The buyer's guide
How to choose an AI consultancy
The market sells three kinds of AI help, and they fail in three different ways. Five questions sort the field in one meeting.
Ask five questions before you sign with any AI consultancy. Do they start with your business model or with the tools? Who actually does the work, seniors or juniors? Do you buy outcomes at a fixed price or hours on a meter? Who owns what gets built? And can you stop after the first step without penalty? The answers separate a partner from a vendor faster than any reference call.
The market
The three things the market sells
The market offers three things: technology vendors selling platform lock-in dressed as transformation, big consultancies selling reports dressed as strategy at three to five times the cost, and freelance AI specialists who understand the tools but not the business model implications. Each fails differently. The vendor optimizes for their platform, the big firm for billable hours, the freelancer for the tech. None of them starts where AI strategy is decided: in the management team.
The three failure modes look different in month one and identical in month twelve: money spent, tools deployed, competitive position unchanged. The vendor's transformation ends where the platform's feature list ends. The big firm's report is often excellent and sits unread, because a recommendation without capability transfer changes nothing on a random Tuesday. The freelancer builds something clever that dies the day they leave, because nobody in the building can hold it.
Who does what
AI consultant, AI agency, data scientist: who does what
An AI agency builds you a system. A data scientist builds you a model. An AI strategy consultant works one level up: what fundamentally changes about your business as AI enters your industry, and what capability your team needs to act on it. Buy the first two when you know exactly what to build. Buy the third when the real question is what the technology changes about how you compete.
The confusion is understandable, because all three put AI on the slide. The test is what they ask about in the first meeting. An agency asks about your systems. A data scientist asks about your data. A strategy consultant asks what your customers pay you for, and which parts of that AI just made cheap, fast or free. If nobody asks that question, you are buying implementation, whatever the proposal says.
Five questions
The five questions
Every criterion below is checkable in the first meeting. A good firm answers all five without flinching. A weak one negotiates.
Do they start with your business or with their tools?
A good firm opens with what changes at the core of your industry: the billable hour, the data moat, the commercial layer. A weak one opens with a platform demo. The first meeting should feel like strategy, not procurement.
Who is actually in the room: partners or juniors?
The pyramid existed to train analysts, and AI just automated most of what those analysts did. If the proposal still carries a delivery team of eight, you are paying for a structure the technology made obsolete. Ask who is in the room in week three, not at the pitch.
Do you buy outcomes at a fixed price, or hours on a meter?
Hourly pricing makes the firm's revenue grow with your problem. A fixed price scoped to outcomes makes it grow with your result. The firms that run on AI themselves can afford to fix the price, because the analysis no longer costs them a pyramid.
Who owns what gets built: you, or the vendor?
Everything built during the engagement, agents, workflows, training material, should belong to you from day one, with no licence in between. If the value walks out the door with the vendor, you rented a capability instead of building one.
Can you stop after step one and keep everything?
No lock-in is a confidence signal. A firm that earns every next step will let the first one stand alone. A firm that needs a twelve-month contract to make the economics work is telling you something about the economics.
What you buy
What you are actually buying
Strip away the market noise and buyers choose advisors on trust. One useful model puts it as an equation: credibility plus reliability plus intimacy, divided by self-orientation. AI just commoditized the first two variables. Every firm now has instant access to every framework ever published, and analysis that does not make arithmetic errors. What remains scarce is the human part: whether the advisor will tell you an uncomfortable truth, whether they are in the room when the hard decision gets made, and whose success they are actually optimizing for.
That has a practical consequence for the analysis itself: it is heading toward commodity pricing. Genius-level research and synthesis are becoming as cheap as cloud computing. Pay a premium for judgment, diagnosis and someone willing to put their name on the recommendation, because that is increasingly the only part that is scarce. The industry view behind this is in Your Consulting Firm Just Hired a Country of Geniuses.
Red flags
The red flags
The red flags are consistent: a pilot proposal with no management involvement, pricing that only makes sense if the engagement never ends, a licence sitting between you and the thing you paid to build, and a case library that is all technology and no P&L. Any one of them is worth a hard question. Two of them are worth another firm.
One more, specific to this market: a firm that sells AI transformation but visibly does not run on AI itself. Ask how their own delivery works. If the answer is a methodology slide rather than a working system, the expertise is theoretical, and you will be the experiment.
The price
What it should cost
An AI consultancy should be able to tell you the price before the work starts. Hourly and day rates put the risk on you and reward slow delivery; a fixed fee scoped to outcomes puts the risk on the firm. Whatever the model, the total should be justified by senior work reaching your P&L, not by the size of the team they bring.
The market context matters here. AI compressed the cost of the analysis layer, which is most of what a traditional engagement billed for. A firm built on the new cost structure delivers in weeks what used to take quarters, at a fraction of the old price. If a quote still assumes a pyramid of billed hours, the price is carrying their structure, not your outcome. How GRAIL prices, and what you own at the end, is on what it costs.
The real test
The one thing all five questions test
Read the five questions together and they are really one question: is this firm built on the old model or the new one? Management consulting was a business built on the assumption that transformation takes large teams of expensive people. AI broke that assumption. A senior advisor running on AI infrastructure now does what used to take a team of eight to twelve, which means the firm you should want is not a smaller version of the pyramid. It is what someone would build if they started a McKinsey from scratch today: a few senior operators on top of an AI delivery engine, pricing outcomes, handing you everything they build.
This is why the incumbents cannot simply match the answers. Their revenue depends on billing the junior hours AI just replaced, so the honest firm and the pyramid firm give opposite answers to all five questions, and they cannot both be telling the truth about the same cost structure. The test also has a clean tell: a firm that has genuinely rebuilt around AI can run its own delivery on it, and can prove speed, cost and quality on itself before it promises them to you. Ask to see that. The full comparison, including when a Big-4 firm is still the right choice, is on GRAIL vs the alternatives.
The right question is not which firm is best. It is which one you can measure.
A firm confident in its answers will put them in writing before the work starts.
Questions buyers ask
How do I choose an AI consultancy?
Ask five questions before you sign with any AI consultancy. Do they start with your business model or with the tools? Who actually does the work, seniors or juniors? Do you buy outcomes at a fixed price or hours on a meter? Who owns what gets built? And can you stop after the first step without penalty? The answers separate a partner from a vendor faster than any reference call.
What kinds of AI consultants are there?
The market offers three things: technology vendors selling platform lock-in dressed as transformation, big consultancies selling reports dressed as strategy at three to five times the cost, and freelance AI specialists who understand the tools but not the business model implications. Each fails differently. The vendor optimizes for their platform, the big firm for billable hours, the freelancer for the tech. None of them starts where AI strategy is decided: in the management team.
What is the difference between an AI consultant, an AI agency and a data scientist?
An AI agency builds you a system. A data scientist builds you a model. An AI strategy consultant works one level up: what fundamentally changes about your business as AI enters your industry, and what capability your team needs to act on it. Buy the first two when you know exactly what to build. Buy the third when the real question is what the technology changes about how you compete.
What are the red flags when hiring an AI consultancy?
The red flags are consistent: a pilot proposal with no management involvement, pricing that only makes sense if the engagement never ends, a licence sitting between you and the thing you paid to build, and a case library that is all technology and no P&L. Any one of them is worth a hard question. Two of them are worth another firm.
What should an AI consultancy cost?
An AI consultancy should be able to tell you the price before the work starts. Hourly and day rates put the risk on you and reward slow delivery; a fixed fee scoped to outcomes puts the risk on the firm. Whatever the model, the total should be justified by senior work reaching your P&L, not by the size of the team they bring.