AI and the capability question in consulting
AI has impacted the consulting market in a multitude of ways. A full assessment has to consider strategic positioning, the economics of leverage, the technology and product roadmap, governance and risk, competitive dynamics, and the changing role of the partner.
This piece takes one narrower question. If AI changes the work through which consultants have historically developed expertise and judgment, how should firms think about developing the next generation?
The consulting delivery model has long done two jobs at once. It has generated revenue while developing professionals. As AI changes the first, it also changes the second.
Firms that use AI to change delivery without deliberately redesigning apprenticeship will accumulate capability debt. The immediate gain may be speed and margin. The later cost will be a thinner pipeline of people able to set direction, challenge weak reasoning and take responsibility for AI-augmented work.
In the MCA’s 2026 survey of more than 1,000 consultants, 77% said their firm had either integrated AI into its systems or enabled employees to use AI models. Seventy-eight per cent identified digital technology and AI as drivers of growth.1 These figures do not describe every firm or every type of work. They do, however, make the capability question a current one rather than a distant concern.
The effects will be uneven across firms, practices and client situations, depending on the nature of the client need and the work required to address it. Consulting fees buy different things, including expertise, capacity, legitimacy, accountability and signalling. AI will affect those sources of value differently.
A delivery model that also develops people
The traditional pyramid was not simply an economic model. Partners won work and managed senior client relationships. Directors, managers and juniors did much of the execution. The same staffing structure that generated revenue also provided the training ground for the next generation. Juniors learned by doing work that, over time and under feedback, gave them the capability to become managers and eventually partners.
The base of the pyramid existed not just to deliver work but to manufacture experienced professionals.
AI changes that arrangement. Tasks that previously justified substantial billable effort can now be completed much faster. Scale can increasingly come from systems, workflows, models and data that allow a smaller group to produce more. But where AI compresses work at the base of the pyramid, it can also reduce the experiences through which professional capability was built.
Some of the work AI removes was never especially developmental. Removing it may improve apprenticeship if the capacity released is reinvested in more challenging work, earlier client exposure and better review. The risk is not automation itself. It is allowing automation to remove developmental work without replacing it.
Client expectations sharpen the commercial tension. Source Global Research found that 82% of UK and US clients expect AI to speed consulting projects. Yet clients were almost evenly divided on whether it would make the resulting work better or more generic.2 Faster output is not, in itself, a durable value proposition. The question is whether teams can use the time and capacity AI releases to produce work that is more useful, more rigorous and better attuned to the client situation.
That is the capability question.
of more than 1,000 consultants say their firm has integrated AI or enabled employees to use AI models.
of UK and US clients expect AI to speed consulting projects, while splitting evenly on whether the work gets better or more generic.
hires recorded across 80-plus member firms in 2025, with graduate recruitment up 10% and apprentice hiring up 31%.
What apprenticeship produced, and what is at risk
A natural response is to ask which capabilities will remain permanently human. It is an understandable question, but not a particularly reliable one. The boundary of what AI can do has moved repeatedly, and it will continue to do so.
The more useful question is developmental rather than ontological. The purpose of apprenticeship is not to preserve work for its own sake. It is to develop people capable of supervising, integrating and taking responsibility for work to a professional standard, even when that work is increasingly AI-augmented.
Consulting apprenticeship developed a specific set of capabilities through volume, repetition and feedback. Juniors learned to structure messy problems from scratch. They built hypothesis-driven analysis, drafted and modelled, recognised patterns across multiple engagements, read client rooms, and formed judgments about what mattered in a sea of evidence.
Three capabilities are particularly vulnerable when AI becomes the default first action.
Problem structuring
The previous loop required a junior consultant to face a messy problem with no perfect template, use previous work as inspiration where helpful, stay with it long enough to find a structure, propose a starting frame and revise it in response to challenge. When AI provides a credible first structure on request, that loop may never begin. Juniors learn to evaluate and edit structures rather than create them. The former is valuable, but it depends on some ability to do the latter.
View-taking
Consulting work often requires people to form a view before they have complete information, defend it under questioning and revise it when they are wrong. AI can present several balanced perspectives quickly and fluently. That may make it easier to sound informed about every angle while never having committed to a position that has to withstand challenge.
The instinct for what is missing
AI output can appear complete, coherent and persuasive even when the underlying reasoning is shallow, generic or incorrect. Juniors who only ever review polished output may not develop the same instinct as people who have built first cuts themselves and learned what their own drafts were missing. As they become more senior, that instinct becomes part of the ability to interrogate AI output at a professional standard.
There is a related issue around client judgment. The previous apprenticeship was not only about individual task repetition. It was also built through review cycles with managers and partners. This involved anticipating client perspectives and challenges, deciding what to say and what to leave unsaid, and seeing how analysis becomes a position that can be used in a live situation. Leaner teams and faster production can reduce those developmental moments as well.
An experiment involving 758 BCG consultants illustrates how sharply performance can vary between tasks. On tasks within GPT-4’s capabilities, consultants using it completed work more than 25% faster, with quality scores more than 40% higher. On a task outside those capabilities, AI use reduced the likelihood of a correct answer. Conducted in 2023 using GPT-4, this was a bounded experiment rather than a test of the entire partner role. The frontier has moved since. Its enduring point is not where it then sat, but that AI performance can differ sharply from task to task.3
What becomes more important
AI does not make these capabilities newly valuable. It changes their relative weight.
The work becomes less about producing a plausible first answer and more about deciding which question matters, what can be provisionally assumed, what must be tested, and which trade-off the team should recommend. This is synthesis under underdetermined evidence. It is making responsible progress when there is no complete dataset or mechanically correct answer.
Contextual judgment matters for the same reason. It is the ability to know when an answer that is technically correct is wrong for this client, at this moment, in this situation.
Neither capability should be presented as permanently beyond AI. The immediate point is that firms need people who can exercise them now, and development takes time. The bottleneck during the transition may be less about what the technology can eventually do than whether enough people are being developed to supervise and take responsibility for the work it is already changing.
Development will need to be designed
The practical response is straightforward to describe and difficult to maintain under commercial pressure.
Juniors need to do meaningful cognitive work before AI enters the process. An analyst might build the first issue tree, storyline or hypothesis set before using AI to stress-test it, identify missing branches or generate alternatives. In that role, AI acts as critic and amplifier rather than substitute.
Tasks should not all be treated the same way. Work with high learning value or greater risk and consequences should normally involve a human first attempt and structured review. This may include setting hypotheses, designing client interviews or building arguments from incomplete evidence. Work with lower learning value and risk can be automated more aggressively, including routine data pulls, formatting and other repeatable production tasks.
The useful question is not whether AI should be used. It is what experience the work needs to provide, where judgment needs to be exercised, and where the consequence of weak reasoning is greatest.
Without explicit design, teams can drift towards over-reliance, where AI supplies too much of the thinking, or towards blanket restriction, where its useful contribution is lost. Neither produces the required capability.
Three design choices matter.
Juniors should be able to explain, in their own words, what they took from AI output, what they rejected and why. That builds the habit of committing and revising.
Reviewers need to see the thinking behind the artefact, not only the final artefact. This includes the pre-AI cut, the initial hypotheses, the prompts used, and the alternatives that were rejected. A polished slide deck is no longer sufficient evidence of good work.
Firms need to make a conscious decision about protected developmental time and AI capacity. Some of that work will not serve the immediate needs of a client engagement. It is an investment in future capability, not simply a margin cost.
On a recent transformation engagement I led, a team agreed that analysts would spend the first ninety minutes developing their own issue tree and initial hypotheses before using AI. The tool then generated additional branches, counterarguments and a first pass at relevant evidence. In review, the manager noticed that one apparently persuasive branch of the output rested on an assumption none of the analysts had tested. The team could see the gap because the original reasoning was visible. The final output was faster than the team’s previous process, but the more useful result was that the analysts had each formed a position before encountering the machine’s. The manager could see where the reasoning was strong, where it was borrowed and where further work was needed.
The risk of doing nothing
Inaction is not neutral. Firms may gain near-term efficiency while accumulating capability debt. This leads to junior professionals who can produce and edit fluent work, but who have had too little practice structuring problems, taking a view, identifying weaknesses or exercising judgment.
Some effects can appear immediately. Teams that cannot explain or defend the logic behind AI-generated work can produce recommendations that look polished but do not withstand challenge. That can lead to rework, poor client conversations and a loss of confidence in the senior team’s assurance of the work.
The longer-term risk is different. Managers and partners may eventually have less depth from which to supervise AI-augmented work credibly, because the people coming through have not had the experiences that previous generations acquired almost automatically.
In a conversation with a partner I have been coaching, we explored a live engagement where a competitor had promised a first draft in an afternoon using AI, while she was under pressure to improve her own margin. Rather than let the team begin with the tool, she asked each analyst to bring an initial view and its key uncertainties to review first. It cost time, but the review showed that the team could describe several AI-generated options without being able to say which they recommended to the client or what trade-off they were prepared to make. The point was not to reject AI. It was to ensure the team learned to take responsibility for the position it put forward.
This is why the issue cannot be left to individual leadership choice alone. Pricing, staffing targets, utilisation measures and partner incentives can make short-term efficiency the rational decision, even where it weakens the firm’s future capability. Firms that want a developmental stance need to decide how that stance will be funded and rewarded.
Uneven exposure
The capability question will not have the same significance everywhere.
Consulting fees pay for several distinct client needs. These include expertise the client cannot economically develop in-house, capacity at peak, third-party legitimacy for politically difficult internal decisions, accountability transfer when something controversial is being recommended, and signalling that an issue is being taken seriously.
AI threatens these needs unevenly. The expertise and capacity elements are most directly exposed. The legitimacy, accountability and signalling elements may retain value, and in some circumstances become more important, as analytical work becomes cheaper and more available.
The same is true of different practice models. A heavily leveraged practice, in which scale comes from teams doing substantial volumes of analysis and production work, faces a different set of pressures from a specialist advisory practice with lower leverage, or work where value rests more heavily on senior client relationships, judgment and accountability.
Some firms may decide that a leaner model, supplemented by senior hiring, is rational. That is a strategic choice. But firms that rely on internally developed managers and partners cannot simply assume those capabilities will still emerge.
The early-talent picture is not one of uniform contraction. The MCA’s 2026 Annual Industry Report, drawing on data from more than 80 member firms, recorded more than 5,400 hires in 2025. Graduate recruitment had risen by 10% and apprentice hiring by 31% on a like-for-like basis.4 This is not a measure of the whole market, nor does it settle how particular firms will change. It does reinforce the point that AI’s effects will vary by business model, practice and client situation.
Wider implications
If apprenticeship requires more design than it once did, the structures around it will change too. Firms may need to reconsider hiring, rotations, shadowing, early client exposure, career pathways and the allocation of partner attention.
The distribution of developmental work will become a strategic issue. Some cohorts may receive rich experience of problem solving, client exposure and structured review. Others may become efficient operators of AI-enabled workflows with fewer opportunities to practise judgment. Over time, that difference can shape promotion pipelines, the availability of credible future leaders and the distribution of capability across a firm.
These are important implications, but they are not all answered here. The point is not to offer a complete model of the future consulting firm. It is to identify a capability problem that can otherwise remain hidden inside a productivity story.
Managing partners should not begin with an abstract AI policy. They should start with one live engagement and understand what AI is doing to the work itself. Specifically, they should ask four questions.
Which tasks is it compressing and which is it relocating?
What should the team do for itself before turning to the tools?
What learning should this engagement produce for the juniors involved?
What would allow the senior team to see the reasoning behind the final output, rather than simply admire its fluency?
Making those choices explicit in one case is more useful than holding vague views about AI in general. The firms that develop the next generation well will not necessarily be those that use the least AI, or the most. They will be those that make deliberate judgments about what to preserve, what to redesign and what kind of professionals they are trying to build.
Sources
Saba Arab coaches partners, directors and founders through senior transitions, and advises leadership teams on AI, the partner model and how firms are organised. Start a conversation.
