Goldman Sachs Says India Faces Lower AI Job Risks, but Services Remain Exposed

Goldman Sachs says India faces less widespread AI job displacement than richer economies, despite clear risks for clerical work and technology-enabled services. Santanu Sengupta, the bank’s chief India economist, delivered that assessment on August 14, according to a labor risk report. The conclusion challenges predictions that generative AI will quickly erase jobs across India’s enormous workforce.

The reassuring headline comes with an important qualification. India’s lower exposure reflects the structure of its labor market, not immunity from increasingly capable AI systems. Agriculture, construction, manufacturing, and other physical occupations still employ millions of people whose daily tasks remain difficult to automate with software.

The pressure is concentrated elsewhere. Routine clerical work, business process services, telecommunications, financial services, media, and some professional roles contain more tasks that software can perform. These sectors also support India’s position as a major exporter of technology and business services.

That creates the central tension. The country appears less vulnerable at the national level, while parts of its internationally important services economy sit close to the technology’s impact zone. India’s AI labor story is therefore about an uneven transition, not a simple contest between mass unemployment and uninterrupted growth.

Goldman’s Forecast Points to Transformation, Not Mass Replacement

The bank’s central argument is that AI will change more Indian jobs than it eliminates.

A Goldman Sachs analysis released in late July estimated that generative AI can perform between 9% and 17% of tasks across India’s non-agricultural workforce. That range measures task exposure, which describes work that AI can assist or automate. It does not represent a forecast that the same percentage of workers will lose their jobs.

The distinction matters because occupations bundle many different tasks. A financial analyst might use AI to summarize documents while retaining responsibility for judgment, client communication, and regulatory decisions. A teacher might generate lesson materials faster without transferring classroom management or student assessment to a model.

Goldman estimated that between 42% and 48% of non-agricultural jobs are more likely to receive AI assistance than face direct replacement. It placed between 8% and 12% at substitution risk, according to a detailed employment analysis. The remainder should experience relatively limited direct effects under the report’s assumptions.

Those estimates support Sengupta’s broader claim. India has a smaller share of workers in highly exposed office occupations than advanced economies. A large agricultural workforce further reduces the countrywide exposure rate because current generative AI primarily operates through language, images, data, and software.

Goldman expects adoption to lift India’s annual labor productivity growth by about 0.4 percentage points over a decade. Its estimated range runs from 0.1 to 0.8 percentage points, depending partly on how capable future systems become. Those gains are projections, not observed outcomes.

Productivity can rise when workers complete existing tasks faster, deliver more output, or move toward higher-value responsibilities. However, employers can also capture efficiency gains by reducing headcount. The same technology can complement workers in one organization and replace similar positions in another.

Implementation costs will shape that decision. Businesses must connect models to internal data, redesign workflows, verify outputs, manage security, and assign accountability. A demonstration that completes one task does not automatically become a dependable production system.

This helps explain why occupation-level exposure rarely maps directly onto immediate job losses. Companies usually automate portions of a workflow before restructuring entire positions. Hiring can weaken before layoffs become visible, especially when employers stop replacing departing workers.

India’s scale also slows any nationwide transition. Large enterprises, small businesses, farms, factories, public agencies, and informal employers adopt technology at different speeds. Access to computing, reliable electricity, useful data, and trained managers remains uneven.

Goldman’s forecast is best read as a map of pressure points. It suggests where tasks can change and where productivity might improve. It does not establish how quickly employers will deploy AI or how they will distribute the resulting gains.

India’s Labor Structure Provides a Buffer

India looks less exposed because much of its workforce performs physical, agricultural, or interpersonal work that generative AI cannot directly execute.

An International Monetary Fund assessment found that around 43% of India’s labor force worked in agriculture. That employment structure makes the country less exposed than economies that moved further toward office-based services. Software can advise a farmer, but it cannot independently complete most physical work in a field.

The IMF estimated that 26% of Indian workers held occupations with high AI exposure. It classified 14% of employment as highly exposed but potentially complemented by AI. Another 12% sat in highly exposed occupations where displacement risks were greater, according to its India assessment.

These figures differ from Goldman’s ranges because the institutions use different datasets, definitions, and occupational classifications. One estimate may cover the entire workforce, while another examines non-agricultural employment or task shares. The variations reinforce why exposure statistics should not be treated as precise layoff forecasts.

Cross-country research points in the same general direction. Advanced economies employ larger proportions of managers, professionals, administrators, and other knowledge workers. Those occupations use the language and information-processing tasks that current AI systems handle most effectively.

The IMF has estimated that about 60% of jobs in advanced economies are exposed to AI. Its corresponding estimates were 40% for emerging markets and 26% for low-income countries. Exposure includes opportunities for productivity growth as well as risks of displacement.

In Asia, approximately half of employment in advanced economies is exposed, compared with about one-quarter in emerging and developing economies. The regional comparison also found that richer economies possess more jobs where AI can complement workers. Higher exposure therefore creates both greater risk and greater potential upside.

India’s lower exposure can still conceal serious distributional problems. Workers in agriculture may be protected from direct generative AI substitution yet remain vulnerable to low productivity, unstable income, and limited social protection. Being outside AI’s immediate reach does not guarantee a secure or well-paid job.

The same issue appears within cities. A software developer and a delivery worker participate in the digital economy, but their exposure differs sharply. A model can produce code or documentation, while it cannot drive a motorcycle through traffic or physically hand a parcel to a customer.

Women and highly educated workers often have greater occupational exposure because they are more represented in administrative and professional roles. Higher-income workers also appear frequently in exposed occupations. Their work sometimes offers greater scope for augmentation, although routine components can still be automated.

Young graduates face a separate concern. Entry-level roles often contain documentation, research, scheduling, testing, and basic analysis. These tasks give new workers experience, but they also fit the capabilities of generative AI tools.

A company does not need to dismiss existing employees to reduce those opportunities. It can maintain a smaller intake class because experienced workers produce more with AI assistance. That pattern would affect career entry without appearing immediately as mass unemployment.

India’s buffer is therefore real but incomplete. The country has fewer exposed jobs as a share of total employment, yet the exposed portion includes strategic industries and common pathways into the middle class. National averages cannot capture that concentration.

The Services Engine Sits Closest to the Risk

India’s most exposed workers are concentrated in services that helped the country become a global destination for outsourced business operations.

Routine IT-enabled business processes face greater substitution risk because their work is repetitive and can be represented digitally. Common examples include standardized support responses, document processing, basic reporting, transcription, data entry, and rule-based quality checks.

Clerical support positions face similar pressure. Generative AI can draft correspondence, categorize documents, extract information, create summaries, and answer routine questions. Systems connected to business software can also move information between applications with limited human involvement.

These capabilities do not eliminate every role. Models still make factual errors, mishandle unusual cases, and struggle with accountability. Organizations need people to define workflows, review sensitive outputs, resolve exceptions, and communicate with customers.

However, the number and composition of positions can change. One employee equipped with an effective system might complete work that previously required several people. Employers may then hire fewer junior workers while adding specialists in data, security, workflow design, and model evaluation.

India has substantial exposure to this transition because services are central to its growth model. Goldman Sachs has described the country’s IT and business services industry as a roughly $250 billion sector. It also reported that services exports reached about $387 billion in 2024.

Global capability centers deepen that connection. These operations allow multinational companies to place engineering, finance, analytics, customer support, and other functions in India. Goldman counted more than 1,700 such centers and projected further expansion through 2030.

The centers can create AI-related work because enterprises need teams to integrate models into real operations. Employees can develop evaluation methods, adapt systems to specialized domains, manage data access, and monitor regulatory compliance. India’s large pool of English-speaking technology workers provides a strong foundation for those activities.

The same centers can also standardize and automate existing work. A multinational firm can deploy one system across multiple offices after testing it in a centralized operation. Efficiency gains can reduce demand for repetitive services even when higher-value technical roles expand.

This creates a contest between upgrading and commoditization. India can move toward engineering, research, model deployment, and complex consulting. Alternatively, customers can use AI to perform more routine work internally, weakening the advantage of labor-cost differences.

Protectionism adds another complication. Restrictions on cross-border data, models, or digital services can limit how Indian teams support overseas clients. Divergent rules can also increase compliance costs and fragment systems across markets.

Goldman has argued that the effects of trade restrictions on India’s AI-led services exports remained limited in its earlier assessment. It still identified data localization, regulatory divergence, and geopolitical tension as risks. Those pressures sit outside any model’s technical ability.

Demand for AI skills offers one positive signal. Research based on Indian job advertisements documented rapid growth in postings that requested AI-related capabilities after 2016. It also found evidence that AI adoption can support demand for complementary non-AI roles.

Yet job advertisements represent employer intentions, not completed hires. Online postings also overrepresent formal, white-collar employers. They reveal where demand is moving but provide a limited view of India’s entire labor market.

The likely outcome is not the disappearance of services. It is a change in what clients consider valuable. Routine output becomes easier to produce, while domain knowledge, trusted execution, proprietary data, and responsibility become more important.

Indian service providers will face pressure to sell outcomes instead of labor hours. Billing models based mainly on headcount become harder to defend when software reduces the time required for each task. Companies must show that their teams deliver judgment, integration, and measurable business results.

Workers face the same shift personally. Knowing how to operate a general-purpose chatbot will not provide a lasting advantage. Employees need domain expertise and the ability to verify, refine, and apply AI-generated work in consequential settings.

That transition can create better jobs, but it does not happen automatically. Reskilling programs must connect training to actual employer needs. Otherwise, workers receive generic credentials while companies continue struggling to fill specialized positions.

Lower Exposure Does Not Mean Lower Economic Risk

India can avoid widespread direct displacement and still suffer if AI weakens entry-level hiring, services exports, or wage growth.

The first uncertainty concerns the gap between tasks and jobs. Exposure studies break occupations into activities that models can perform. Employers decide whether those activities justify redesigned roles, slower hiring, outsourcing changes, or workforce reductions.

Those decisions depend on reliability. A model that succeeds during a controlled test can fail when inputs become ambiguous or customers behave unpredictably. Errors in healthcare, banking, law, and government services carry consequences that prevent unrestricted automation.

Regulation can preserve human oversight, but it can also change where work occurs. Companies might centralize approved systems while reducing local teams. Alternatively, compliance requirements can generate new work in auditing, documentation, risk management, and security.

The International Labour Organization has stressed that exposure is not equivalent to job loss. Its global index found that one in four jobs had some exposure to generative AI. Transformation was more likely than full replacement, according to the occupational exposure index.

Transformation can still be painful. An employee may retain a job while losing bargaining power, autonomy, or opportunities for promotion. Digital monitoring can increase when companies use AI systems to measure output and enforce standardized processes.

Entry-level employment deserves particular attention. Junior workers often handle first drafts, basic research, simple code, meeting notes, and document review. Automating those tasks can remove the work through which people traditionally learn a profession.

Organizations then face a pipeline problem. They still need experienced employees, but they may train fewer beginners. Without deliberate apprenticeships, companies can save money today while creating future shortages of senior talent.

The second uncertainty concerns productivity distribution. Goldman’s projected gains describe additional output per worker. They do not determine whether employees receive higher wages, companies earn larger margins, customers pay less, or working hours decline.

Competitive markets can spread benefits through lower costs and new services. Concentrated markets can allow a smaller group of firms to retain more value. Workers with scarce skills may command higher pay, while employees in standardized roles face weaker leverage.

The third uncertainty involves infrastructure. Advanced models require computing capacity, reliable electricity, network access, secure data systems, and skilled implementation teams. Goldman has identified power, water, data centers, and computing resources as necessary investments.

India is expanding its AI infrastructure, including public initiatives under the IndiaAI Mission. However, infrastructure must reach beyond large technology companies. Smaller firms need affordable access and practical support if productivity gains are to spread across the economy.

The fourth uncertainty is whether India can produce more of the underlying technology. The country has strengths in deployment and engineering services, but it remains less prominent in advanced chips and foundational model research. Heavy dependence on foreign platforms can send part of the economic value abroad.

Goldman previously estimated that India’s AI sector can generate more than 2.3 million job openings by 2027. It also cited projections placing the available talent pool near 1.2 million. The gap indicates potential demand, but it does not guarantee that training programs will match real positions.

Businesses frequently request combinations of technical and industry knowledge. A bank needs people who understand both models and financial controls. A manufacturer needs engineers who can connect AI outputs with physical operations and safety requirements.

Generic training cannot close every mismatch. Workers need opportunities to practice with relevant data, tools, and workflows. Employers also need credible methods for evaluating skills that change faster than conventional degrees.

Knowledge workers can prepare by documenting decisions, developing domain expertise, and building reliable ways to retrieve prior work. A searchable personal knowledge base can support that process, especially when AI produces more drafts and information to review.

Still, individual preparation cannot replace labor policy. Governments and employers must improve transition support, education, and access to training. Workers cannot independently solve infrastructure shortages, weak social protection, or declining demand across an entire occupation.

The cautious conclusion is that lower aggregate exposure gives India time. It does not remove the obligation to use that time well. Poor preparation can turn a manageable transition into concentrated hardship.

Three Signals Will Show Whether Goldman’s Case Holds

The next phase should be judged through hiring, services demand, and measured productivity rather than headline claims about AI capability.

The first signal is entry-level hiring across IT services, business process operations, finance, media, and professional services. Graduate recruitment provides an early view of whether companies are augmenting existing teams or eliminating the bottom rung of career ladders.

A sustained decline in junior openings would weaken the reassuring interpretation of Goldman’s forecast. It would show displacement through missing jobs rather than highly visible layoffs. Stable or expanding intake, paired with new AI responsibilities, would support the augmentation case.

The composition of hiring matters as much as the total. Employers may recruit fewer generalists while seeking more workers with cybersecurity, data engineering, model evaluation, and industry-specific expertise. That shift would validate the expectation of transformation while creating an urgent training challenge.

The second signal is the performance of India’s technology and business-services exports. Revenue, contract size, client retention, and hiring across major providers can reveal whether AI expands demand or reduces labor-intensive outsourcing.

Growth in complex engineering and AI integration work would strengthen India’s position. Falling demand for routine processing, without offsetting high-value contracts, would expose the vulnerability hidden by the national employment average.

Investors should also examine how companies describe productivity. A provider can report more output per worker because it moved into valuable services, automated repetitive work, or reduced staffing. Those paths have very different consequences for employees and long-term competitiveness.

The third signal is measured productivity outside controlled demonstrations. Goldman’s baseline forecast anticipates an annual boost of about 0.4 percentage points over a decade. Real gains should eventually appear in output, delivery times, service quality, or business formation.

Productivity without broad adoption would provide weaker support. A few large companies can achieve substantial efficiencies while smaller firms fall behind. That outcome can increase concentration without transforming the wider economy.

Evidence of wage growth would make the case stronger. If workers become more productive but compensation remains flat, the benefits are not reaching labor. If wages rise alongside output and employment, augmentation becomes more convincing.

Policymakers should watch differences across gender, education, occupation, and region. A stable national unemployment rate can coexist with serious disruption in clerical centers or among recent graduates. Granular labor data will reveal those effects sooner than a single headline number.

The debate should also separate temporary adjustment from structural change. Employers can pause hiring while testing new systems, then resume once demand grows. A multiyear decline across exposed roles would indicate a deeper reorganization.

Goldman’s argument is plausible because India employs so many people in work that current generative AI cannot directly perform. Agriculture, construction, manufacturing, transport, and personal services provide a substantial statistical buffer.

Yet India’s economic ambitions depend on more than protecting existing jobs. The country wants higher productivity, stronger services exports, improved wages, and better opportunities for its young workforce. Success requires exposed sectors to create valuable new work faster than routine positions contract.

The most useful question is therefore not whether AI will take all Indian jobs. It will not affect every occupation equally, and exposure does not automatically become unemployment. The sharper question is whether businesses can turn automation into expansion rather than a narrower search for labor savings.

Over the coming months, watch graduate recruitment, high-value services contracts, and productivity evidence together. If all three improve, Goldman’s lower-risk forecast will look increasingly credible. If hiring and exports weaken while efficiency rises, India’s apparent protection will have concealed a more difficult transition.