's going to own the thing once the consultant leaves.

The ONS findings show that adoption and depth of use are different measures. They do not tell you whether a particular firm needs more AI. A small, well-used capability can be enough; the test is whether it improves the work you care about.

What are you actually paying for?

Judgement, mostly. And a set of things you can hold in your hand at the end. If a proposal doesn't list deliverables like these, ask why.

Stage What you should receive What it lets you do
Discovery and assessment A ranked list of use cases with value, effort and risk for each; a readiness verdict on data and systems Decide whether to spend more, and where
Strategy and roadmap A sequenced plan for the next 6 to 12 months, with owners, costs and decision points Budget it and hold people to it
Tool selection A short comparison of options against your requirements, including data handling and total cost Buy once, not three times
Proof of concept A working pilot on your real data, a baseline, results against an agreed pass mark Make a go or no-go call on evidence
Implementation A system in production, documentation, access you control, a handover Run it without the consultant
Governance and training An AI usage policy, an approvals process, training for the people who'll use it Use AI safely at scale

Notice the last column. Every deliverable should leave you able to do something you couldn't do before. A 60-page "AI strategy" that nobody can act on is the most expensive document in the building.

For what each of these typically costs, see how much an AI consultant costs.

What types of AI consultants are there?

"AI consultant" covers everyone from a former Big Four partner to someone who discovered automation tools last spring. The label tells you very little. What they actually do tells you a lot.

Type What they're good at Hire them when
Strategy adviser Board-level priorities, roadmaps, operating model, business cases You don't yet know where AI fits or how much to spend
Technical implementer Integrations, custom workflows, agents, data pipelines You know the use case and need it built properly
Automation specialist Process automation on tools like Zapier, Make or Power Automate, with AI steps You have clear, repetitive admin work to remove
Sector specialist AI applied to one industry's workflows and regulations You're in a regulated sector such as health, legal or finance
Trainer and change lead Literacy programmes, usage policies, adoption You've bought the licences and nobody's using them well
Fractional AI lead Ongoing ownership of the AI agenda, a few days a month You need a senior owner but not a full-time hire

The best people span two or three of these. Be wary of anyone who claims all six. And if you're weighing a solo consultant against a firm that builds things, read AI consultant vs AI agency first, because the incentives are different.

Do you actually need an AI consultant?

Not always. Some honest cases where you don't:

  • You want your team using ChatGPT, Copilot or Claude better day to day. A good training programme and a usage policy will do more than a consultant. See AI training for employees.
  • You've already got a capable technical lead who has the time. Give them a budget and a deadline instead.
  • You're buying a single, well-understood tool, such as an AI receptionist or meeting notes. Talk to three vendors and check references.

You probably do need one when the decision is expensive or hard to reverse. Custom builds, anything touching customer data, regulated workflows, and situations where three departments each want a different AI tool and someone has to choose. Also when you've already run a pilot that went nowhere and can't say why.

The McKinsey 2026 survey puts AI high performers at about 6% of respondents. Nearly three-quarters of them had fundamentally redesigned workflows around AI, against a quarter of everyone else. Redesigning work is the hard part, and it's rarely the part a software vendor will do for you.

Not sure whether you're ready? Run through the AI readiness checklist before you call anyone.

How do you tell a good AI consultant from a bad one?

Watch what they ask in the first meeting. A good one asks about your business: margins, bottlenecks, where customers complain, what your team hates doing. A weak one tells you about models.

Green flags:

  • They talk you out of at least one idea.
  • They ask about your data before they talk about tools.
  • They propose a small, fixed-scope first step with a clear stop point.
  • They can show something they've built and run, not just advised on.
  • They plan for the handover from day one: documentation, access, an internal owner.

Red flags:

  • A tool recommendation before they've seen a single process.
  • Undisclosed referral fees or reseller margins on the software they recommend.
  • No mention of data protection, security or who can see what.
  • Success described as "efficiency" or "transformation" with no number attached.
  • They own the accounts, the code or the prompts, and you'd have to ask permission to leave.

The full list of questions to ask is in how to choose an AI consultancy.

What I've seen in practice

I've spent more than 20 years turning technology into working businesses, including co-founding Telenor Health, whose Tonic platform reached around 9 million subscribers. The technology was never the bit that decided success. Distribution, operations and people were.

AI is the same, only faster. BCG's research found that around 70% of the challenges in AI programmes are people and process problems, 20% technology and only 10% the algorithms. Yet the algorithms get most of the airtime in sales meetings.

The common mistake is starting with the tool. A business buys licences, runs a lunch-and-learn and waits for productivity to appear. Six months later, usage has drifted down to a handful of enthusiasts. Nothing broke. Nothing changed either.

The second mistake is the pilot that can't fail. No baseline, no pass mark, so every pilot is declared a success and none of them reach production. I cover how to avoid that in how to run an AI proof of concept.

What works is narrower than people expect. One process, one owner, one number. Get that live, then do the next one. I run my own consulting this way: AI handles research, synthesis, first drafts and repeatable workflows, and I've built a working lab of AI agents, Ascendancy IQ, that do real tasks under human approval gates. The lesson from both is the same. The controls matter more than the cleverness.

How to hire an AI consultant

  1. Write down the problem, not the solution. "Quotes take four days to turn round" beats "we want an AI chatbot".
  2. Pick one or two processes to start. High volume, repetitive, measurable. Note how they perform today.
  3. Set a budget range for a first phase. A discovery or assessment, not the whole programme. Use the cost guide to sanity-check it.
  4. Shortlist three. Mix types if you're unsure: one adviser, one implementer, one specialist in your sector.
  5. Ask each for a written first-phase proposal with deliverables, people, days, price and a stop point.
  6. Check two references for work that went to production, not just pilots.
  7. Agree ownership up front. Your accounts, your data, your code, your prompts.
  8. Name an internal owner before the consultant starts. If nobody inside owns it, it won't survive their departure.

Frequently asked questions

What does an AI business consultant do? They identify where AI can improve how your business runs, then help you get it working: choosing use cases, checking data, selecting tools, piloting and training staff. The good ones measure the result against a baseline so you know whether it paid off.

What's the difference between an AI consultant and an AI developer? A developer builds what you specify. A consultant helps you decide what's worth building in the first place, and whether to build or buy. Many independent consultants do both, so ask which you're paying for.

Are AI consultants worth it for small businesses? They can be, if the engagement is small and specific. A short, fixed-price assessment that ranks your best two or three opportunities is often worth more than a long programme. For basic tool use, training is usually better value.

What do AI consulting services usually include? Typically an assessment or discovery, a roadmap, tool selection, a proof of concept, implementation support, governance and staff training. Few businesses need all of them at once.

How long does an AI consulting engagement take? A discovery or assessment usually takes a few weeks. A pilot on one use case often runs a further month or two. Ongoing support is normally a monthly retainer or a few days a month.

Sources

If you're weighing up whether to bring in outside AI help and want a straight answer on where to start, let's talk.

Choose from all 20 guides ↗

Keith De Alwis
WRITTEN BY

Keith De Alwis

Independent thinking. Connected expertise.

Meet the mind behind the notes ↗