How to Choose an AI Consultant: 7 Questions to Ask

Three proposals are on your desk. One is sixty pages long. One quotes a day rate and no end date. One promises results in a number you cannot check. All three come from someone who calls themselves an AI consultant. This guide gives you seven questions to ask before you sign, with a good answer and a warning sign for each.

It is written for the buyer: the CEO who signs, the CFO who checks the number, and the IT and Legal leads who have to approve. If you are here because you want to become an AI consultant, the last question in the FAQ is for you.

Key takeaways

An AI consultant helps a company decide where AI is worth using, and gets it working. Many only do the first half.
Ask what will be working on your own systems at the end. A report is not an answer.
Ask how each euro figure is calculated. If you cannot check the arithmetic, it is a guess.
Prefer a fixed price and a first step you can stop after.

What does an AI consultant actually do?

An AI consultant is an outside person or firm that helps a company decide where to use AI and then put it to work. In practice that covers three different jobs:

The three jobs of an AI consultant: advise, build and train

Advise. Find the work AI should take over, and say what it is worth.
Build. Make it run in the systems your teams already use: the CRM, email, the document store.
Train. Teach the people who will use it, on their own tasks.

Many consultants do only the first job. That is where the trouble starts, because advice alone rarely changes how work gets done.

Many companies get little from AI. In a report published in September 2025, based on a survey of 1,250 senior executives, BCG found that 60% of companies get “hardly any material value” from AI despite substantial investment.^[1] BCG sells consulting itself and the answers are self-reported, so treat the figure as an indication. Still, it matches what many leaders say in private: we bought it, and nothing changed.

The reason companies look outside is also clear. Among EU companies with ten or more employees that considered AI and did not adopt it, the most common reason was “the lack of relevant expertise”, named by 70.89%.^[2] Only 30.36% of medium-sized EU enterprises used AI at all in 2025.^[2]

Does outside help work better than building alone? One study suggests it can. MIT’s Project NANDA reports that, in its sample, external partnerships with customised tools that learn reached deployment about 67% of the time, against about 33% for tools built in-house.^[3] The authors add their own caution. The figures come from interviews with 52 organisations and the findings are preliminary. The difference “may reflect organizational capabilities rather than implementation approach alone”, and the link “does not necessarily prove causation”. So outside help can work. The study does not prove that it will.

The 7 questions to ask an AI consultant

Before PromptPartner I spent more than twenty years in software go-to-market, and helped grow WebPros (cPanel, Plesk, WHMCS) to a €1.5 billion exit to CVC. These are the questions I would ask as the buyer.

The seven questions, with the answer to look for and the warning sign.
Question	A good answer	A warning sign
1. What will be working on our systems at the end?	A named, working prototype on your real systems	A report, a roadmap or a “maturity score”
2. How do you put a number on a use case?	The arithmetic is visible and every assumption is named	A benchmark percentage or a promised return
3. Who does the work, and where?	Named people, working inside your environment	Seniors sell, juniors deliver, on the consultant’s platform
4. What happens to what you build if we stop?	It lives in your systems from day one. You keep it	You lose access when the contract ends
5. How do you handle our data and our rules?	Legal and IT set the rules first. Read-only access to start	“We will sort out security later”
6. How do our people learn to use it?	Regular training on their own tasks, while the build runs	One workshop at the end
7. How is it priced, and can we stop?	A fixed price, a fixed scope, a first step you can stop after	Open-ended day rates, or a promise of results

1. What will be working on our systems at the end?

This is the most useful question, so ask it first. A good consultant names one thing that will run on your real systems by a date. For example: “By week four, your account managers get a prepared brief before every customer call, built from your CRM and your email.”

The warning sign is a deliverable made of documents only. I have been that buyer. At a former company, consultancies delivered 750 slides, and then we used nothing from them.

2. How do you put a number on a use case?

Every consultant will show you a number. Ask where it comes from. A good answer shows the arithmetic: how many people, how many hours, which cost, which assumption. Your CFO should be able to argue with each line.

The warning sign is a number that arrives before anyone has looked at your business. Be careful with anyone who promises a return before they have seen your numbers. They are applying their figure to your company.

3. Who does the work, and where?

Ask for names. Ask whether the people in the sales meeting are the people who will do the work. Then ask where the work happens. A good answer: inside your own environment, with your people, from the first day.

The warning sign: senior people sell, junior people deliver, and the work sits on the consultant’s own platform.

4. What happens to what you build if we stop?

You should keep everything. If the work lives in your own systems from day one, there is nothing to hand over at the end and nothing to lose if you stop. If it lives on the consultant’s platform, you are renting it.

5. How do you handle our data and our rules?

An AI agent with access to your systems can read your mailbox and your CRM. That is why the rules come first. A good consultant sits down with Legal, IT and Security before anything goes live, and starts with read-only access. Rules that are set once make every later use case faster to approve. That is how we build it.

The warning sign is the sentence “we will sort out security later”.

6. How do our people learn to use it?

BCG has a rule of thumb for where the value of AI comes from: about 10% from the AI application, about 20% from data and technology, and about 70% from how people and processes change.^[4] It is one consulting firm’s framework, not a measurement. But it explains why a tool that nobody was trained on changes nothing.

A good answer is regular training on real tasks, while the build is running. A warning sign is a single workshop at the end.

7. How is it priced, and can we stop?

Look for a fixed price, a fixed scope, and a first step that is its own contract. Then you can stop after it without owing anything further.

There are two warning signs. One is an open-ended day rate with no end date. The other is a promised result. We won’t guarantee a result, because half of the result is what your organisation does with it, and we don’t control that half. The same is true for every consultant.

Consultant, agency or in-house team: which fits?

An AI consultant is one of three ways to get the work done. Each fits a different situation.

Three ways to get AI work done: a strategy consultancy, an agency, or your own team, with when each fits

A strategy consultancy fits when the board needs an outside view on a large decision. The risk is that you get analysis and no working system.
An agency fits when you already know exactly what you want built. The risk is that often nobody measured whether it was the right thing, and often nobody makes sure your people use it.
Your own team fits when you have engineers with time and experience in AI. The risk is the common one: licences for everyone, then little use.

PromptPartner works in a fourth way: a forward-deployed engineer, which means an engineer who works inside your systems, and a business operator, working inside your team. We measure first, build in your systems, and train the people who use it. Here is how that works, step by step.

What does an AI consultant cost?

We found no reliable public survey of what AI consultants charge, so be careful with any “average day rate” you read. What you can compare is the pricing model. A study of nearly 1,000 general consultants found four common ones: by project, by the hour or day, by monthly retainer, and by value.^[5] That study covers consultants in general, not AI, and comes from a firm that coaches consultants.

Fixed price per project. You know the cost before you start. If the work takes longer, the consultant pays for it.
Day rate. Flexible, and the total can grow without limit. Ask for a maximum.
Monthly retainer. Useful for ongoing work after something is live.
Value-based. Tied to an outcome. Check carefully who measures the outcome.

Our own first step has a fixed price. The four-week AI audit is priced by company size and by whether it covers one department or several. It is its own contract, and you can stop after it.

What good looks like: one example

At a European software company above €100 million in annual recurring revenue, the first step was an audit. Eleven discovery sessions. Thirty-six opportunities scored. Then the build: five working tools, shipped into the systems the team already used.

One of those tools turns a week of work on sales material into minutes. Read the case study.

That is one company. Yours will look different, which is the point of measuring first.

Frequently asked questions

What does an AI consultant actually do?

Three jobs: advise on where AI is worth using, build it into the systems your teams already use, and train the people who will use it. Many consultants do only the first. Ask which of the three is in the contract.

How much is an AI consultant?

It depends on the pricing model: a fixed price per project, a day rate, a monthly retainer or a fee tied to value. We found no reliable public survey of AI consulting rates. Our own four-week AI audit has a fixed price, set by company size and by whether it covers one department or several.

Do we need an AI consultant, or can we do it ourselves?

You can do it yourselves if you have engineers with time and experience in AI, and someone who owns the business case. If either is missing, consider outside help. Either way, measure first.

What is the difference between an AI consultant and an AI agency?

A consultant usually starts with the question of what is worth doing. An agency usually starts with a brief and builds it. Consider a partner who does both and stays until your people use the result.

How long does it take to see something working?

Ask every consultant for a date. In our method, the audit takes four weeks and ends with a working prototype on your real systems. What happens after that is your decision.

How do I become an AI consultant?

This guide is for buyers, so we keep it short: build things that work in real companies, and learn the business side as well as the technology. If that sounds like you, look at our open roles on the careers page.

Looking for a role and not a consultant? See our open positions.