AI can do a SWOT in 30 seconds. It still cannot tell you why your MES does not talk to SAP.

Last year I sat in a meeting where a management consultant presented an AI-generated competitive analysis. It was correct. Every data point checked out. The framing was sharp. The recommendation was logical.

It was also useless.

The recommendation assumed our production planning system communicated with our ERP. It did not. The recommendation assumed the logistics manager who controlled 60% of the operational budget would support the change. She was leaving in March. The recommendation assumed the factory floor team had capacity for a new process. They were running double shifts to cover two sick leaves.

None of this was in the data. All of it determined whether the recommendation would work.

This is what I think about when people say AI is eating consulting. They are right, but only about a specific kind of consulting.

Execution has become cheap. A SWOT analysis, a market scan, a first draft of a strategy deck. AI produces these in minutes. What took a junior analyst a week now takes a prompt and thirty seconds. The information is the same. The speed is different. The cost approaches zero.

That changes the economics of consulting. But it changes the economics of a specific layer: the finding-and-formatting layer. Search, structure, first draft. McKinsey’s own research (before AI) showed knowledge workers spent roughly 20% of their week just looking for information. AI compresses that to almost nothing.

What remains is framing and sequencing. Framing: what is the actual question? Sequencing: what do we do first, second, third? These are the two things AI cannot do without the context that only exists inside the organization.

I have worked in three types of organizations. A startup with ten employees where every decision was a week away from cash running out. Public sector where a procurement process took nine months and a project group had twenty people with different mandates. A corporate venture inside Volvo Group where every initiative required alignment across five departments and two countries.

In all three, the thing that determined project success was not the quality of the analysis. It was whether someone understood the actual constraints. The production manager who was leaving. The IT system that did not integrate. The union agreement that prevented the proposed shift change. The procurement rule that made the best solution impossible to buy.

AI does not know these things. AI cannot know these things. They exist in conversations, in relationships, in the accumulated knowledge of people who have worked in the building for fifteen years.

My assessment is that consulting splits into two categories now.

The first category is commoditized. Data gathering, benchmarking, slide production, first-draft strategy. AI does this better, faster and cheaper than junior consultants. Any firm still charging 2 000 euros per day for this work is selling something that costs nearly nothing to produce.

The second category is the opposite of commoditized. Understanding the actual situation. Knowing which stakeholder will block the project and why. Seeing that the obvious technical solution fails because the incentive structure points in the wrong direction. Sequencing the implementation so that the first win builds trust for the second.

This category requires field experience. Not the kind you get from case studies. The kind you get from having stood at the loading dock at five in the morning, from having built a system that a thousand truck drivers actually used, from having sat in a boardroom explaining to a CEO why the expensive platform nobody uses is not a technology problem.

AI makes the first category worthless. AI makes the second category more valuable.

If you lead an organization with operational complexity and someone offers you “AI-powered consulting”, ask one question: has the person giving the advice ever done the job they are advising you on?

If the answer is no, you can get the same advice from ChatGPT. For free.