Forty per cent of jobs are exposed to AI. Exposure is not the same as loss.

The International Monetary Fund estimates that almost 40 per cent of jobs worldwide are exposed to AI. The exposure is highest in the richest economies, at about 60 per cent. The IMF gives the reason: earlier waves of automation mostly affected routine manual work, but AI can reach high-skilled cognitive work as well. That is the kind of work advanced economies depend on most.

AI exposure by type of economy

Share of employment exposed to AI, per cent

Exposure is not the same as job loss

The IMF's more useful finding is how exposure divides. In advanced economies, roughly half of exposed jobs are likely to benefit: AI complements the work and raises productivity. In the other half, AI can perform key tasks now done by people. That can mean lower demand for labour, lower wages and less hiring. Two roles with the same exposure score can end up in very different places. What decides it is whether AI is used to augment the person or to replace the task.

National studies show the same range of outcomes. In India, a NITI Aayog roadmap published in 2025 finds that the technology and customer service workforce could lose about 1.5 million jobs by 2031. With the right skilling it could instead gain up to 4 million. The spread of possible outcomes is as wide as the risk.

What employers should do now

• Redesign at task level, not headcount level. Break roles into tasks and identify which ones AI can do. Then decide deliberately what the released time is for. A headcount target set from the top repeats the mistakes we described in our note on cost programmes.
• Choose augmentation where it creates value. The half of exposed jobs that gain from AI do not gain automatically. They gain when the process is rebuilt around the combination of person and tool.
• Build new roles before retiring old ones. Staff roles in AI operations, evaluation and domain-specific model work with people who already understand the business.
• Fund reskilling like a capital programme. It needs named cohorts, measured outcomes and its own budget line. A learning-platform licence on its own is not enough.

Exposure is a property of the task. Whether it becomes a gain or a loss is a management decision.

← Back to all insights