World Bank: Only 7% of South Asian Jobs Face High AI Substitution

Only 7% of jobs in South Asia combine high exposure to AI with low complementarity, the profile most open to substitution. The average for other developing regions is 15%. The World Bank's October 2026 South Asia Economic Update, published on October 6, adds the first firm survey of AI use in India and new data on outsourcing links with US companies.

Adoption is half the US rate, and shallower

[Fact] The World Bank Enterprise Surveys AI module interviewed 1,355 formal Indian firms and 401 US firms with at least five employees in 2025-26. 23.4% of Indian firms use some form of AI, against 42.7% in the United States. On an index that also counts paid software, advanced tools, tasks, business functions and sourcing, India averages 0.27 and the US 0.85.

[Fact] Only 3.5% of Indian firms pay for AI, against 23.1% of US firms. [Estimate] If every paying firm is also a user, about 15% of Indian AI users pay for it, compared with about 54% in the US. Most Indian use runs on free online platforms.

[Fact] Indian adoption was near zero in mid-2024 and has accelerated since early 2025, while US growth slowed. Asked about productivity over three years, Indian firms expect an average gain of 3% from faster AI development and US firms 1.3%. Almost three-quarters of Indian firms expect close to nothing, 8% expect 20% or more, and 15% expect a decline of up to 10%.

Outsourcing links grew where AI exposure was highest

[Fact] Since ChatGPT's release, a one-standard-deviation increase in a US firm's industry AI exposure is associated with a 27.5% larger rise in its customer and partner links with South Asian firms. For customer links, the figure was 63.1% when the South Asian counterparties were themselves above the median for exposure, and 7.4% when they were below it. [Estimate] That is a gap of roughly 8.5 times.

[Fact] Earlier World Bank work found that Indian suppliers of highly exposed services cut hiring sharply after ChatGPT, mostly where foreign buyers were also highly exposed. [Claim] The authors read the new link data as showing that growth in demand outweighs reshoring. One candidate mechanism they name is new work adjacent to old outsourcing, such as expert human review of AI-generated software and text used to fine-tune models.

Hiring is shifting toward complementary skills

[Fact] Firms whose workforces were highly exposed slowed their hiring after ChatGPT and shifted postings toward jobs that are less exposed and more complementary to AI. Postings that ask for AI skills carry a wage premium of nearly 30%. In South Asia 15% of workers hold jobs with both high exposure and high complementarity. [Estimate] That is about two such jobs for every substitutable one.

Teachers: the AI app did not beat the basics

[Fact] A randomized trial in Quetta, Pakistan, ran a six-week winter school for 4,938 students in grades 2 to 5. An app generated lesson plans, built and graded assessments, and coached teachers from classroom audio. The teacher-led arm raised test scores by 0.10 standard deviations, a significant effect. The AI-assisted arm raised them by 0.05, not distinguishable from zero. Average attendance was 26.3%.

What to prepare

• Customer service representatives, computer programmers and data entry keyers in outsourcing. [Estimate] These match the report's examples of customer support, routine software development and back-office processing. Review and quality checking of AI output is the adjacent work the report names.
• Software developers and analysts. Listing AI skills on applications targets the 30% premium measured in postings.
• Elementary school teachers. In Quetta the tool added nothing measurable where students did not show up. Attendance and targeted instruction came first.

Limits

The link analysis counts relationships, not their value, applies US-based exposure measures to South Asian firms, and may oversample large exporters. The firm survey covers India only within South Asia, and the teaching trial results are preliminary. The hiring-cut finding comes from a separate World Bank study that was summarized here from this report rather than read in full.

Sources

• World Bank, "Adopting AI for Growth", South Asia Economic Update (October 2026), October 6, 2026, CC BY 3.0 IGO: https://documents.worldbank.org/curated/en/099552210132636511

About this analysis

This article was written with AI assistance, from the full text of the report (Executive Summary and Chapter 2). Figures marked [Estimate] (the 15% and 54% paying shares, the 8.5-times gap, the two-to-one ratio, the occupation mapping) are our own arithmetic or reading of the report's numbers. This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank.