What Does an AI Consultant Actually Do?

Updated June 2026. Rewritten with a practitioner’s view of what an AI consultant actually does, when you need one, and what the work costs.

Ask ten people what an AI consultant does and you will get ten answers, most of them wrong. Some picture a strategy deck and a workshop. Some picture a developer who fine-tunes models all day. The honest answer is that a good AI consultant spends most of their time on two unglamorous things: figuring out which problems are worth solving with AI, and making the solution survive contact with a real business. The model is the easy part.

We are Osher Digital, an AI consultancy based in Brisbane. We have run discovery, built systems, and cleaned up after failed projects across healthcare, recruitment, finance, and professional services. This is the inside view of what the job is, where an AI consultant earns their fee, and where hiring one is a waste of money.

If you are weighing whether to bring someone in, this should give you a clear picture of the work, how to tell a good AI consultant from a confident one, and what a sensible engagement costs. For the kinds of systems this work produces, our pages on AI consulting and AI agent development go into the build side.

What an AI Consultant Actually Does

Strip away the marketing and an AI consultant does four things:

1. Works out where AI actually fits in your business, and where it does not.
2. Designs a solution that fits your data, your systems, and your budget.
3. Builds it, or oversees the build, with the failure cases handled from day one.
4. Makes sure it keeps working and proves it is saving money once the launch glow fades.

The split of time is not even. On most engagements, the thinking and the wiring-into-the-business take far longer than the model work. The title on the contract barely matters here. Whether someone is sold to you as an AI consultant, an AI business consultant, or an AI architect, judge them on this same four-part job, not the label.

The reason is simple. Foundation models are now very good and very accessible. Calling Claude or GPT is a few lines of code. The hard parts are the ones around it: which process to target, what data the model needs, how to handle the cases it gets wrong, how to connect it to your CRM and your accounting system, and how to prove it is actually saving money. That is the work. An AI consultant who only wants to talk about models is telling you what they find interesting, not what you need.

Good consultants are also willing to tell you not to use AI. Plenty of problems are better solved by a database query, a fixed rule, or deleting a process step. If everything looks like an AI problem to the person you hired, check whose interest that serves.

The Work of an AI Consultant, Stage by Stage

Discovery: finding the problem worth solving

This is where the value is decided. The consultant sits with your teams, finds the work that is repetitive, high volume, and rule-bound, and separates it from the work that needs human judgment. They look at where your data lives and whether it is clean enough to be useful. The output is a shortlist of candidates ranked by value and feasibility, not a single grand plan. The biggest failures we are asked to rescue almost always skipped this stage and started building the most exciting idea instead of the most valuable one.

Design: choosing the approach

Once a target is chosen, the consultant designs the solution. Does this need an AI agent that takes actions, or a simple extraction step that reads a document and stops? Which model fits the accuracy and cost profile? Where does a human stay in the loop? What happens when the model is unsure? These decisions shape the cost and the reliability of everything that follows. Our primer on what an AI agent is covers the agent-versus-simple-task call in more depth.

Build: making it real

Now the engineering. Connecting the model to your systems, writing the validation that catches bad outputs before they hit your database, building the human-review path, and setting up the logging you will need when something goes wrong at 2am. An AI consultant who has shipped production systems builds for the failure cases from day one. One who has only built demos finds out about the failure cases in production, on your dime.

Deploy and measure

The last stage is the one amateurs skip. Roll out carefully, watch a sample of real outputs, measure against the baseline you captured before you started, and tune. A document classification agent we deployed for a healthcare client took staff out of hours of manual sorting, but we only know that because we measured the before and after. Without measurement, you have a system you hope is working and no way to defend the spend.

AI Consultant vs Agency vs Hiring In-House

An independent AI consultant or a small AI consultancy gives you senior judgment and moves fast, because the person scoping the work is usually the person doing it. You get experience across many businesses, which means pattern recognition you cannot buy in-house. The trade-off is that they leave when the work is done, so handover and documentation matter.

A larger AI agency brings more hands and can staff a bigger program, but you often pay for layers, and the person who impressed you in the pitch is rarely the one writing the code. Ask who actually does the work.

Hiring an in-house AI engineer makes sense once you have a steady pipeline of AI work and the systems to maintain. Before that, a full-time hire spends most of their time waiting for the next project or building things that do not need building. Most mid-sized businesses are better served by a consultant for the first few projects, then an internal hire once the volume justifies it. A good AI consultant will tell you when you have reached that point, even though it ends their engagement.

When You Actually Need an AI Consultant

You need one when you can see that AI should help somewhere but you cannot tell which problem to start with, or when you have a clear target but no one in-house who has shipped a production AI system. Those are the two genuine cases. A consultant earns their fee by skipping the expensive year of learning what does and does not work.

You do not need one when the task is small and well-defined and your existing developers can handle it with an API call. You do not need one when what you actually want is an off-the-shelf tool that already exists; our rundown of the best AI tools for business covers cases where buying beats building. And you do not need a strategy consultant to produce a forty-slide AI vision deck. If the deliverable is a document rather than a working system, you are buying reassurance, not capability.

The quick test:

• Bring one in when you know AI should help but cannot tell which problem to start with.
• Bring one in when you have a clear target but no one in-house who has shipped a production AI system.
• Skip it when the task is small and your developers can do it with an API call.
• Skip it when an off-the-shelf tool already does the job.
• Skip it when the only deliverable on offer is a deck.

How to Tell a Good AI Consultant From a Confident One

The market is full of people who learned the vocabulary last year. A few questions separate the practitioners from the rest.

• Ask for a system that is still running. Not a pilot, not a demo. Something that has been in production for six months or more, with the warts that come with it.
• Ask what they have told a client not to build. Anyone with real experience has talked a client out of a bad idea. If they cannot name one, they say yes to everything.
• Ask how they handle the cases the model gets wrong. The answer should be specific: confidence thresholds, validation rules, human review for the uncertain cases. Vagueness here means they have not run anything at scale.
• Ask how they measure success. A good AI consultant captures a baseline and reports against it. A weak one points at activity and adjectives.
• Ask which model they would use and why. The answer should reference specific models and the cost and accuracy trade-offs, not “the latest one”.

The tell is honesty about limits. The people who have actually deployed AI are quick to tell you what it is bad at, because they have been burned. The ones selling the dream never mention a downside.

What an AI Consultant Costs

Pricing varies, but here are the ranges we see in the Australian market. Independent AI consultants and small consultancies charge roughly $200 to $400 AUD an hour, or run on a project basis.

A scoped discovery engagement, the piece that tells you what to build and what to skip, typically runs $8,000 to $25,000 AUD and is the best money you will spend, because it stops you building the wrong thing.

A first production build, something focused like a document extraction pipeline or a support agent, generally lands between $25,000 and $80,000 AUD depending on the number of systems it touches and how much validation it needs. Ongoing running costs are usually modest, $50 to $500 AUD a month in model and software fees for most mid-volume systems, with retainers for support and improvement on top if you want them.

Judge it on payback, not rate. A senior AI consultant at $350 an hour who scopes the right project and ships it in a month is far cheaper than a $150 an hour generalist who spends six months learning on your budget. If you want a sense of what your first project would cost, book a call and we will talk specifics.

Red Flags to Walk Away From

A few signals that tell us to keep looking, offered so you can spot them too. A consultant who promises a fixed outcome before discovery has either done this exact thing before or is guessing. One who cannot explain their approach in plain language is hiding behind jargon. One whose proposal is all strategy and no working software is selling slides. And one who never mentions data quality has not thought about where your project will actually break, because that is where it usually does.

The good news is that the same honesty that makes a consultant useful also makes them easy to spot. They will tell you what could go wrong before you ask.

Frequently Asked Questions

What does an AI consultant do day to day?

Day to day, an AI consultant moves between discovery conversations, solution design, hands-on building, and measurement. The visible output is working software, but most of the hours go into deciding what to build and wiring it into your existing systems. The model itself is usually the smallest part of the job.

What is the difference between an AI consultant and a data scientist?

A data scientist typically builds and analyses models from data. An AI consultant is broader: they decide which business problems are worth solving, design the solution, often build it using existing foundation models, and connect it to your operation. Consulting leans toward business outcomes and delivery; data science leans toward modelling.

How much does an AI consultant cost in Australia?

Independent AI consultants and small consultancies usually charge $200 to $400 AUD an hour. A discovery engagement runs about $8,000 to $25,000 AUD, and a first production build commonly lands between $25,000 and $80,000 AUD. Running costs for most mid-volume systems are $50 to $500 AUD a month.

Do I need an AI consultant or an in-house hire?

A consultant suits your first few projects, when you need senior judgment and speed without a long-term commitment. An in-house AI engineer makes sense once you have a steady pipeline of AI work to build and maintain. Many businesses start with a consultant and hire internally once the volume justifies it.

What should I ask an AI consultant before hiring them?

Ask to see a system that has run in production for six months, ask what they have told a client not to build, and ask how they handle the cases the model gets wrong. The answers should be specific and should include honest limits. Confident vagueness is the warning sign.

Can an AI consultant help with strategy, or only building?

Both, but be wary of strategy with no path to a working system. The useful strategy work is choosing which problems to solve and in what order. A vision deck with no buildable output attached is the kind of AI consultancy deliverable that looks impressive and changes nothing.

How long does an AI consulting project take?

Discovery is usually two to four weeks. A focused first build typically takes four to twelve weeks depending on complexity and the number of systems involved. Anyone promising a production AI system in a few days is showing you a demo, not a deployment.

The short version: an AI consultant earns their fee by choosing the right problem and shipping something that survives in production, not by knowing the most about models. If you want that kind of help, the kind that starts with your business and not with the technology, get in touch. We will tell you early if AI is not the answer, which is more than the deck-sellers will.