The seventy percent problem.

The three numbers that define this problem.

Adoption is no longer the question. Impact is. The gap between the two is the seventy percent of AI transformation effort that consulting firms structurally cannot deliver — and that mid-market operators have to figure out for themselves, or hire Mezura.

In Q3 2025, McKinsey surveyed 1,993 organizations across 105 countries for its State of AI 2025 report. The headline finding: 88% of organizations reported using AI in at least one function, up from 78% the previous year. Adoption is solved.

The same report measured impact differently. 39% reported any measurable EBIT impact from AI, and the majority of those said the impact was less than 5% of EBIT. The cohort attributing 5% or more of EBIT directly to AI — McKinsey's "AI high performers" — was just 6% of organizations (McKinsey, 2025).

MIT NANDA's GenAI Divide (July 2025) measured it from the project side: 95% of generative AI projects deliver no measurable ROI to the deploying enterprise. The 5% that do are concentrated in organizations that rebuilt their operating models around AI economics — not those that deployed AI on top of existing operating models.

These three numbers — 88% adopting, 6% impacting, 95% of projects failing — define the shape of the problem. The technology works. The deployment does not.

BCG's 10/20/70 framework.

Boston Consulting Group has spent a decade quantifying where AI transformation value comes from. The answer, validated repeatedly: 10% algorithms, 20% technology and data, 70% people and processes. The seventy percent is the work most firms cannot sell.

Reaffirmed in BCG's January 2026 publication Scaling AI Requires New Processes, Not Just New Tools:

• 10% — Algorithms. Model selection, fine-tuning, prompts, RAG. What AI labs and technical boutiques sell.
• 20% — Technology and data. Integration, pipelines, vendor-stack management, governance infrastructure. What Big Four firms sell at scale.
• 70% — People and processes. Workflow architecture, decision rights, incentive structure, governance, operational measurement. What determines whether AI compounds or sits unused.

A buyer can spend 80% of their AI budget on the 30% of effort that drives value. McKinsey's 2025 report confirms it from another angle: out of 25 organizational attributes tested for correlation with EBIT impact, fundamental workflow redesign was the single highest-correlated. Only 21% of organizations had done it.

McKinsey's 6% cohort: what high performers actually do.

The high-performer cohort was:

• 3.6× more likely than other organizations to pursue transformative change with AI.
• More likely to have fundamentally redesigned workflows for AI economics.
• More likely to track well-defined KPIs for AI deployment.
• More likely to have senior leadership directly accountable for AI outcomes.
• More likely to make AI deployment a board-level discussion.

None of these are technology variables. All of them are operating-model variables. The other 94% are not making mistakes that better algorithms would fix — they are running operating models built before AI economics existed, with AI bolted on top.

Why most consulting firms cannot sell the seventy percent.

Their economics are wrong for it. Big firms sell projects, not outcomes. Boutique firms are paid by the vendors they recommend. Neither has skin in the game on the result.

The 70% does not scale with people. It requires senior judgment applied to specific buyer contexts; it cannot be templated, delegated to associate teams, or sold as a fixed-scope deliverable with confidence — because the work depends on what the operating-model audit reveals, unknown at the time of sale. (Mezura's answer: productize the diagnostic — fixed scope, firm price — and keep the rebuild scoped to what the diagnostic actually surfaces.)

The seventy percent requires a structurally different model: senior-only delivery, no vendor partnership economics, skin in the game on outcomes the firm has moved before, and selectivity by design. That model is uncommon. It is what Mezura is.

What working on the seventy percent looks like in practice.

In one 2024 engagement, a YC-backed career-services company moved application-to-interview conversion from 0.5% to 2.3% — a 4.6× lift, no new product, no new team — by rebuilding four layers of the operating model in a ten-week sprint, using AI-driven workflow automation to strip out the manual drag. Input quality. Decision protocols. Automation handoffs. Incentive alignment. The technology was not the bottleneck. The operating model around it was.

Read the full case study →  ·  See how Mezura's methodology runs →

One more thing about timing.

The seventy percent is not an AI problem. It is an operating-model problem that AI exposes and amplifies. The operating models that produce the McKinsey 6% cohort would produce above-market performance with or without AI. AI is the multiplier; the operating model is the base. A buyer who waits for AI to stabilize before fixing the operating model is waiting on the wrong variable.