The Analysts Who Predicted AI Would Kill Consulting Were Right About Everything Except the Direction

The prediction was specific, confident, and everywhere by late 2023: generative AI would disintermediate management consulting. If a $20-per-month subscription to ChatGPT could produce competitive analyses, synthesize market research, and draft strategic frameworks on demand, the margin justification for six-figure consulting engagements would evaporate. Democratized expertise, the argument went, would replace purchased expertise.

The prediction failed. Not narrowly — inverted. The global AI consulting market was valued at $11.13 billion in 2025 and is forecast to reach $116.8 billion by 2035 at a 26.49% CAGR. BCG attributed 20% of its record 2024 revenue to AI consulting, with projections to 40% by 2026. McKinsey estimates 40% of its total revenue now derives from AI and technology advisory. And in the most structurally revealing moment of the current cycle, OpenAI — the company whose tools supposedly rendered consultants redundant — announced multi-year Frontier Alliance partnerships with McKinsey, BCG, Accenture, and Capgemini in February 2026. The disruptors hired the disrupted. The question worth answering is why the original thesis was so wrong, and what the inversion tells us about where consulting value actually lives.

The Prediction Was Specific, Confident, and Almost Perfectly Backwards

The death-of-consulting thesis rested on a coherent model: consultants sell information and analytical frameworks packaged as expertise. AI produces both at near-zero marginal cost. Therefore the consulting addressable market should contract as clients self-serve.

What this model got right is real. AI genuinely automates large portions of what junior consultants did. The slide deck synthesis, the market sizing exercises, the competitor landscape mapping — these are now faster, cheaper, and increasingly client-executable without a consulting invoice. A landmark 2023 study by Harvard Business School and BCG of 758 BCG consultants found that GPT-4 users completed 12.2% more tasks, finished them 25.1% faster, and achieved over 40% better output quality compared to non-users. The productivity disruption argument had empirical backing.

What the model got catastrophically wrong was the assumption that the consulting value chain terminates at analysis. It never did. Analysis was always the table stakes — the minimum required to enter a client engagement. The actual product, what clients have always been buying at premium rates, is implementation confidence in environments of organizational, regulatory, and technical complexity. AI did not simplify that environment. It made it orders of magnitude more complex and raised the stakes on every governance and integration decision.

Why the Disintermediation Logic Broke Down at the Deployment Layer

The disintermediation thesis modeled consulting as a research and synthesis business. It should have modeled it as a change management, systems integration, and risk mitigation business. That category error explains the full prediction failure.

Enterprise AI deployment is not a software installation. It requires reconciling AI capabilities with fragmented legacy data architectures, redesigning core operational workflows, managing workforce transitions at scale, navigating a rapidly evolving regulatory environment across multiple jurisdictions, and constructing governance frameworks for systems that can hallucinate, discriminate, and fail in legally consequential ways. A 2025 MIT-cited analysis found that 95% of enterprise generative AI pilots fail to deliver measurable P&L impact — almost entirely due to integration, data, and governance gaps rather than model capability limitations. S&P Global Market Intelligence reported that 42% of companies abandoned most of their AI initiatives in 2025, up from just 17% in 2024.

The failure rate is the consulting market. Every stalled pilot, every abandoned initiative, every board that approved an AI budget and watched it dissolve into a proof-of-concept with no production path is a client who now needs professional help diagnosing what went wrong. Meanwhile, only 25% of executives strongly agree their IT infrastructure can support enterprise-wide AI scaling. The gap between organizational AI ambition and organizational AI capability is the consulting addressable market, and that gap is widening with every wave of AI capability that outpaces internal implementation readiness.

Complexity as a Consulting Engine: The $11 Billion Case Study

Capgemini's Chief Strategy Officer Fernando Alvarez articulated the structural dynamic precisely in a March 2026 interview with Fortune. Enterprise clients are not asking whether to use AI — that decision is made at the board level across virtually every major organization. The question is how: how to govern AI agents, how to secure them against adversarial inputs, how to connect them to fragmented internal data sources, how to satisfy compliance requirements, and how to integrate them with legacy systems that were never designed for interoperability with autonomous decision-making software.

This is not a question OpenAI's forward deployed engineers can answer at enterprise scale. OpenAI employs roughly 70 such engineers for all customer implementation work globally. Against a market where thousands of enterprises are simultaneously attempting AI transformation programs, that is not a deployment function — it is a proof-of-concept showcase. The Frontier Alliance announcement in February 2026 was OpenAI acknowledging what the market already knew: scaling enterprise AI adoption requires the organizational reach, sector-specific domain expertise, and pre-existing C-suite relationships that only the major consulting incumbents possess. The AI company needs the consulting firm. The disintermediation thesis assumed the opposite.

The mandate structure this creates is new. Clients are now commissioning three distinct categories of consulting work that barely existed three years ago: AI readiness assessments that audit data governance and infrastructure before any model deployment; AI governance architecture engagements that build the oversight, audit, and compliance frameworks that regulators and boards are demanding; and AI change management programs that address the workforce redeployment and process redesign challenges that follow every successful implementation. None of these existed as meaningful practice areas in 2022. All three are growing faster than the core strategy and operations practices they sit alongside.

What Alvarez Gets Right That Most Critics Miss

Alvarez frames Capgemini's competitive positioning around domain expertise rather than generic advisory capability. "The conversation is, do you have the domain expertise to understand my problem?" he told Fortune. This is the epistemological insight the disintermediation thesis never modeled: AI commoditized generic expertise, not domain expertise.

A general-purpose language model can produce a go-to-market strategy memo. It cannot tell a global pharmaceutical company how to restructure its clinical trial data workflows for AI-assisted analysis in a way that satisfies FDA 21 CFR Part 11 requirements, preserves data provenance across CRO partnerships, and avoids rebuilding the LIMS platform that cost $200 million to implement in 2019. The domain knowledge, regulatory context, and organizational history required for that engagement are not available in a training corpus. They require years of sector-specific practice and client relationship depth.

Alvarez's observation about outcome-based pricing is the commercial confirmation of this thesis. Capgemini is shifting away from time-and-materials billing toward accountability for delivered results. That pricing model is only viable when the firm has enough domain and implementation confidence to accept outcome risk. Generic advisory firms billing for hours cannot make that shift. The move to outcome-based contracts is a structural signal that consulting value is concentrating in firms with deep specialization, not dispersing toward commoditized analysis products.

The Consultants Who Will Actually Lose — and Why That Proves the Thesis

The death-of-consulting narrative will be proven partially correct, but about precisely the wrong tier. The junior generalist analyst producing slide decks that synthesize publicly available information faces genuine structural pressure. McKinsey has deployed 12,000 AI agents internally, with work that previously required 14 consultants now requiring two or three alongside AI assistance. The commodity analysis work is being automated, and the headcount attached to it will contract meaningfully over the next several years.

But this compression at the generalist tier validates rather than refutes the broader argument. The industry is not dying; it is stratifying. The premium on domain expertise, governance architecture, enterprise change management, and senior client relationship capital is rising precisely because the junior-analyst labor arbitrage model is disappearing. Consulting firms that have invested in AI capability platforms and deep sector specialization — McKinsey, BCG, Accenture, Capgemini — are capturing the value gap at the top. Boutique generalists with no differentiated practice are the actual casualty class.

The 86% of consulting buyers who now actively seek AI-integrated services, and the 66% who will terminate relationships with firms that fail to integrate AI into their offerings, are not describing a market that is contracting. They are describing a market that is demanding a higher standard from its advisors. The analysts who predicted AI would kill consulting were correct that AI would eliminate the work that was purely information synthesis. They missed the far larger reality: eliminating that work makes the remaining consulting more valuable, more defensible, and more necessary than at any prior point in the industry's history. The $11 billion market and the OpenAI Frontier Alliance are the forensic evidence.