An article in last week’s FT caught our eye. Robert Armstrong posed an uncomfortable question — well, certainly uncomfortable for those of us with “consultant” on our CV … who needs consultants in the age of AI?
It is a question that the consulting industry has rather more reason to take seriously than most professions. Strategy consultants have traditionally charged handsomely for activities that generative AI is becoming increasingly capable of performing: researching markets, synthesizing evidence, identifying patterns, structuring problems, generating strategic options and turning the resulting argument into a polished presentation. Armstrong captures the threat rather neatly when he suggests that if something can go into a slide deck, it is probably becoming a commodity.
There is clearly something in this. AI does not mean that every consultant is about to disappear, but it does challenge an economic model built partly around expensive human production of analysis. If a client can generate a reasonable market overview, competitive assessment or first-pass strategic argument internally and in minutes, it becomes harder to justify a team of consultants spending weeks producing something recognizably similar. This is not simply about making PowerPoint decks faster; it potentially removes a substantial part of the intellectual production process around which consulting firms have historically organized themselves.
Armstrong’s answer is that the most valuable part of consulting has, for some time, lain elsewhere. Working out what an organization ought to do may be difficult, but getting an organization to do it is harder. Senior executives have different incentives, beliefs and interpretations of what is happening. Knowledge is dispersed around the organization, much of it tacit, some of it political and often not widely socialized. The consultant’s role therefore involves building trust, surfacing what is really going on, reconciling competing perspectives, creating sufficient consensus and helping an organization move from recommendation to action. Its about framing choices … and strategy is about making choices. As one of Armstrong’s interviewees puts it, consulting is a “contact sport”.
His description culminates in a memorable metaphor. Analytical and technical work once provided the bulk of the meal, while the bespoke human work of persuasion, alignment and change was the special sauce. AI is now eating much of the meal, leaving, as Armstrong puts it, “just sauce”.
But perhaps we have mistaken volume for value. The fact that research, analysis and slide production consumed much of a consulting team’s time does not necessarily mean that this was where most of the value was created. AI may be doing something more interesting than simply stripping labor out of consulting: it may be forcing the industry to confront what clients were really paying for in the first place.
Think about the moments of truth in a consulting engagement. Is the question the client has asked actually the question that needs answering? Which of several contradictory pieces of evidence should carry most weight? Is the emerging pattern meaningful? Why does something in the data not quite fit? How much confidence should the team place in an analysis built on incomplete information? Eventually, somebody has to decide whether the accumulating evidence is sufficiently compelling to recommend that a client acts upon it.
These are questions of judgement, and they sit awkwardly between Armstrong’s two categories of analysis and implementation. Judgement runs through the consulting process from beginning to end. It shapes how the problem is framed, determines which evidence carries weight and which is discounted, influences when an assumption is challenged and ultimately governs the transition from “this looks plausible” to “we are prepared to put our professional weight behind this recommendation”.
This is where AI introduces a more complicated problem than the simple commoditization of analysis. In the first of a series of articles we recently wrote for Consulting Magazine, we called it The Judgement Gap. The space between AI producing a plausible answer and the human doing enough work to know whether that answer deserves confidence.
Generative AI is remarkably good at producing the appearance of completed thinking. Imagine a situation familiar to almost anyone who has worked in consulting. The client problem is messy, the evidence incomplete, the team tired and a meeting is looming. Someone asks AI to summarize the data, sharpen the issue analysis and help structure the argument. Within seconds the team has something coherent, plausible and impressively polished. It has saved time and may well have improved the work, but has AI improved the thinking, or merely improved the appearance of thinking?
The obvious concern is hallucination, but the greater risk may be the answer that is broadly sensible and only subtly mistaken. One that frames the problem too narrowly, over-weights one piece of evidence, suppresses uncertainty or proposes something perfectly rational in the abstract but ill-suited to the reality of this particular client. These answers are dangerous precisely because they do not look dangerous. Fluency and structure create their own sense of authority, and under the familiar pressures of consulting work the plausible answer can travel surprisingly far before somebody questions it.
AI therefore changes the economics of intelligence in two directions at once. It reduces the cost of producing analysis while increasing the importance of evaluating what has been produced. When analysis was expensive and relatively scarce, producing it carried considerable value. As plausible analysis becomes abundant, the scarce capability increasingly becomes discrimination … knowing what deserves attention, what requires challenge, what fits the context and what should ultimately be acted upon.
That takes us slightly beyond Armstrong’s conclusion that strategic consultants increasingly sell a process. They certainly do, but perhaps a better description is that they sell judgement exercised through a process. The human advantage does not begin once AI has produced the strategy and somebody has to persuade the organization to implement it. It starts much earlier, in determining what problem is worth solving, how conflicting signals should be interpreted and whether the answer emerging from the machine survives contact with reality.
There is another observation in Armstrong’s article that may prove still more consequential. If AI removes much of the “bulk analytic work” traditionally undertaken by junior consultants, how will firms train the next generation? The traditional consulting pyramid was not simply an economic model; it was also an apprenticeship system. Junior consultants learned by conducting interviews, gathering evidence, building analyses, defending conclusions and discovering that arguments which, on the face of it, looked convincing were often rather less convincing once challenged by an experienced colleague.
Much of that work was laborious, and AI should remove some of it. But embedded within it were thousands of small experiences through which consultants gradually developed professional judgement. What looked like inefficient analytical labor was also the environment in which expertise was being formed. The risk is therefore not simply that AI does more of the junior consultant’s work, but that firms inadvertently automate parts of the developmental pathway through which junior consultants became senior consultants capable of judging that work.
This is why the distinction between cognitive offloading and cognitive surrender matters. Consultants have always used tools to extend their capabilities, from frameworks and spreadsheets to databases and search engines. AI belongs in that tradition, but it differs in one important respect: rather than simply storing information or performing a calculation, it can return something that resembles completed thought. The temptation is not merely to offload part of the cognitive process, but to accept the finished-looking result without sufficiently engaging with the reasoning beneath it.
The question facing consulting firms is therefore no longer whether their people should use AI. That argument is largely over. The harder question is what kind of consultants widespread AI use will produce. It is possible to imagine firms becoming dramatically more productive while simultaneously weakening some of the human capabilities on which their future differentiation depends. Equally, AI can widen the range of possibilities consultants consider, challenge their initial framing, stress-test emerging conclusions and create more space for the parts of consulting that require distinctly human judgement.
Which version emerges will depend on how deliberately firms manage the transition. They will need to become much more intentional about cultivating judgement, critical thinking, curiosity, communication, empathy and leadership, together with the metacognitive ability to recognize when their own thinking has become overly dependent on the machine. These capabilities are still routinely described as soft skills, but that description makes progressively less sense. When competent cognitive output becomes abundant, the ability to interrogate it, contextualize it, challenge it and turn it into action becomes a much harder source of competitive advantage.
So perhaps Armstrong is right that AI is removing much of the traditional consulting meal. Where we would differ is in assuming that what remains is merely the sauce. AI may instead be unbundling consulting in a way that finally reveals which parts of the work were genuinely scarce. The research mattered, the analysis mattered and the slides sometimes mattered. But underneath all of them sat something harder to see and harder to price: the accumulated human judgement required to decide what mattered, what could be trusted and what should be done.