Case Study 1: McKinsey's AI ROI Research — What Separates AI Leaders from Laggards

The Survey

Every year since 2017, McKinsey & Company has conducted a global survey on the state of AI adoption in business. The survey — formally titled The State of AI — is one of the largest and most methodologically rigorous studies of its kind, typically surveying 1,400 to 1,800 respondents from companies across industries, geographies, and size categories. The respondents are senior executives and functional leaders who are directly involved in AI strategy and implementation.

The survey does not measure what companies say they will do with AI. It measures what they are actually doing: which AI capabilities they have deployed, how they are organized for AI, how much they are investing, and — critically — whether they are generating measurable financial returns.

Over the years, the survey has produced one of the most robust longitudinal datasets on AI ROI in existence. And its central finding is both clear and uncomfortable: the distribution of AI value is highly skewed. A relatively small number of organizations capture the majority of value from AI, while the majority of organizations report modest or negligible financial returns.

Understanding what separates the top performers from the rest is the most practical question a business leader can ask about AI strategy. This case study examines McKinsey's findings across multiple survey years to identify the practices that correlate with superior AI ROI.

The Distribution Problem

The 2023 and 2024 surveys revealed a stark distribution:

• About 25 percent of respondents reported that AI contributed more than 20 percent of their EBIT (earnings before interest and taxes). This group — the "AI leaders" or "AI high performers" — had advanced significantly from 2021, when only 15 percent of respondents reported this level of impact.
• The remaining 75 percent reported AI contributions below 20 percent of EBIT, with a significant proportion (roughly 30 to 40 percent) reporting that AI's financial impact was difficult to measure or negligible.

This is not a normal distribution with a few outliers. It is a power-law distribution: a small number of organizations are capturing enormous value, and the long tail includes many organizations that have invested meaningfully in AI but cannot demonstrate clear returns.

The gap is widening, not closing. In 2020, the revenue gap between AI leaders and laggards was approximately 1.5x. By 2024, Accenture's complementary research estimated the gap at 2.5x in revenue growth. Companies that figure out how to generate value from AI pull further ahead; companies that do not fall further behind.

Business Insight: The skewed distribution of AI value is the single most important finding in AI ROI research. It means that simply investing in AI is not sufficient. The differentiator is not whether you invest but how you invest — the practices, processes, and organizational structures that translate AI investment into business value.

What AI Leaders Do Differently

McKinsey's research identifies several practices that consistently distinguish high performers from the rest. These practices are not technological — they are organizational, strategic, and operational.

1. They Scale Beyond Pilots

The most common pattern among underperforming AI programs is what McKinsey calls "pilot purgatory" — a cycle of proof-of-concept projects that demonstrate technical feasibility but never reach production deployment or enterprise scale.

AI leaders have broken this cycle. The 2023 survey found that high performers had deployed AI at scale (embedded in core business processes, affecting a significant portion of revenue or operations) in an average of 4 or more business functions, compared to 1 or fewer for laggards. The distinction is not the number of experiments — it is the number of experiments that become operational systems.

Ravi's portfolio review at Athena illustrates this. The portfolio includes twelve active projects, but the value — the $22.8 million in measurable annual impact — comes from the four projects that reached production scale. The experiments and in-development projects consume resources and generate learning, but they do not generate financial returns until they are deployed and integrated into business operations.

2. They Invest in Enabling Infrastructure

High performers invest disproportionately in the "invisible" layers of AI: data infrastructure, MLOps platforms, feature stores, and monitoring systems. These investments do not produce direct revenue. They reduce the cost and time of every subsequent AI project.

McKinsey's data shows that AI leaders allocate 20 to 30 percent of their AI budget to data and infrastructure, compared to 5 to 15 percent for laggards. This investment creates the option value discussed in Section 34.5 — the ability to deploy new AI capabilities quickly because the foundation already exists.

The implication for ROI measurement is significant. If you measure AI ROI only at the project level, infrastructure investments will always look expensive and unproductive. You need platform-level and program-level ROI measurement to capture their value.

3. They Embed AI in Business Processes

A model sitting in a notebook is not an AI deployment. A model integrated into a CRM system that automatically triggers retention offers — that is a deployment. The difference is not technical; it is organizational.

AI leaders ensure that AI outputs are connected to business actions through clear workflows, defined decision rights, and trained end users. McKinsey found that organizations where "AI recommendations are integrated into day-to-day workflows" were 2.4 times more likely to report significant financial impact from AI.

This echoes a theme from Chapter 6: the business of machine learning is the business first. Tom's pricing engine achieved 94 percent accuracy but generated zero value because it was not integrated into the business process it was meant to serve. High performers avoid this trap by involving business process owners from the earliest stages of AI development.

4. They Practice Active Portfolio Management

AI leaders do not simply launch projects and hope for the best. They actively manage their AI portfolio — prioritizing, accelerating, and killing projects based on ongoing assessment of value, risk, and strategic alignment.

The 2024 survey found that high performers were three times more likely to have a formal AI portfolio review process (at least quarterly) compared to laggards. They were also twice as likely to have killed at least one AI project in the past year — a counterintuitive finding that demonstrates the discipline of cutting losses and reallocating resources.

Ravi's quarterly review process — assessing each project on technical progress, business engagement, and assumption validity — is a textbook example of the practice McKinsey's research validates.

5. They Measure and Communicate ROI Rigorously

Perhaps the most important finding: high performers have established robust methods for measuring and communicating AI ROI. This includes:

• Pre-defined success metrics tied to business outcomes (not just model metrics)
• Formal attribution methodologies (A/B testing, before/after analysis, or modeling-based attribution)
• Regular reporting to senior leadership in business language (revenue, cost savings, risk reduction) rather than technical language (accuracy, F1 score, AUC)
• Honest treatment of uncertainty, including confidence intervals and scenario analysis

Organizations with formal AI ROI measurement processes were 2.1 times more likely to increase their AI budgets — not because they inflated their numbers, but because they could demonstrate value credibly to financial decision-makers.

Business Insight: The measurement paradox: organizations that invest in rigorous ROI measurement spend more on the measurement process, but they also get more AI funding — because they can prove the funding is justified. The ROI of measuring ROI is remarkably high.

6. They Invest in AI Talent and Organizational Capabilities

AI leaders invest not just in hiring AI specialists but in building AI literacy across the organization. The 2024 survey found that high performers were four times more likely to have provided AI training to more than 20 percent of their workforce.

This investment in organizational capability connects directly to the adoption and process integration findings above. An organization where business users understand what AI can and cannot do is an organization where AI projects are more likely to be properly scoped, effectively integrated, and enthusiastically adopted.

The Talent Paradox

McKinsey's surveys consistently identify talent as the most frequently cited barrier to AI adoption. But the talent challenge is more nuanced than "we can't hire enough data scientists."

The most critical talent gaps, according to high performers, are not in data science but in:

1. AI translators — people who can bridge the gap between technical AI teams and business stakeholders (the role NK is growing into)
2. ML engineers — people who can take models from notebooks to production (the gap between experimentation and deployment)
3. Data engineers — people who can build and maintain the data pipelines that feed AI systems (the least glamorous and most essential role)

AI leaders address these gaps through a combination of external hiring, internal training, and partnerships. They also compensate AI roles competitively — McKinsey found that organizations offering above-market compensation for AI roles were 1.5 times more likely to be high performers, likely because they could attract and retain the scarce talent that makes AI programs successful.

The Risk Factor

High performers do not avoid risk. They manage it differently.

The 2024 survey found that AI leaders were more likely to report having encountered AI-related risks (bias, security vulnerabilities, regulatory challenges) than laggards. This is not because their AI systems are riskier — it is because they are more actively monitoring for risks and identifying them before they become crises.

AI leaders are also more likely to have formal AI governance structures — risk assessment processes, ethical review boards, model monitoring systems, and incident response plans. These governance investments are part of the TCO that Section 34.9 describes, and they are a significant reason why AI leaders' TCO is higher on an absolute basis but lower as a percentage of value created.

Lessons for Athena and Beyond

Professor Okonkwo uses the McKinsey findings to frame a discussion for the class.

"The most important takeaway," she says, "is that AI ROI is not primarily a technology problem. The technology is increasingly commoditized — the same algorithms, the same cloud platforms, the same open-source libraries are available to everyone. What differentiates leaders from laggards is execution: how they organize for AI, how they measure value, how they manage the portfolio, and how they build organizational capability."

She poses a question: "Ravi's portfolio review — the one with the $22.8 million in measurable value, the two killed projects, and the three accelerated projects — which of McKinsey's six practices does it demonstrate?"

NK answers: "All six. They've scaled beyond pilots — four projects in production. They've invested in enabling infrastructure — the data platform. They've embedded AI in business processes — the churn model feeds directly into the retention workflow. They actively manage the portfolio — the quarterly review. They measure and communicate ROI rigorously — Ravi's presentation to the board with methodology, uncertainty, and honest treatment of what they can and can't measure. And they've invested in organizational capability — Ravi's team has grown from three people to fifteen, and business teams are increasingly data-literate."

"Correct," Professor Okonkwo says. "And notice something else. Ravi's presentation did not claim that AI was transforming the company. He claimed — with evidence — that specific AI projects were generating specific, measurable value. That is a much stronger claim, precisely because it is more modest."

What the Research Does Not Tell Us

McKinsey's research is valuable but imperfect. Several important caveats apply:

Selection bias. The survey respondents are disproportionately large companies with formal AI programs. Smaller companies and organizations just beginning their AI journey are underrepresented. The findings may not generalize to early-stage AI adopters.

Self-reporting. All data is self-reported. Companies may overstate their AI maturity, overestimate their financial returns, or selectively report successful projects. There is no independent verification of the claimed financial impacts.

Correlation, not causation. The practices associated with high performers may not cause high performance. Companies with strong management, generous budgets, and competent teams may naturally adopt good AI practices and generate strong returns — with the good practices being a consequence rather than a cause of organizational excellence.

Survivorship bias. Failed AI programs are less likely to be represented in the survey. The "laggard" category in the survey includes organizations that are still trying — not organizations that have abandoned AI entirely.

Despite these limitations, the McKinsey data provides the most comprehensive available picture of what separates AI leaders from the rest. The practices it identifies — scaling beyond pilots, investing in infrastructure, embedding AI in processes, managing the portfolio, measuring ROI, and building capability — are actionable, verifiable, and consistent with the framework developed in this chapter.

Discussion Questions

1. McKinsey's data shows that about 25 percent of organizations capture the majority of AI value. Is this distribution likely to become more concentrated (a "winner-take-all" dynamic) or more distributed over time? What factors would drive each outcome?

2. Why might companies that actively kill AI projects achieve higher overall portfolio returns than companies that let all projects continue? Connect your answer to concepts from portfolio management in Section 34.8.

3. If the most critical AI talent gaps are in "translator" roles (people who bridge business and technology) rather than data science, what are the implications for MBA education? What skills from this course are most relevant to the translator role?

4. How would you validate the causal claim that the six practices identified by McKinsey cause higher AI ROI, rather than merely correlating with it? Design a study that would provide stronger evidence.

This case study is based on publicly available research from McKinsey & Company's annual State of AI surveys (2017-2024), supplemented by findings from Accenture, MIT Sloan Management Review, and BCG. All specific figures are sourced from published survey reports and should be verified against the original sources for precision.