AI is not hitting India’s offshored work all at once. It is hitting customer support, application maintenance and labour-billed back-office work first and hardest, with insurance and bank operations following, while some lines, such as airline back office and data labelling for AI itself, are still moving to India.

Across sectors the pattern so far is the same: few layoffs, slower hiring of freshers, and clients demanding the productivity gain as a lower price. Global firms are also shifting the human work that remains to India, so India’s share is rising even where the total is shrinking. The Indian government, meanwhile, has large backlogs of exactly this kind of work, in courts, claims, grievances and tax, and buys almost none of it the way foreign clients do.

How much to trust it. This piece pulls together nine sets of research notes compiled on 18 October 2026. Almost every figure comes from search-result summaries of filings, earnings calls, official releases and press reports; the original pages, including all Indian government sites, could not be opened. Company statements are quoted as reported. Several sector sizes rest on a single estimate and are marked. No source measures how many Indian jobs AI has actually removed, and this piece does not claim a number.

What other countries’ job reports say is most exposed

Official reports from several countries name the same jobs, and they are the jobs India took on through outsourcing: customer service, claims, clerical and data-entry work, bookkeeping, transcription and junior software work. Most of these reports measure exposure or project the future rather than count jobs lost. The two that measure outcomes, from the United States and South Korea, both find the damage concentrated among young workers in software and information services.

NASSCOM, the industry body, has published on AI talent shortages rather than on job losses. Its reports with Deloitte, BCG and Indeed count growing demand for AI engineers and say most employers expect roles to change. A widely shared figure of a “20-25% cut in entry-level IT roles per NASSCOM” could not be traced to any NASSCOM document and is not used here. The only official Indian estimate found is NITI Aayog’s October 2025 roadmap, written with NASSCOM and BCG, and press summaries of it disagree on whether the possible loss is 15 lakh or 20 lakh jobs.

Where it shows in India’s export numbers

The Reserve Bank’s annual survey splits software exports into computer services and IT-enabled services, most of which is BPO. In FY26 the first grew about 11% and the second about 2%, and BPO itself was almost flat at US$56 billion. That fits AI pressure on voice and back-office work, but it does not prove it: currency movements, pricing and survey revisions could explain part of it.

The wider balance-of-payments category of business services, which includes most capability-centre work, grew faster, from US$107.2 billion to US$124.2 billion, as this blog’s earlier look at Indian IT in the AI era found.

Sector by sector

Customer support is the clearest case. The big global contact-centre companies are shrinking or flat and say why. Concentrix cut its 2026 forecast in June and its shares fell about a fifth; its finance chief said revenue reflected “an accelerated deployment of AI for clients, as well as client decisions to reduce support for certain customer groups”. TTEC’s customer-care revenue fell 12.1% in the second quarter. Clients are also cutting their own support staff: Salesforce went from about 9,000 support staff to about 5,000 in 2025.

But the same companies report that clients move the human work that remains to cheaper locations. TP, the French firm that is one of India’s largest contact-centre employers, says growth in India “continued to accelerate” because of it, and now has more than 90,000 staff in the country. Gartner expects no Fortune 500 company to have done away with human customer service by 2028, and half of those planning big AI-led cuts to drop the plans by 2027.

In IT services, the pressure is on price and on fresher hiring. Broker estimates put application services at 40-70% of Indian IT firms’ revenue, and that is where coding assistants bite. Persistent’s chief executive told Reuters in August 2026 that clients want the same work for 25-30% less.

TCS made 25,000 campus offers for FY27, against about 44,000 freshers hired in FY26, though its human-resources chief has reportedly said it remains on course for about 40,000 a year; its chairman said in June that the company was heading towards having as many AI agents as employees and would “hire less”. Wipro set no fresher target for FY27; Infosys kept its at 20,000. Software testing, often described as the most exposed line, has no official headcount; one staffing firm’s estimate is about 3.75 lakh people, growing under 1% a year.

In finance, insurance and banking operations, labour-billed revenue is shrinking inside growing firms. EXL’s digital-operations revenue fell in both quarters of 2026 that it has reported, which management linked to AI being built into clients’ operations, while its total revenue grew 16%. Genpact’s chief executive said in May that headcount growth is “decoupling from revenue”; 48% of its revenue is no longer priced per person.

On the bank side, JPMorgan guided to 10% fewer operations staff and had about 4% fewer by February 2026, yet still plans to grow its 55,000-strong Indian workforce by 5-7% a year. Standard Chartered is the one bank found where reporting names Indian centres, in Chennai and Bengaluru, as the most affected by planned cuts of about 7,000 roles by 2030.

Some work is still arriving. American carriers are expanding Hyderabad centres, and Qantas is considering moving up to 1,000 roles to India. US accounting firms short of staff are adding Indians: RSM US plans to go from about 2,000 to 5,000. And AI itself creates work: NASSCOM counts about 70,000 data-labelling workers in India, although that number may be old, with the work dividing between low-paid volume labelling and much better-paid expert grading of AI output by doctors, lawyers and engineers.

How the firms are adapting

1. Selling AI as a service line. TCS reports an AI revenue run-rate of about US$3.1 billion, Infosys says AI work is 8.2% of revenue, and Genpact’s AI-led unit earns more than twice the company average per employee.
2. Pricing on outcomes rather than seats. Firstsource calls its model “UnBPO” and grew revenue 19.7% in FY26 with headcount almost flat. HCLTech’s deal with E.ON was reported to pay nothing in its first year, with fees tied to efficiency gains.
3. Buying their way in. Capgemini paid US$3.3 billion for WNS to build “agentic AI-powered intelligent operations”; EXL bought iMerit (the deal closed in the third quarter of 2026); publishing-services firms are buying analytics companies.
4. Using AI to defend Indian seats. TP invested in Sanas, whose software adjusts agents’ accents in real time, and deployed it in Indian centres serving foreign clients, an advantage against the Philippines rather than a threat to Indian jobs.
5. Hiring fewer, different people. Naukri’s September 2026 index shows IT-services postings down 4% and AI and machine-learning postings up 20%, with fresher postings up just 1%.

Fewer hires, not layoffs

Every source that measures outcomes points the same way. The New York Fed finds about 1% of US service firms laying people off because of AI but 12% of AI users hiring fewer. A Dallas Fed analysis finds the fall in young workers in exposed jobs comes from fewer new entrants rather than dismissals. The Bank of Korea finds youth employment in AI-exposed sectors falling while jobs for people in their fifties in the same sectors rise.

In India, the large IT firms say they will hire fewer freshers, not cut staff. For a country whose advantage was a supply of trainable graduates, that is the most important channel: the first rung of the ladder is narrowing before anyone is fired.

How the other offshoring hubs see it

India’s rivals tell the same story in public, of an “AI-enabled workforce” climbing a value ladder, but their numbers differ. Central Europe is meeting AI together with rising wages, and part of what it loses comes to India. The Philippines, whose product is mostly voice support, is the hub where AI hits the core business, and it is the only one that has cut its own jobs target.

Three differences matter for India. First, Kraków’s losses are partly India’s gains. The cuts there are concentrated in transactional finance work, and Bloomberg and Polish press report several companies moving it to India; Heineken (up to 700 roles to Hyderabad) and Electrolux (up to 200 accounting roles) were reported, not confirmed.

Polish centres have already moved up: more than half of their processes are mid-office and only about a quarter of staff are entry-level. Second, that makes India the destination for the most automatable work. Work moving from Kraków to India is the accounts-payable and reconciliation work that AI automates fastest, so the inflow buys time rather than security. Third, India has not done what the Philippines did. IBPAP publicly cut its headcount target and moved its yardstick from jobs to value per worker. NASSCOM has published no equivalent revision, and India has no official forecast of its own export workforce to revise.

The comparison has limits. Definitions differ between associations, the Polish growth figures do not reconcile exactly between ABSL’s two reports, the latest Bulgarian headcount is from 2022, and no credible like-for-like wage comparison between Poland and India was found.

The same work, waiting at home

The Indian state processes, at home, much of the kind of work its companies sell abroad: grievances, claims, document review, transcription and translation. It mostly does this through its own staff, through automation, or not at all, leaving backlogs.

The provident-fund numbers are the warning. EPFO settled 38% more claims in FY26 than in FY25 and settled 71% of advance claims automatically, with no reported rise in staff. Where work is rules-based and the data is clean, software absorbs the volume and creates few jobs.

The jobs are in the parts software cannot finish alone: translating High Court and district court judgments, most of which have not been translated; checking crop-yield data behind delayed insurance claims; reconciling land records in the villages where they are still not integrated; reviewing the tax returns and health claims that automated checks flag; and answering grievances in the citizen’s own language.

That is the “AI plus human review” model Indian firms now sell to American banks and insurers, and Bhashini, the government’s language platform, already provides much of the translation layer. What is missing is a buyer that pays for outcomes.

No credible estimate exists of how many jobs such contracts would create, and this piece does not offer one. The e-Courts budget is a sign of the scale of current ambition: of Rs 7,210 crore, Rs 53.57 crore, about 0.7%, is earmarked for AI.

What is not known

1. India’s job split. NASSCOM has not published, in any source found, a recent split of its roughly 60 lakh workforce into customer support, back-office, testing and other lines. The last found split is from about 2021.
2. Jobs lost. No source measures Indian jobs removed by AI. The figures in circulation are exposure estimates, projections or company statements.
3. Where cuts land. Salesforce, Klarna, Meta and the banks rarely say how many affected roles were in India or at Indian vendors.
4. Price versus volume. Contracts renegotiated at lower prices are described, but no large renewal has been disclosed with its price.

What would fix this

• MeitY and NASSCOM should publish an annual headcount split by service line (customer support, finance and accounting, testing, infrastructure, data labelling), so that the effect of AI can be measured instead of guessed.
• The labour ministry should add an occupation-and-age table for IT and business services to the Periodic Labour Force Survey, so that India can run the kind of analysis that found the youth decline in Korea and the United States.
• The Centre and states should tender court translation, claims review, land-record reconciliation and grievance handling as outcome-priced managed services in Indian languages, with human review, the way foreign banks and insurers buy them from Indian firms, and the way the tax department bought its processing centre.
• A larger share of the e-Courts budget should go to translation and transcription with human post-editing, starting with High Court judgments.
• Link these contracts to apprenticeships, so that public work replaces some of the fresher hiring that the export firms are cutting.

These are proposals from this blog, not recommendations any agency has adopted.

Sources and caveats

About this article: Researched, written and edited by Umashankar Triplicane Dwarakanathan, with AI research assistance; every figure is meant to trace to the primary source cited. See the Editorial Policy for how sourcing, AI use and corrections work.