McKinsey has deployed 12,000 AI agents to assist its 40,000 employees. At the same time, headcount has fallen from 45,000 to 40,000. This dual movement—more AI, fewer employees—encapsulates the tension running through consulting today. AI for consulting firms is no longer a marginal experiment: it is an operational capability reshaping the deliverable production chain, team structures and the value proposition presented to clients.
BCG generated $2.7 billion in AI-related revenue in 2024, 20% of its total revenue—a share it projects will reach 40% by 2026. Accenture reports $3.6 billion in AI bookings for fiscal 2025, doubling year over year. The question is no longer whether firms should adopt AI, but how to do so without sacrificing what makes them valuable: human expertise, contextual judgment and the ability to turn analysis into decisions.
TL;DR — Generative AI lets consulting firms reduce time spent on research, synthesis and document production by 50 to 70%. But gains are real only if integration preserves junior development and senior quality control. This article details concrete use cases, pitfalls to avoid and architectures that work.
Consulting at an Economic Turning Point
A Market Under Pressure Despite Apparent Growth
France's consulting sector generates nearly €20 billion in revenue across more than 15,000 companies. Behind that robust exterior, however, signs of strain are accumulating. According to Syntec Conseil, strategy and management consulting recorded zero growth in 2024. French IT services firms grew just 0.7% over the same period, well below Numeum forecasts. Nearly 40% of sector professionals expect further economic deterioration in 2025.
Shrinking public-sector budgets, tougher procurement constraints and geopolitical instability squeeze margins. In this environment, hiring to deliver more is no longer viable for most firms. The historical model—billing person-days and adding layers of junior consultants—is reaching its structural limits.
The Pyramid Model Nears the End of Its Cycle
Consulting's conventional structure is a pyramid: many juniors executing, a few managers supervising and partners selling. AI destabilizes this model from the base. Tasks traditionally assigned to analysts and junior consultants—desk research, data compilation and drafting summary slides—are precisely those generative AI automates best.
According to Napta, the sector is moving toward a diamond structure: fewer execution roles and more experienced specialists. This transformation is not cosmetic. It changes recruitment policy, pricing policy and the entire business model. Firms clinging to the pyramid without integrating AI lose competitiveness against leaner organizations delivering as much—or more—with fewer people.
Pressure on Time-Based Billing
The traditional person-day billing model directly conflicts with AI. If a deliverable that once required five days now takes two through automation, billing for five becomes difficult to defend to informed clients. The sector is moving toward hybrid models: fixed fees, success fees and value-based rather than time-based billing.
This transition forces firms to rethink their value proposition. It is no longer hours worked that justify the price, but analytical quality, relevant recommendations and delivery speed. AI then becomes a margin accelerator—provided it is deployed without weakening the substance of the advice.
Concrete AI Use Cases in Consulting
Research and Market Intelligence: From Two Weeks to a Few Hours
Desk research absorbs a disproportionate share of consultants' time, especially at the start of an engagement. Identifying relevant studies, cross-referencing sector data and synthesizing 200-page reports historically occupied junior teams for weeks.
Bain & Company observes that research tasks requiring two weeks of junior consultant work are now performed “almost instantly” through targeted queries to specialized AI agents. McKinsey's Lilli platform handles over 500,000 queries a month and saves approximately 30% of research and synthesis time.
The gain extends beyond speed. AI queries dozens of sources simultaneously, detects correlations a human eye might miss and produces structured summaries project teams can use directly.
Deliverable Production: Automating Without Standardizing
Drafting commercial proposals, audit reports, industry benchmarks and client presentations is central to a firm's production. It is also where measured productivity gains are most striking.
CMI has developed around thirty specialized AI agents covering consulting's main recurring tasks. Measured results: 60 to 70% time savings in preparing commercial proposals and 50% in summarizing large documents. The principle: each agent combines an optimized prompt with an internal document collection—reference deliverables, proprietary methodologies and sector data.
| Deliverable | Traditional time | Time with AI | Measured gain |
|---|---|---|---|
| Commercial proposal | 5–8 days | 1.5–3 days | 60–70% |
| Report summary (200 slides) | 2–3 days | 0.5–1 day | 50% |
| Industry benchmark | 3–5 days | 1–2 days | 40–60% |
| Document due diligence | 2–4 weeks | 3–7 days | 50–65% |
| Structured meeting notes | 2–4 hours | 15–30 minutes | 80% |
The main risk is standardization. An AI-produced deliverable without supervision resembles every other AI-produced deliverable. Differentiation comes from the review, enrichment and contextualization only an experienced consultant can provide.
Data Analysis and Decision Support
AI does more than produce text. Analytical agents process complex datasets, identify patterns, model scenarios and generate usable visualizations. For a consulting firm, this changes the depth of analysis available on a standard engagement.
A consultant equipped with AI tools can explore multiple hypotheses in a few hours where manual analysis would force prioritization—and therefore blind spots. A study by Harvard, Wharton, Warwick and MIT involving 758 BCG consultants measured a 40% improvement in deliverable quality on creative and structured tasks when AI was used.
But the same study reveals a clear limit: on analytical tasks requiring nuanced contextual judgment—interpreting ambiguous qualitative data and formulating subtle strategic recommendations—consultants using AI saw performance fall by 23%. AI excels in structured processing; it fails when reasoning requires situational understanding absent from the data alone.
The Technical Architecture of an AI-Augmented Firm
From Generic Chatbot to Specialized Agents
The first adoption wave—giving all consultants ChatGPT access—produced uneven results. Without structure, every consultant reinvents prompts, gets outputs of varying quality and spends as much time correcting them as they would producing them manually.
Firms gaining a real advantage from AI have moved beyond this stage. They deploy specialized agents, each dedicated to a specific task: one for sector intelligence, another for proposal writing, a third for financial analysis. PMP Strategy creates agents by pairing an optimized prompt with internal documents specific to each expertise area. Access is through a simple interface, such as an “@” command in a collaborative environment.
McKinsey illustrates the next stage with 12,000 AI agents deployed organization-wide, integrated into the daily workflows of 72% of its professionals. Lilli generates over 500,000 monthly queries—a volume demonstrating real adoption, not a pilot confined to a few teams.
The Importance of the Proprietary Document Base
An AI agent performs well only if its knowledge base is relevant, current and structured. For a consulting firm, this includes past engagement deliverables, proprietary methodologies, up-to-date sector data and internal lessons learned.
Establishing a consulting-specific RAG (Retrieval-Augmented Generation) system requires preparatory document structuring that is often underestimated. Firms getting this right obtain agents capable of producing deliverables consistent with their methodological identity. Others get generic content, indistinguishable from what any competitor could produce with the same public tools.
Checklist: Technical Prerequisites for Effective AI Deployment in a Consulting Firm
- Structured, indexed document base (deliverables, methodologies, sector data)
- RAG pipeline connected to internal repositories with automated updates
- Specialized agents by task type (research, writing, analysis, intelligence)
- Mandatory human validation before client distribution
- Strict confidentiality policy (client data never exposed to public LLMs)
- Monitoring of AI usage and output quality
- Training differentiated by experience level (junior versus senior)
Client Data Confidentiality and Governance
The Xerfi study highlights a critical point: using public AI solutions exposes firms to confidentiality and cybersecurity risks. The client data consultants handle—strategic plans, financial data, acquisition projects—is inherently sensitive. Sending it to third-party APIs without a non-retention guarantee constitutes potential professional misconduct.
Large firms invest in private AI instances hosted on their own infrastructure or sovereign clouds. For independent and mid-sized firms, the cost of these proprietary solutions is a major obstacle. This is precisely where custom tools—AI agents deployed in private environments and connected only to authorized data—make sense.
The Deskilling Trap: Preserving Human Expertise
The Harvard-BCG Study and the “Technological Frontier”
The experiment conducted by Harvard, Wharton, Warwick and MIT researchers with 758 BCG consultants revealed a phenomenon firm leaders cannot ignore. On structured and creative tasks (product design, strategic brainstorming), consultants using AI completed 12.5% more tasks with 40% higher quality.
But on complex analytical tasks—those requiring nuanced interpretation of qualitative data and contextual judgment—performance fell 23%. Consultants relied on AI outputs without challenging them sufficiently, accepting plausible but incorrect conclusions.
This “technological frontier” defines the dividing line between what AI can and cannot do in consulting. Ignoring it risks delivering strategic recommendations whose quality deteriorates precisely where it matters most.
The Junior Training Deficit
A Stanford study published in August 2025 confirmed an average 16% decline in junior positions since late 2022, particularly in software development, administrative functions and consulting. If tasks that traditionally trained young consultants—exhaustive research, data compilation and first drafts—are automated, how will these professionals acquire the expertise needed to become seniors?
Stanford research on 5,000 customer support agents had already shown that AI increases productivity by 14% on average, but asymmetrically: +35% for beginners and a modest gain for experts. AI raises the floor in the short term—but risks producing a generation of consultants who never learned to analyze without a technological crutch.
Strategies to Maintain Skills Development
Firms managing this transition with clear judgment establish explicit safeguards:
Supervised rotation. Juniors use AI to produce a first draft, then must critique, enrich and defend it before a senior. The final deliverable bears their intellectual stamp, not the tool's.
AI-free zones. Some engagements or phases deliberately remain unassisted. The aim is to force the acquisition of fundamental analytical habits AI cannot teach.
Assessment of reasoning, not volume. Junior evaluation criteria refocus on reasoning quality, recommendation relevance and the ability to challenge AI outputs—rather than the number of slides produced.
Bain & Company now targets 30% “tech-enabled consultants” in recruitment: people who understand technology without being developers, able to use AI to extend their capabilities rather than substitute for them.
Transforming the Business Model Without Sacrificing Quality
From Time-Based to Value-Based Billing
AI accelerates production but compresses the traditional billing base. An industry benchmark delivered in two days instead of five no longer justifies the same billed days. Yet its client value is identical—or higher if quality improves.
This tension pushes the sector toward billing aligned with value delivered rather than time spent. According to Napta, firms are experimenting with fixed fees, success fees tied to client performance indicators and subscriptions to continuous intelligence or analysis services.
| Billing model | Principle | AI compatibility | Main risk |
|---|---|---|---|
| Person-days | Time spent × daily rate | Low—AI reduces time | Revenue loss if the client demands transparency |
| Fixed fee | Fixed price per deliverable | High—AI gains improve margin | Underestimating actual complexity |
| Success fee | Compensation tied to outcomes | Medium—depends on external factors | Difficulty isolating the firm's contribution |
| Subscription | Continuous access to a consulting service | High—AI reduces service costs | Requires sufficient client volume |
| Hybrid | Fixed fee + outcome-based variable | Optimal—combines predictability and alignment | Contractual complexity |
The Economics for a 50-Consultant Firm
Consider a strategy consulting firm with 50 consultants, an average daily rate of €1,500 and 75% utilization. Theoretical annual revenue is approximately €12.4 million.
If AI reduces deliverable production time by 30% without quality loss, two scenarios emerge. In the first, the firm retains headcount and takes on more engagements: potential revenue rises to €16–17 million. In the second, it reduces headcount by 20% while maintaining the same engagement volume: operating margin increases significantly at unchanged revenue.
Reality will fall between the two, shaped by the firm's ability to sell more (scenario 1) or change its pricing structure (moving to fixed fees). Either way, a firm not integrating AI experiences gradual competitive erosion against rivals delivering faster for the same price—or at the same pace for less.
Transparency with Clients
Should clients be told that AI helped produce their deliverables? Practices differ. Bain and PMP Strategy embrace full transparency. Oliver Wyman considers that clients buy an outcome, not a production method. Publicis Sapient openly incorporates AI tools into delivery.
The trend is toward greater transparency, driven by growing procurement demands. A firm concealing AI use takes a reputational risk disproportionate to the benefit—especially if a competitor uses transparency as a selling point.
Roadmap: Integrating AI into a Firm Without Breaking Everything
Phase 1—Equip Low-Value Tasks (Months 1–3)
Start with tasks nobody enjoys—the “no-joy tasks.” Data extraction, large-document summaries, meeting notes and routine benchmark compilation. These absorb 30 to 40% of consultants' time without forming the core of their added value.
Deploy AI tools in these areas first. ROI is immediate, resistance is minimal (nobody defends the right to write minutes) and quality degradation risks are low because these intermediate deliverables are not exposed directly to clients.
Phase 2—Build Specialized Business Agents (Months 3–6)
Once initial productivity gains are validated, build specialized agents aligned with proprietary methodologies. A proposal-writing agent fed your 50 best past proposals. A sector-analysis agent connected to your reference databases. A due diligence agent applying your internal evaluation grid.
This phase requires investment in document structuring and technical development. CMI built around thirty such agents—a realistic order of magnitude for a medium-sized firm.
Phase 3—Redesign Pricing and Training (Months 6–12)
Technical integration is not enough. The business model and talent development must align with the new operational reality. Experiment with fixed-fee offerings in clearly defined areas. Revise junior evaluation criteria to include challenging AI outputs. Train seniors to manage AI agents, not merely use ChatGPT.
EY trained 83% of its 400,000 employees in AI—a massive effort illustrating the scale of transformation required. PwC invested $1 billion over three years, measuring efficiency gains of 20 to 30% across its workforce. Training investment is not optional: it is the prerequisite for technical gains to become business gains.
Phase 4—Manage, Measure, Adjust (Ongoing)
Only 34% of leaders report tangible AI-related profitability improvements, according to the PwC UK CEO Survey 2025. The gap between promises and measured results often stems from a lack of rigorous management. Define clear indicators: production time by deliverable type, rework rate after AI validation, client satisfaction and margin per engagement. Without metrics, AI remains an expense—not an investment.
What AI Will Never Replace in Consulting
Strategic Judgment Under Uncertainty
Data does not contain its own interpretation. A senior consultant advising a client to abandon an acquisition despite favorable financial indicators—because they perceive cultural risk, a weak market signal or unfavorable internal politics—exercises judgment AI cannot reproduce. The Harvard-BCG study demonstrated it: AI performance collapses by 23% on tasks requiring this kind of contextual reasoning.
Client Relationships and Change Management
Persuading an executive committee to transform its organization, supporting teams adopting a new process and navigating the internal politics of a large company: these relational and emotional skills remain beyond AI's reach. They are precisely the value layer clients pay most for—and the one firms should strengthen rather than automate.
Accountability for Recommendations
An AI agent does not bear responsibility for a strategic recommendation. When a firm stakes its reputation on a €50 million transformation plan, a partner signs—not an algorithm. This human accountability underpins the trust model on which consulting rests. AI increases production and analytical capacity; it does not replace the commitment to accountability that gives consulting its fiduciary value.
FAQ
Will AI replace consultants? No, but it redefines their role. The Harvard-BCG study shows AI improving productivity by 12.5% and quality by 40% on structured tasks while reducing performance by 23% on complex strategic analysis. Consultants move from execution toward oversight and judgment—skills AI has not mastered.
What productivity gains are realistic for a consulting firm? Practical data ranges from 30 to 70% depending on deliverable type. CMI measures 60–70% savings on commercial proposals and 50% on document summaries. McKinsey estimates 30% time savings on research and synthesis through Lilli. What matters is measuring by task type, not a global average.
How do you protect client data when using AI tools? Private AI instances hosted on controlled infrastructure are standard for large firms. For smaller organizations, custom AI agents deployed in secure environments offer a viable alternative. Client data must never pass through public APIs without contractual non-retention guarantees.
How much does AI integration cost for a medium-sized firm? Investment ranges from €50,000 to €300,000 for a firm with 30 to 80 consultants, depending on ambition. Initial steps (tools for low-value tasks) are accessible for a few thousand euros a month in SaaS licenses. Specialized business agents with proprietary RAG account for most of the budget.
Should clients be informed about AI use in producing deliverables? The market trend is toward full transparency. Firms such as Bain and PMP Strategy openly embrace it. Concealing AI use creates increasing reputational risk, especially as procurement teams become better informed and more demanding on this point.
How do you prevent junior consultants from becoming deskilled? Three main measures: retain “AI-free zones” in certain engagement phases to require foundational learning, require juniors to critique and enrich AI outputs rather than accept them unchanged, and assess reasoning quality instead of output volume. Bain now recruits 30% “tech-enabled” profiles capable of directing AI without being replaced by it.
AI Coder Squad: AI Agents Tailored to Consulting
Consulting firms seeking to move beyond shared ChatGPT access need specialized agents connected to proprietary data and deployed in a secure environment. This is exactly the kind of solution a senior development team experienced in AI architectures can deliver in a few weeks.
AI Coder Squad designs custom applications and AI agents for companies that want to move fast without sacrificing quality—with senior developers and an AI-powered approach.
→ Start your project and discover how AI Coder Squad can accelerate your next delivery.