McKinsey's GenAI economic potential estimate is real and grounded in methodology. The way it gets used in boardrooms strips out the conditions that make it meaningful. As an IS researcher, that gap bothers me.
I was reading the McKinsey State of AI 2025 report this spring and one data point kept pulling me back. Only 7 percent of organizations have fully scaled AI across their operations. That figure sits in the same research program as McKinsey's estimate that generative AI has the potential to add between $2.6 trillion and $4.4 trillion in annual value across the global economy. I keep thinking about those two numbers together, because they are measuring very different things and almost nobody uses them together in the same sentence.
The $4.4 trillion figure is real and it is grounded in real methodology. McKinsey's analysis looked at use cases across industries, estimated the impact of generative AI on the tasks within those use cases, and calculated what that task-level impact translates to at economic scale. The use cases cover customer operations, marketing and sales, software engineering, and research and development. The numbers are large partly because the use cases are broad and partly because the underlying economy they are drawing from is enormous. If AI meaningfully accelerates even a modest fraction of global knowledge work output, the aggregate total reaches trillions quickly. That is not a rhetorical trick. It is arithmetic applied to a large base.
What the number represents is a ceiling estimate of value capture if adoption reaches full scale, if organizations successfully redesign their processes around AI capabilities, and if workers redirect the time freed by automation toward productive new activities. Each of those conditions is non-trivial. McKinsey's own research on AI adoption finds that 88 percent of organizations report using AI in some form, but only 7 percent have fully scaled it. The potential estimate and the scaling reality live in the same firm's research, and they are measuring completely different things. The potential estimate asks what is theoretically achievable. The 7 percent figure describes where organizations actually are right now.
The automation framing is where I think the number gets most systematically misread. McKinsey estimates that 60 to 70 percent of current work activities could theoretically be automated using existing or near-term technology. The same analysis projects that the midpoint for half of today's work activities being automated sits around 2045. These are very different timeframes than what you hear when the $4.4 trillion headline gets deployed to justify a quarterly AI investment roadmap. The potential is real. The timeline is two decades long. The realization of that potential depends on adoption curves, workforce adjustment, regulatory environments, and organizational change dynamics that no one can forecast with precision. The economic potential analysis is a what-if exercise. It is not a when-will-it-happen forecast.
The distinction between automating tasks and automating jobs is where I think IS researchers need to be careful, because it matters for how we frame the research questions. The McKinsey analysis is explicitly about work activities, not jobs, and that difference is substantial. Almost every job contains activities that are theoretically automatable. Most jobs also involve judgment, relationship management, exception handling, and contextual improvisation that is much harder to automate than any individual task within it. When McKinsey says 60 to 70 percent of work activities could be automated, it does not mean 60 to 70 percent of jobs will disappear. It means that parts of most jobs could theoretically be done differently or faster. What happens to the overall job depends on what organizations and workers do with the time those automatable activities no longer require, and that is an organizational design question, not a technology question.
The IS theory I want to apply here is Melville, Kraemer, and Gurbaxani's 2004 integrative IT value model. In that framework, IT value reaches organizational performance through business process performance, mediated by complementary organizational resources and moderated by the competitive environment. The $4.4 trillion potential assumes all of those mediators and moderators work out favorably. Complementary resources need to be in place: the right data, the right processes, the right skills, the right structures. Business processes need to actually change, not just incorporate a new tool into an existing workflow. The competitive environment needs to allow firms to capture value rather than passing all productivity gains to customers in the form of lower prices, which is what happens in competitive markets. Brynjolfsson's productivity paradox showed exactly this dynamic for earlier waves of IT investment: the macroeconomic numbers lagged for years after the technology adoption because the complementary organizational investments took time. Every link in Melville's chain is an assumption embedded in the potential estimate.
For the organizations that are using the $4.4 trillion figure to justify AI budgets right now, I think the honest translation is this: there is a very large amount of potential value, realized value will be a fraction of potential value, and the interesting question is not whether the potential is real but what determines which organizations capture it. The gap between potential and realized is where absorptive capacity, process redesign, governance design, and skill development all do their work. I have been reading Cohen and Levinthal (1990) on absorptive capacity alongside the McKinsey data, and the fit is uncomfortable. Organizations cannot absorb the value of generative AI without the prior knowledge base to recognize what is actually valuable, assimilate how it works, and apply it to real problems. Building that knowledge base takes time and deliberate investment. Announcing a $4.4 trillion addressable market does not build it.
McKinsey's labor productivity estimate is the number I think gets least attention and deserves more. They estimate AI-driven productivity increases of 0.1 to 0.6 percent annually through 2040. That is a real and meaningful economic effect. It is also, when you say it out loud next to the $4.4 trillion headline, a fairly modest annual increment relative to the size of the opportunity. The productivity paradox that Brynjolfsson documented for the IT wave of the 1980s and 1990s showed that aggregate productivity gains from technology investment are slower, more distributed, and more conditional than the technology's potential suggests. My read is that the generative AI story will follow a similar pattern: real value, unevenly distributed, slower to appear in aggregate than in specific high-implementation firms, and heavily dependent on complementary organizational investments that are harder and less glamorous than the technology purchase itself.
What surprises me, honestly, is how rarely the 7 percent scaling figure gets used to contextualize the $4.4 trillion number in public conversation. These are not contradictory findings. They are describing the same phenomenon at two different levels of analysis. If only 7 percent of organizations have fully scaled AI, and the economic potential of AI runs into the trillions, then the research question worth pursuing is not whether the potential is real. It is what separates the 7 percent from the 93 percent, and whether the gap is narrowing fast enough to matter at the macroeconomic scale the headline number implies.
---
claims_checked:
- "GenAI potential value $2.6T to $4.4T annually across industries": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"
- "60-70% of current work activities could theoretically be automated": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"
- "Half of today's work activities automated with midpoint around 2045": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"
- "88% of organizations report using AI; only 7% have fully scaled it": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"
- "Labor productivity increase 0.1-0.6% annually through 2040": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"
claims_unverified:
- "The exact number of use cases McKinsey analyzed (often cited as 63) is widely reported in summaries but I have not independently verified the exact count from the primary report page; not referenced in this post body"
sources_used:
- "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"
word_count: 1050
About the author
Share
More notes
Related notes