1. Employees work faster with AI, but employers don't see a corresponding lift in revenue or profit. This dominated an Oct. 30 summit in Singapore, where enthusiasm was unmistakable. A survey by the Federal Reserve Bank of Atlanta, Bank of England, Bundesbank, and others, covering nearly 9,000 executives in seven countries, found 78% used AI, yet executives estimated it added only about 0.2 percentage point annually to labor-productivity growth over three years. [para. 1][para. 2][para. 3]
2. The problem isn't that AI fails to deliver, but that gains at the desk don't automatically become gains on the income statement. Managers must ask where the time goes. Evidence of individual speed is compelling: Brynjolfsson's study of 5,000 customer-service agents showed AI increased issues resolved per hour by ~14% (34% for newcomers). Consultants completed tasks 25% faster with 40% higher quality (Harvard/BCG). SMU's index across 2,000 occupations estimates 40-50% time savings, but these are technical potentials. [para. 4][para. 5][para. 6][para. 7]
3. The gap stems from bottlenecks. A Microsoft experiment with 7,137 knowledge workers at 66 companies found frequent users (10%) spent 3.6 fewer hours on email, but meeting time and output didn't change. An employee can answer email faster alone, but shortening a meeting requires colleagues to change. Kremer's O-ring theory explains that a weak link undermines the whole. Doubling coding speed doesn't help if testing or approval lags. An executive bottleneck is common: AI saves 65% of time reviewing financial info but only 10% making decisions, so employees wait. [para. 8][para. 9][para. 10][para. 11][para. 12]
4. The remedy is organizational redesign, not another software subscription. Paul David's electrification analogy: electric motors appeared in the 1880s, but productivity gains came decades later when factories rearranged layouts. AI can also repair weak links: a 2024 Procter & Gamble experiment with 776 employees showed individuals with AI matched two-person teams without it, and helped technical and marketing employees incorporate each other's considerations. [para. 13][para. 14]
5. Turning capacity into growth is a further hurdle. Saving labor only lowers costs if staffing or hours fall; it doesn't create demand. Freed capacity has three uses: produce more (easy to copy, drives prices down), improve quality, or develop new offerings (differentiation). Tania Babina's research on U.S. public firms found AI investment drove faster sales and employment growth via product innovation, not lower costs, with returns after 2-3 years. [para. 15][para. 16][para. 17][para. 18]
6. Small firms face disadvantages: Singapore's SME adoption is ~23% vs ~70% for large firms, with less money for experiments and less room for failure. Governments can help by publishing credible evaluations and promoting low-cost standard solutions, as Singapore does with subsidies; China's stakes are greater. The right test has two parts: where did the saved time go, and can customers tell the difference? As AI tools become ubiquitous, access isn't an advantage—the scarce capability is reorganizing work. Companies can buy software, not management. [para. 19][para. 20][para. 21][para. 22][para. 23]
AI generated, for reference only