Most companies are now experimenting with AI, yet clear productivity gains remain uneven. AI delivers when four capabilities develop together. A 2024 OECD working paper finds adoption limited and patchy, and describes AI as a production technology that bundles skills, software, data and infrastructure. Each part depends on the others, so if one is weak, returns fall and risks rise.
Data and infrastructure come first. AI depends on high-quality data, connected systems and computing infrastructure that can scale. Many firms still run on fragmented data, old systems and manual handoffs, so their AI projects spend most of the budget cleaning data and connecting systems. Rollouts slow down, each new use costs more, and people trust the results less.
The second capability is models tuned to the business. General AI models are widely available, so the advantage comes from tuning them to a firm’s own products, customers and decisions. The OECD notes that the same core technology can produce very different results depending on how it is trained and deployed. The Productivity Institute finds that banking and retail gain most when AI is aimed at specific performance measures, such as credit loss rates, inventory turns and customer conversion. Generic proofs of concept rarely reach the profit and loss statement.
AI changes how people work, so the workforce and operating model are the third capability. Research from MIT shows that many US manufacturers suffer a short-term productivity dip after adopting industrial AI. They outperform their peers later, once they rewire processes and management practices. Older, more complex firms see the biggest early setbacks, because of rigid routines and deep hierarchies. This capability means deciding before rollout how roles will change, providing targeted training and naming an owner for human oversight of AI.
Leadership focus is the fourth. The Productivity Institute notes that many published studies find positive results, but strong long-term evidence is still limited, which makes disciplined capital allocation essential. Focused leaders pick one or two areas where AI has to cut cost, lift margins or speed up work, then line up capital, risk and governance behind them. They also hold course through the adjustment phase, when early results are messy.
