In short

AI in management consulting changes more than individual tasks: it changes structure, value creation, offerings and training. AI agents take over research, drafts and routine work, the pyramid narrows, and partners and senior consultants move back closer to the content. Offerings shift from day rates to tools and outcomes. Firms that switch now set shared tool standards, run a lighthouse project with the new way of working and train juniors so that judgment comes before AI speed. LEADBeyond, founded by former partners at McKinsey & Company and Roland Berger, trains consultants at leading firms, from analyst to partner level.

Key takeaways

  • Many firms use AI in pockets: a use case here, a tool there. The leverage lies in changing how the firm works, roles included.
  • The pyramid narrows: less routine for analysts, more agents, and partners and senior consultants closer to the work again.
  • Every step of the project chain can be rethought, from research and data requests to the deliverable, which no longer has to be a deck.
  • Offerings become agent-backed and paid by outcome: cockpits, portals and prototypes instead of lists of measures, with handover and training so they keep running at the client.
  • Every AI answer is a draft: five checks (confidentiality, sources, plausibility, status of hypotheses, accountability) turn it into a result.
  • Juniors are still needed, because without juniors there are no seniors. They learn the method first, then AI as accelerator and sparring partner.
  • The fastest start: shared standards, a lighthouse project next to the old way of working, then roles, training and offerings.

Where consulting firms stand with AI today

In many consulting firms AI runs in pockets: a group-wide chat assistant with prompt templates, Copilot for part of the staff, individual use cases that keen colleagues try out at the weekend. That is a start, but not yet a new way of working. Sooner or later the partner group asks the real question: does AI change something fundamental about our business? The honest answer is yes, and faster than earlier technology waves. McKinsey shows how far the largest firms have come: global managing partner Bob Sternfels cited about 25,000 AI agents alongside roughly 40,000 employees in early 2026 (CNBC). On top of that comes new competition: AI labs such as Anthropic and OpenAI are building and buying consulting firms of their own in 2026 (why AI labs are buying consulting firms).

We see two patterns again and again. Without a shared standard, shadow use grows: whoever lacks a good tool at work uses a personal account for the market scan behind a proposal. And without time and guidance, usage stays low, even among those you'd expect to be hungriest: someone who tried a tool once and got a weak result often gives up. Where a firm stands can be read from five stages:

Self-assessment

Five stages: where does your firm stand?

1StageIndividual tools

How to recognise it

Use depends on individuals. There are hardly any shared rules.

Next step

Approved tools, a data traffic light, basics for everyone.

2StageShared standards

How to recognise it

Everyone knows which data may go into which tool.

Next step

Map the steps of each offering: what does the AI draft, who checks?

3StageAI in the project flow

How to recognise it

The project plan names, step by step, what the AI drafts and who checks.

Next step

Agents with fixed roles. Specifications, not ad-hoc requests.

4StageHybrid project teams

How to recognise it

Staffing lists people and agents. People sign off.

Next step

Recut the offerings: what stays with the client?

5StageAgent-backed offerings

How to recognise it

The handover covers model, source register and operation.

Next step

Portfolio and pricing: handover, further development or operation.

LEADBeyond assessment. People as circles, agents as dashed squares; numbers illustrative.

Five stages of AI in a consulting firm. Pick a stage to see how to recognise it and what the next step is.

The four areas below map what changes on the way from one stage to the next.

1. Structure: how the consulting pyramid changes

For decades it was clear who did what in the pyramid. With AI agents that shifts: agents take over research, data preparation, first drafts and cross-checks. Partners and principals are pulled back into the substance, because with agents they can often do things faster themselves than delegate them. The pyramid narrows and rests on a base of agents.

Team shape

AI turns the consulting pyramid 180 degrees.

PartnerJudgment and client
Engagement managersSteering
ConsultantsAnalysis and storyline
AnalystsResearch, numbers, slides
People moderateClient, judgment, sign-off
People checkSource, plausibility, quality
A few lead agentsBrief, test case, acceptance
Agents carry the baseResearch, model, slides, code

Lever: many juniors per partnerLever: many agents per person with judgment

Turned 180 degrees

The wide side is on top. Judgment becomes the majority of the team.

Base
Agents draft. A few people lead them.
Lever
Agents per person with judgment.
Learning
Check early, push back, take responsibility.

How we work ourselves: no pyramid, with agents as the base.

LEADBeyond thinking model, not a headcount forecast. Roles illustrative.

The consulting pyramid, classic and with AI. Switch between the two views. A thinking model, not a headcount.

That changes roles, not just headcount:

  • Partners write the storyline themselves again and have the team verify the facts behind it, rather than the other way round.
  • Engagement managers lead people and agents: they break hypotheses into analyses, steer the agents and hold the checkpoints.
  • Juniors check, verify and sharpen results, run interviews and build judgment instead of formatting slides.
  • A central AI function, such as a head of AI, steers the internal shift as a cross-cutting role, tightly linked to the practices and initially without revenue or utilisation targets of its own.

Recurring workflows need rethinking too. A response to a tender used to tie up a team for days. With a well-built proposal engine of agents for research, references and drafting, a high-quality draft can be ready in a day, and partners spend their time on the solution rather than the layout.

2. Value creation: rethinking the project chain

The biggest lever lies not in automating single tasks but in rethinking the whole value chain of a project once, radically. The pattern is the same at every step: an AI agent drafts, a checkpoint secures it, a person decides.

Value creation

AI drafts. People check and decide.

  1. Person decidesCheck against the primary sourceAI agent draftsScreen sources, collect references
  2. Person decidesChoose, set the barAI agent draftsOptions and counter-positions
  3. Person decidesOwn the assumptionsAI agent draftsModel and scenarios
  4. Person decidesListen, read between the linesAI agent draftsGuide and analysis
  5. Person decidesJudgment and recommendationAI agent draftsSort findings, dissenting views
  6. Person decidesOwn the upkeepAI agent draftsModel and source register

Research. The AI delivers drafts with references. A person checks them against the primary source.

Hypotheses. The AI widens the solution space. A person chooses and sets the bar.

Model. The AI runs every scenario. A person sets and owns the assumptions.

Interviews. The AI prepares and analyses. A person runs the conversation.

Synthesis. The AI maps findings to the hypotheses. A person makes the call.

Handover. Instead of a deck, a tool stays. One person owns it.

The time saved goes intoPrimary sourcesInterviewsTools that stay

Assessment from our project work, not a measurement of time or cost.

The project chain as a relay: at every step an AI agent drafts, a person checks and decides. Click a step.

What matters is where the time saved goes: into checking against primary sources, into more conversations with clients and experts, and into tools that stay after the project. The data request to the client can also be derived from the analyses and prioritised this way, instead of starting as a long list up front.

Hypothesis-driven work becomes more important, not less: whoever breaks the question cleanly into hypotheses can deploy agents precisely, and less experienced engagement managers grow into the role faster. What this looks like from the client side is covered in AI-aided strategy consulting.

3. Offerings: from day rates to tools and outcomes

The most interesting question for consulting firms is not how to become more efficient internally, but what they will sell. Three directions are emerging:

  • Agent-backed services. Recurring work becomes a service with agents in the background: live cockpits for performance management, ongoing monitoring, client portals where clients work with the results.
  • A prototype instead of a list of measures. Whoever shows a measure as a working prototype rather than just recommending it is more convincing and more often wins the implementation too. One example is our AI order intake.
  • Paid by outcome. When agents replace hours, the day rate measures the wrong thing: it rewards effort rather than impact and turns every gain in speed into lost revenue. Fixed prices for clearly defined results, building blocks for tool and operation and, where measurable, a share tied to an agreed KPI take its place. What two invoices for the same project look like is shown in The future of consulting.
  • Enablement as part of the service. A tool without a trained team goes stale. Handover, training and one owner at the client therefore belong in the proposal, not in a change request.

The market is there: in a Boston Consulting Group survey of 1,250 executives, only 5% of companies qualify as "future-built" for AI, and 60% report minimal revenue and cost gains (BCG). These companies need fewer recommendations and more tools that run, plus teams that can operate them.

Using AI quietly in the background only buys a seat at the table: firms that don't use it will soon fall behind on price and speed. The difference is made by firms that show their AI capability, tailor their own assistants to their business and sell that as a strength. What that looks like in a project is shown in the scenes in The future of consulting. How clients compare AI-native and traditional firms is shown in AI-native vs. traditional consulting.

There are limits: consultancies within audit or legal groups are bound by professional rules, and not every tool is allowed there. Two principles still help: build vendor-agnostic, so that switching models does not force a rebuild, and check which parts, such as transcription and minutes, can run on locally hosted models.

4. Training: judgment before AI speed

Do consulting firms still need juniors? Yes. Without juniors today there are no seniors tomorrow, and AI-native joiners bring ways of working that experienced teams still have to learn. The question is how juniors without experience can still work well with AI. Our answer from our training programmes: method first, then AI.

  • Method first. Problem statement, hypothesis tree, pyramid principle: whoever has learned this by hand can tell when an AI draft is good and when it only looks good. The hypothesis tree in particular is regularly the aha moment in our training, at partner level too.
  • Then AI as accelerator and sparring partner. Not like a shotgun that sprays in every direction, but step by step: to the next milestone, adjust, carry on.
  • Built in, not bolted on. AI belongs in the regular curriculum, from Lead Self through Lead Others to Lead Change, not in a separate AI seminar.
  • Time and permission. People who never get time to try tools don't use them. Training needs approved tools and real cases.

Research also shows that experience changes the effect of AI: in a field study, less experienced workers gained a lot from an AI assistant and the most experienced barely (Brynjolfsson, Li and Raymond). For consulting firms that means AI can make juniors productive faster, but only if they learn the judgment along with it.

What "judgment before AI speed" means in practice comes down to five checks. Every AI answer is a draft; only checking turns it into a result. A second model that argues the counter-position on instruction helps, because language models tend to agree with their users (Sharma et al.).

Judgment

Every AI answer is a draft. Five checks turn it into a result.

AI draftInvented scenario

„The market for lab equipment in the new application field will reach about €1.84bn by 2030, according to the Lab Market Industry Report 2026.“

After the five checks: a range of €1.1bn to €1.5bn from two independent derivations, marked "to be validated", checked and signed off by a named person.

Or click the checks one by one

Which data may go where

  • Green. Public information
  • Yellow. Cleanly anonymised
  • Red. Only in the approved tool
  • Black. Off limits: insider and personal data

A second model argues the counter-position on instruction, because language models tend to agree with their users.

Checks and traffic light from how we work. Not legal advice.

The five checks on an invented example. Click the checks or "Check all": only when all are met does the draft become a paper.

Which check demands most of whom depends on the level. The higher the level, the more the focus shifts: from working with AI through steering hybrid teams to judging and owning the result.

Training

The higher the level, the more it is about leading change.

Three dimensionsLead Selfleading yourselfLead Othersleading teamsLead Changeleading change

Level 1

Analyst & consultant

60 %Lead Self25 %Lead Others15 %Lead Change
AI focusWorking with AI

Research with sources, AI as a sparring partner, every number checked against the primary source.

Level 2

Senior consultant

40 %Lead Self35 %Lead Others25 %Lead Change
AI focusChecking and pushing back

Apply the five checks reliably, seek counter-positions, ranges instead of false precision.

Level 3

Engagement manager

20 %Lead Self40 %Lead Others40 %Lead Change
AI focusSteering hybrid teams

Cut tasks for agents like a good delegation: with a specification, a test case and sign-off.

Level 4

Principal & partner

15 %Lead Self30 %Lead Others55 %Lead Change
AI focusJudging and owning

Agree data rules with clients, recut offerings, sign off results by name.

Weights of the three dimensions per level in percent, as in our training architecture. AI focus areas: our proposal, to be adapted to your firm.

Career ladder with an AI focus per level. Pick a level.

Will AI replace consultants?

AI replaces tasks, not consulting. Agents increasingly take over research, drafts, formatting and many analyses. What remains is the part clients actually pay for: the right question, judgment under uncertainty, trust in the boardroom and accountability for the recommendation. A field experiment with employees of a management consultancy also shows that AI can lead to worse results on tasks beyond its capabilities (Dell’Acqua et al.). What gets replaced is consultancies that keep working as before, by consultancies that use AI consistently.

Five steps

How consulting firms make the switch to AI

  1. Set standardsWhich tools are allowed in the firm, which data may go where, and how conversations are recorded. Without a standard, shadow use grows.
  2. Start a lighthouse projectOne team works a real project the new way: conversations captured, agents on the team, results as interactive pages. Ideally next to a team working the classic way, so the difference shows.
  3. Redraw the pyramid and rolesWho does what once agents take over the routine? Adjust role profiles, staffing and career paths.
  4. Build AI into the curriculumFrom the first week to partner level: method first, then AI, with real cases and the tools the firm actually allows.
  5. Turn offerings into agent-backed services and price themWhich services become tools, cockpits or portals, and how are they paid for by outcome?

Then the cycle starts again: what the lighthouse project learned becomes the standard for the whole firm.

Frequently asked questions

How is AI changing management consulting?

In four areas: structure (the pyramid narrows, agents take over routine), value creation (every project step from research to deliverable is rethought), offerings (tools, cockpits and portals instead of decks, pricing more by outcome) and training (method and judgment first, then AI as accelerator).

Will AI replace consultants?

AI replaces tasks, not consulting. Agents take over research, drafts and many analyses. The right question, judgment under uncertainty, trust and accountability stay human. The firms under pressure are mainly those that do not use AI consistently.

How does AI change the consulting pyramid?

It narrows and rests on a base of AI agents. Fewer analysts are needed for research and slides; partners and senior consultants move closer to the content, and engagement managers lead people and agents at the same time.

Do consulting firms still need junior consultants?

Yes. Without juniors there are no seniors, and AI-native joiners bring valuable ways of working. Their role shifts from formatting to checking, interviewing and judging. To get there they learn the method first, then AI.

How should a consulting firm start with AI?

With shared standards for tools, data and recording, then a lighthouse project in which one team works the new way consistently. New roles, training in the curriculum and agent-backed offerings follow.

How do consultants check AI output?

With five checks: confidentiality (was the content allowed into this tool?), sources (is every number backed by the primary source?), plausibility (do independent derivations reach a similar result?), status of hypotheses (is it clear what is confirmed?) and accountability (has a named person signed it off?). Only when all five are met does an AI draft become a result.

What AI tools do consultants use?

Typically a chat assistant on a company account, such as Microsoft Copilot, ChatGPT or Claude, plus tools for research and knowledge search, meeting transcription and, increasingly, AI agents for recurring workflows such as proposals or data requests. More important than any single tool is a shared standard across the firm.

Sources

  1. Rosenbaum, E. (2026, January 29). How CEOs bringing AI agents to work are preparing customers and employees. CNBC., CNBC (2026). Cited only for the figure given by McKinsey's global managing partner Bob Sternfels (CES, January 2026): about 25,000 personalised AI agents alongside roughly 40,000 employees. The firm's own statement, not an independent count.
  2. Dell’Acqua, F., McFowland III, E., Mollick, E., Lifshitz, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science, 37(2), 403–423., INFORMS (2026). Pre-registered field experiment with employees of a management consultancy; cited qualitatively only, for AI's 'jagged' capability frontier. Effect sizes deliberately not used.
  3. Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942., Oxford University Press (2025). Field study in customer support; cited qualitatively only, for less experienced workers gaining a lot and the most experienced barely. Not directly transferable to consulting.
  4. Boston Consulting Group. (2025, September 30). AI Leaders Outpace Laggards with Double the Revenue Growth and 40% More Cost Savings. Press release on the report "The Widening AI Value Gap: Build for the Future 2025"., Boston Consulting Group (2025). Global survey of 1,250 senior executives and AI decision makers. Cited for: 5% of companies qualify as "future-built", 60% report minimal revenue and cost gains from AI. Self-reported, categories set by BCG.
  5. Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S. R., et al. (2024). Towards Understanding Sycophancy in Language Models. International Conference on Learning Representations (ICLR 2024)., ICLR (2024). Cited for language models tending to agree with their users; that is why a second model argues the counter-position on instruction in our work.

Discuss a programme for your leaders

30 minutes, no pitch. We listen, ask the right questions and tell you honestly whether and how we can help.

Discuss a programme