A services business sells whatever it can package fastest when a new technology arrives, and I say that as someone who ran one for long enough to have done it myself. So when OpenAI announced its “Frontier Alliances” in February — multi-year partnerships with McKinsey, BCG, Accenture and Capgemini to deploy its agent platform inside large companies — my reaction was less about the names than about the admission underneath them. OpenAI wrote that the limiting factor in getting value from AI “isn’t model intelligence” but how the work around it gets organised, and BCG’s chief executive went further in the same announcement: “AI alone does not drive transformation.”

When the company selling the model says the model isn’t enough, and the four firms who’ll charge you to make up the difference sign up the same day, you’ve been handed the most useful sentence in the whole announcement, because it tells you what to look for when somebody offers to help.

Training dressed as transformation

Most of what’s sold as AI consulting today is training. Someone comes in, runs a workshop on writing better prompts, sets up a few licences, produces a roadmap with the word “agentic” in the title, and leaves. People do need to learn the tools, and a good internal champion with a couple of curious weeks will teach them; calling that transformation is what lets it carry a transformation price tag. Most of us buy the training first because it’s the visible, purchasable thing, whereas the decisions underneath the work are harder to look at and nobody sells a workshop on them. The business that comes out the other side does what it did before, with the same people making the same decisions on the same evidence, slightly faster. I wrote earlier this year about the efficiency trap: AI pointed at an existing process mostly produces the old output at greater speed, and speeding up the monthly report says nothing about whether anyone reads it.

Transformation, if the word is to mean anything, changes how decisions get made. Take quoting. The training version has your estimator using a chatbot to write quotes faster, and quotes go out the same afternoon. The redesign version starts by asking who decides the price, on what evidence, and how long the evidence takes to arrive, then finds that the discount your estimator applies by habit was set in 2019 and that the model can pull three years of won-and-lost quotes in a morning to show which jobs you should stop bidding on altogether. The first is a tool being adopted; the second changes what the business decides and on what evidence. Almost nobody selling AI consulting is staffed or priced to deliver the second kind of work, because it requires someone who understands your operation well enough to say which decisions matter, and the nerve to say which of your processes should stop. Whether the four alliance firms will deliver that at scale is a fair question. They employ plenty of people who could, and I don’t know whether those people or the licence targets will win.

The COBOL irony

What makes the Frontier Alliances so revealing is the timing. On 23 February, the same day OpenAI made its announcement, Anthropic published a playbook for using Claude Code to analyse and migrate COBOL systems, the decades-old banking and government code that has kept consultants in billable work partly because so few people can still read it. Anthropic’s pitch was that the work “used to require armies of consultants”, and investors read it that way: IBM’s shares fell around 13% that day, its worst session since 2000, and Accenture and Cognizant fell with it. Within the same news cycle, Accenture was announced as a partner selling OpenAI’s agents.

That’s a rational move, and it’s the move the same firms made through Y2K, ERP, cloud and “digital transformation”: when the technology threatens the last engagement, sell the next one. The practical lesson for a buyer is simpler than the irony. Advisers sell what their commercial model rewards, so before you weigh their advice, look at the model.

How to tell who’s selling what

If you’re weighing up an AI adviser, a handful of questions separate the real thing from the rebrand. Start with which decision in your business they’d change first, and what evidence they’d need to change it; a genuine transformation consultant will have an answer within the first meeting, and it will involve your data rather than their slides. Then find out what they’d tell you to stop doing, because the ones who only ever add are selling you more of what you already do. Who is the deliverable for? A roadmap written for the board reads very differently from a process written for the people who run it.

And look at how they’re paid, because the payment structure shapes the recommendation before anyone has looked at your business. A firm paid by seat, by licence or by headcount trained has a reason to recommend more seats, and the four alliance firms are each building teams certified on OpenAI’s technology, which makes it worth asking how often their answer will be somebody else’s platform. The red flags follow from that: a generic workshop, a recommendation that arrives with a vendor attached, a roadmap with nobody’s name on it, and no measure of success beyond how many people logged in.

I’ve written a longer five-point test for separating real transformation from a rebrand with a chatbot, and it applies to advisers as well as vendors: the more work the word “AI” is doing in the pitch, the less the AI will be doing for you.

When to hire, and when not to

The awkward part, for anyone who sells advice for a living, is that most small and mid-sized businesses need permission to experiment far more than they need a consultant. That means someone senior saying, out loud, that it’s fine to pick one recurring decision — how quotes get priced, how leads get qualified, which customers get chased — and let a curious person spend two weeks rebuilding it with the tools now sitting on every desk. The cost is one person’s time, and what comes back will teach you more about where AI fits in your business than any workshop.

The rule I’d use from there is short. Hire an integrator when the redesign needs wiring into systems your own people can’t touch. Hire a consultant when the redesign runs into a decision nobody inside the business can make honestly, whether that’s two departments who both claim to own it, a process the founder built and won’t let go of, or the moment it becomes clear that a good person’s work is now redundant. An outsider with the standing to say what the room can’t is the engagement worth paying for. Everything else is training, and you can buy training a great deal cheaper than you can buy transformation.