The Evidence: Three Estimates, One Direction
Goldman Sachs, McKinsey, and MIT researchers have measured the same phenomenon using different methods. Their three estimates diverge sharply.
A 2023 Goldman Sachs paper projects a productivity increase of 1.5 percentage points over the following decade. McKinsey places gains at up to 3.4 percentage points by 2040. MIT researchers are more cautious, projecting a 0.53% increase by 2034, as reported by CNBC.
The distance between 0.53% and 3.4% is where the leadership challenge lives. That gap depends on the conditions organizations build, before it depends on the tools they purchase.
These figures describe scenarios rather than certainties. Organizations that read the range as a margin for action, rather than a fixed forecast, gain a competitive advantage.
The Infrastructure Phase Explains the Delay
The AI boom, measured from the launch of ChatGPT in November 2022, is nearly four years old. It remains largely a phase of hardware and infrastructure capacity.
Ben Snider, Goldman's chief U.S. equity strategist, writes that the earnings impact of AI adoption will become clearer over the coming quarters. Investors have rewarded companies involved in building the infrastructure, given their visible near-term earnings impact.
Investors prefer to avoid speculation about which companies will implement AI most effectively. This caution creates space for those who build capability before the market prices it in.
Enterprise-level productivity gains are slow to arrive at scale. This delay has an organizational cause, as well as a technological one.
The Bottleneck Is Leadership Readiness
The true constraint on enterprise AI adoption lies with those who commission, govern, and evaluate deployments. Available data indicates that roughly 7% of leaders describe themselves as ready to lead these programs.
That number belongs in the boardroom. It concerns executives first, before the people using the tools day to day.
An organization with excellent AI tools and unprepared leadership will underperform one with mediocre tools and AI-literate leadership. The gap between the 7% of ready leaders and the broader population adopting tools describes two distinct populations with limited overlap.
The board should read this figure as a structural signal. A workforce absorbing AI tools faster than the leadership governing them generates friction, waste, and compliance risk.
The 56% Salary Premium Is the Most Actionable Data Point
The labor market already prices AI skills. The documented salary premium for these skills is around 56%.
This is a present-tense measure, not a forecast. Organizations that delay building internal capability are accumulating a skills debt. That debt becomes an acquisition cost within eighteen months.
For the CFO, the read is straightforward: investment in people development has a documented ROI, derived from the price the market places on these skills. Delaying the investment shifts the cost from the training budget to the recruiting budget, with a multiplier effect.
Converting Internal Talent Beats External Recruiting
Adoption data reveals a consistent pattern. Morgan Stanley reports a 98% adoption rate; JPMorgan reports approximately 200,000 opt-ins.
People already inside the organization bring context, relationships, and process knowledge. They adopt the tools and outperform newly hired AI-specialized employees brought in from outside in isolation.
Converting existing talent, rather than competing exclusively in the external market, produces faster results. This approach also reduces exposure to the 56% salary premium.
For the CHRO, the L&D priority becomes clear. Building internal AI learning pathways activates a talent pool that the external market values at a premium, and keeps process knowledge inside the organization.
Botsitting Is a Design Problem
Some organizations are recording roughly 6.4 hours per week of passive AI output supervision. People monitor what the system produces, with minimal active interaction.
This phenomenon, botsitting, signals a poorly designed workflow. AI has been inserted into the process, while the process itself has remained unchanged.
The remedy belongs to change management, before technology. Redesigning the workflow transforms passive supervision into productive collaboration and frees up the hours currently being lost.
People adapt when leaders create the right conditions. Passive supervision reflects a missing condition, before it reflects individual resistance.
The Exceptions That Point the Way
Some companies have already converted adoption into measurable gains. Shares of medical platform Doximity rose after CEO Jeffrey Tangney reported search revenues of ten times the cost of the tool.
Goldman identifies productivity beneficiaries within the Russell 1000. The methodology cross-references labor intensity with AI sensitivity, and includes companies such as CoStar Group, Dollar Tree, and eBay.
The sectors with the highest labor intensity and AI sensitivity include software, professional services, finance, and biotech. These are the industries where organizational conditions weigh most heavily on outcomes.
MIT economist Daron Acemoglu warns that the concentration of AI tools among a small number of companies could slow adoption by small and medium-sized enterprises. The potential therefore also depends on market structure.
The Design Question for Leaders
The question for CEOs and CHROs is concrete. What conversation about organizational readiness reaches the board this quarter?
The Talent & Compensation Committee has a clear metric to monitor: the share of leaders ready to govern AI deployments. That percentage predicts returns on technology investments better than the number of software licenses activated.
Organizations that close the gap act on three levers: process redesign, internal talent conversion, and leadership AI literacy. The evidence indicates that these conditions, built deliberately, move gains from the territory of promise into the territory of results.
The response pathway exists, and it is documented. Deliberately designed conditions outperform technological luck, systematically.
This article was written by an AI editorial author with human oversight, in compliance with the transparency obligations of Regulation (EU) 2024/1689 (AI Act, Art. 50). Sources are linked in the text.
Article by VERA