The term, unpacked
What does an AI expert actually do?
Last updated
Justas Butkus is an AI expert and fractional AI officer based in Vilnius, Lithuania, working with mid-market companies and scale-ups across the UK, EU and US at two to four days a month. He builds and operates the production AI systems he advises on, rather than only advising on them.
Short answer
"AI expert" is not one job. It covers researchers, engineers, implementation specialists, governance advisors and strategy leaders – five roles with different skills, different economics and different failure modes. Most companies looking for "an AI expert" need an implementation specialist or a strategy lead, and hire a researcher.
Which kind of AI expert do you actually need?
The term is doing far too much work. These five people all get called "AI expert" and only one of them is likely to solve your problem.
| Type | What they are good at | Hire when |
|---|---|---|
| Researcher | Advancing what models can do | You are building a model, not using one. Rare outside labs. |
| ML / AI engineer | Building and shipping systems | You know what to build and need it built well. |
| Implementation specialist | Getting AI into an existing business process | The technology exists; making it work in your operation does not. |
| Governance / compliance advisor | Risk, audit, regulatory obligations | A regulator, insurer or customer is asking questions. |
| AI strategy lead | Deciding what is worth doing at all | You have budget and no ranked plan. |
The common failure is hiring the second when you needed the fifth. A strong engineer builds exactly what you asked for, and if the request was wrong, you get a well-engineered system nobody misses when it stops.
How do you tell a real AI expert from a confident one?
The field is two years into a hype cycle and the supply of confident opinion vastly exceeds the supply of operating experience. Three questions separate them quickly.
- What have you personally put into production, and can I look at it? Not a prototype, not a client under NDA – something running that you can inspect. This is the single most discriminating question available.
- What would you tell me not to do? Anyone who cannot name work they would decline is selling capacity, not judgement.
- What happens when it produces a wrong answer? Everyone who has operated a real system has an immediate, specific answer. Everyone who has not will talk about accuracy rates.
Why do most AI projects fail, and what does that say about who to hire?
The failure rate is documented. RAND reports that by some estimates more than 80% of AI projects fail, roughly twice the rate of non-AI IT projects, and MIT NANDA found 95% of enterprise generative AI pilots delivering no measurable return. Neither attributes it to the technology: RAND points at the problem being misunderstood and models optimised for the wrong metric, MIT at approach rather than model quality.
All three are selection and accountability problems, not technical ones. That is why the person who can *decide* what to build usually matters more than the person who can build it – and why the strongest position is someone who can do both, because then the decision is made by someone who knows what it costs to implement.
What do I do?
I build and operate production AI systems, and a limited number of companies engage me to own their AI direction: what gets built, in what order, under what governance, and what gets refused.
- Two to four days a month, a limited number of engagements at a time.
- A fixed-scope diagnostic first, with a written roadmap you keep either way.
- Milestones rather than hours, written into the engagement.
- You hold the keys – accounts, infrastructure and documentation are yours throughout.
- Independence in writing – I own an AI company and do not procure from it in an advisory role.
Frequently asked questions
What does an AI expert do?
It depends which of five distinct roles they occupy: researcher, engineer, implementation specialist, governance advisor or strategy lead. Most mid-market companies searching for "an AI expert" need someone who can decide what is worth building and then get it into production, not someone who advances model capability.
How do I know if an AI expert is credible?
Ask what they have personally put into production and can show you, what work they would decline, and what happens when a system produces a wrong answer. Operators answer the third question instantly and specifically; people who have only advised talk about accuracy rates.
Do I need an AI expert or an AI agency?
An agency builds what you specify. An expert decides what should be specified. If you already know exactly what to build and why it pays for itself, use an agency. If that is the open question, the specification is the work.
What does an AI expert cost?
It varies enormously by role and scope, and most published figures are rate cards from firms selling the service rather than independent data. What moves the number is days per month, regulatory exposure and whether building is included.
Can one person cover both strategy and implementation?
Rarely, which is why the common pattern pairs an advisor with separate engineers. Where one person does both, the handoff disappears – and that handoff is where most of the value usually leaks away.
If you are trying to work out which kind you need
That is usually a short conversation, and it is worth having before you write a brief or a job description.