It has also had to contend with being consistently wrong about the size of the market.

The 2020 report (PDF) opened with this number:

The Indian e-retail market is primed to reach nearly 300 to 350 million shoppers over the next five years—propelling the online Gross Merchandise Value (GMV) to $100 to 120 billion by 2025.

Calendar year 2025, per Bain’s own 2026 report, ended at $65-66 billion. This week’s report does not acknowledge the forecast made six years ago.

The intellectual foundation of the series is a comparison to China. In 2020, Bain argued that India’s e-retail market looked a lot like China’s had about eight years earlier — before Alibaba and JD.com and cheap smartphones turned it into the world’s largest online shopping market. Data prices in India had just crashed in the same way. Internet users were multiplying rapidly. So the analogy was not unreasonable.

Except, China’s e-retail market was approximately $1.2 trillion in 2019. But despite all the learnings, the 2026 report continues to use China as its primary benchmark.

Each edition also names the disruptions that will define the next chapter of Indian e-commerce. Some of what Bain got wrong here, to be fair, was genuinely unforeseeable.

In 2020, Bain was excited about voice and vernacular — the idea that India’s next hundred million shoppers would arrive not through typed searches but through apps in Hindi, Tamil and Telugu. Back during those days, Google India was reporting 270% year-on-year growth in voice search, and three platforms had each crossed 50 million users:

Several vernacular apps like ShareChat, TikTok and Helo have garnered upwards of 50 million users each.

Shortly after Bain published its maiden India e-commerce report, the Indian government banned TikTok and Helo. BulBul.tv, an Indian live video shopping startup that the same report identified as an emerging model, has since shut down.

The 2021 report cited a second live commerce startup, Simsim, as evidence the model was taking hold in India, describing “9-10x growth in user base over the last year.” Simsim was acquired by YouTube in 2021 and shut down by 2022 — a sad detail worth noting given that even a platform with YouTube’s distribution and resources could not make live commerce work in India.

The 2021 report described voice and vernacular as “vital to win new shoppers.” The 2022 report cited “5x growth in voice-search users and 3x growth in vernacular-search users over the past year.” The 2023 report said “about 25%-30% of first-time shoppers used these offerings.”

The 2025 (PDF) and 2026 reports skip voice and vernacular.

Social commerce got the series’ most precisely quantified forecast. The model Bain had in mind was Pinduoduo, a Chinese platform that built hundreds of millions of users by letting people bargain in groups and rope in friends for discounts. India’s demographics seemed to suit it: young, price-sensitive, deeply networked, spending hours daily on social media. From the 2021 report:

Social commerce GMV (approximately $1.5-2 billion in FY20) could grow at a 55%-60% compound annual growth rate over FY20-25, with the potential to empower 40 million small businesses and turbocharge women entrepreneurs.

At those growth rates, the 2025 social commerce market should have been somewhere between $13 billion and $21 billion.

The 2025 and 2026 reports do not mention social commerce.

The 2021 report also made a prediction that has received less attention than it probably deserves. E-retail, it said, would outgrow India’s modern trade sector — the organised offline retail industry of supermarkets, hypermarket chains and branded stores — by FY26. Modern trade in India is estimated at somewhere around $60-80 billion, a relatively small slice of the total retail market given how much Indian retail still runs through unorganised neighbourhood stores. India’s e-retail market in 2026 will land at somewhere around $78-80 billion. Bain got this right.

Brand aggregators — firms that acquire and scale digitally-native consumer brands, a model that briefly swept global e-commerce on the back of the Thrasio playbook in the U.S. — made their debut in the 2022 report. “Brand aggregators are helping turbocharge these brands,” it said, devoting a section to the model and its potential in India. By 2023, Thrasio had filed for bankruptcy in the U.S. and Indian aggregators were facing similar headwinds. The 2023 report does not mention brand aggregators, and neither do any subsequent editions.

ONDC — the Indian government’s attempt to build open plumbing for e-commerce, modelled on how UPI created a shared payments infrastructure that transformed digital transactions in India — made its debut in the 2022 report:

ONDC aims to create an interoperable network for digital commerce akin to Unified Payment Interface (UPI) network for payments. Although a nascent initiative, its successful and full-scale execution could provide a fillip to the Indian e-retail ecosystem.

ONDC has since struggled to gain meaningful traction, both in the real world and in recent Bain reports.

Quick commerce is where the story runs differently. The model — ordering groceries or household essentials on your phone and receiving them in under ten minutes, from a small warehouse tucked into your neighbourhood — looked shaky in 2022. Global players were burning cash, and Bain was measured:

Long-term sustainability of the model (especially <15-minute model) is unproven, especially in the absence of delivery and/or convenience fees.

In India’s largest cities, it turned out to work. Blinkit, Zepto, Instamart and Flipkart Minutes have become part of the daily texture of urban life in Mumbai, Delhi and Bengaluru in a way that quick commerce has managed almost nowhere else in the world. Bain is now declaring India as having “emerged at the forefront of Q-commerce globally,” projecting the market at $65-70 billion by 2030 — roughly equal to the entire current e-retail market.

In the 2025 report, hyper-value commerce — very cheap goods sold through platforms targeting price-sensitive shoppers in smaller Indian cities — was named as one of three forces reshaping Indian e-retail, having grown from 5% of e-retail GMV in 2021 to more than 12% in 2024. The 2025 report cited Temu’s rapid rise in the U.S. as the global template for the model. Temu is now facing severe U.S. tariff exposure following the elimination of the de minimis exemption and has begun pulling back significantly from the American market. In Bain’s new report, hyper-value commerce does not make a cameo. Its place in the framework is taken by conversational commerce, AI assistants reshaping how people discover and buy products online.

To be sure, Bain has raised some concerns over the years. The 2025 report contained a reality check that sat awkwardly alongside its optimistic recovery forecast. Right next to its projections for resumed e-retail growth, it noted that female real earnings in India were approximately 20% lower in FY24 compared to pre-Covid levels, with male real earnings about 3% lower. These are the people the market needs to convert into online shoppers spending more per order.

From the beginning, Bain has also cautioned that India needs a higher per capita GDP for much of the e-commerce story to fly. The $4,000 GDP per capita inflection point is the series’ most reliably rediscovered insight. The argument is that when a country’s average income crosses roughly $4,000 a year, online discretionary spending accelerates sharply — a pattern observed in China and Indonesia that India is expected to replicate around or by 2030.

Nobody knows whether Indian e-retail can sustain the new projected growth rate and the market has surprised forecasters before. The earlier targets, though, have all passed without being reached, and each time one has, Bain has moved the horizon forward without acknowledging the miss.

Paytm sold its 7.2% stake in PayPay to SoftBank’s Vision Fund 2 for $279M net in December 2024, part of a broader divestment of non-core assets following the Reserve Bank of India’s crackdown on Paytm Payments Bank earlier that year.

Paytm held the stake not as equity but as Stock Acquisition Rights, and had to pay an exercise cost to convert those rights into actual shares before selling them, which reduced its net proceeds. The transaction valued PayPay at $7.06B (PDF). Fifteen months later, PayPay debuted on the Nasdaq at an opening market cap of $12.7B, putting the gross implied value of the same stake at $915M.

The 400 million Indians who were online that year have since grown to over 950 million, per IAMAI and Kantar, and one in five of them now navigates the internet by voice. That number, and what it represents in accent, dialect, language-switching and real-world deployment scale, is what could finally position India at the centre of an AI boom it has largely watched from the outside.

India has mostly been written off in the AI conversation the past three years. Its equity markets sat out the global rally, its indices dominated by banks and consumption companies with no stake in the GPU buildout reshaping markets in the US, Taiwan, South Korea and Japan. Indian IT services employs roughly 5.8 million people and generates an outsized share of corporate India’s profits. When Anthropic released tools that could automate back-office work and modernise COBOL code, Indian IT stocks lost nearly 15% in a month and TCS fell below a market cap it had not seen since 2020. That disruption may have been priced in accurately, but what the market was not pricing was where India sits in the layer of AI that is growing fastest.

Nearly $380 billion in investment pledges have been made toward India’s AI ecosystem this year. Of that, roughly $260 billion is allocated to energy and $115 billion to data centres and networks. The AI model layer has $1.5 billion assigned to it. Another $1.5 billion is earmarked for applications. India generates nearly 20% of the world’s data but holds only 3% of global data centre capacity. The private market is beginning to make a different bet on where value will actually accumulate.

The mismatch that matters most for voice AI is not between languages but mediums. English accounts for 49.5% of websites with a known content language and Hindi for under 0.1%. Yet 870 million Indians accessed the internet in Indic languages in 2024, and among Indic-language internet activity the dominant uses were watching videos and listening to music, not typing queries into websites. The global training corpus is missing not only the language but the way India actually uses the internet.

India’s BPM sector employs 1.65 million workers in call centres, payroll and data handling and accounts for 52% of the global outsourcing market. The country has been running voice-saturated enterprise workflows at scale for two decades, and that history is also a speech-data history.

The most concrete evidence of what that means commercially came from Bajaj Finance’s Q3 FY26 earnings call. Managing director Rajeev Jain told analysts that AI had listened to 20 million customer calls, converted voice to text and generated 100,000 new loan offers for customers the company previously lacked data on. Disbursements through AI-powered call centres came in at $172 million, 10% of the company’s total disbursals that quarter. Jain called it “just our first attempt” and said the company expects to process 100 million AI-powered calls in the coming year. Bajaj Finance did not build a new product but extracted value that had already existed, unread, in two decades of recorded customer conversations.

UPI 123PAY lets India’s roughly 400 million feature-phone users make payments through IVR calls and missed calls without internet access, in 12 Indian languages. BHASHINI has been wired into Indian Railways for multilingual voice assistants across public platforms. BHASHINI supports about three dozen languages across more than 1,600 AI models and processes millions of inferences daily. The government’s CPGRAMS grievance system now accepts complaints by voice in 22 regional languages.

A speaker’s accent, their habit of switching between Hindi and English mid-sentence and the way a language sounds when it is someone’s only tongue rather than a second one learned in school do not survive into a text corpus. Researchers who built a Hindi speech recognition benchmark drawing on 132 speakers from 83 districts found that mainstream commercial and open-source systems broke down badly across India’s accent regions, and worse still on names, places and domain-specific words. Hindi in one part of the country sounds different from Hindi in another part, and the training data that exists does not reflect either very well.

Hinglish, the Hindi-English mixing that characterises hundreds of millions of everyday conversations, produces roughly 42% word error rate on baseline monolingual-trained models. Publicly available code-switched datasets remain scarce. Small amounts of dialectal fine-tuning can outperform much larger quantities of standard-language data.

Building speech systems that actually work in India requires collecting data the way field researchers do, district by district, speaker by speaker. AI4Bharat’s speech effort spans over 400 districts and covers 300,000 hours of raw speech, 6,000 hours of transcribed data and 6,400 hours of mined audio-text pairs. Its IndicVoices dataset drew spontaneous and conversational speech from 51,000 speakers across 400 districts in 22 languages, released under an open license for research and commercial use.

Meesho launched Vaani, a generative AI voice shopping assistant aimed at tier-2 and tier-3 users who find typing-based e-commerce unintuitive, and within its first month 1.5 million users had tried it, conversion rates ran 22% higher than regular browsing and return rates fell. In India, where shopping has always been conversational, the assistant replicates what a kirana shopkeeper does — listens, asks follow-up questions and guides the buyer to a decision.

Sarvam dubbed the Union Budget 2026 live in Kannada and Hindi on Republic TV, the first national budget to be dubbed live using AI. The Prime Minister’s monthly address is dubbed every month into 11 Indian languages using Sarvam’s system while preserving the speaker’s voice. In India, dubbing is how content crosses the language market at all.

ElevenLabs closed 2025 with $330 million in annual recurring revenue and raised $500 million at an $11 billion valuation in February 2026. India is already its second-largest enterprise market globally, ahead of markets it has operated in far longer. Usage in India surged 50% in the three months ending January 2026, and the company has set an India revenue target of $100 million, deploying its technology with enterprises including Meesho, TVS Motor, Mahindra and PocketFM.

Every company building the next generation of voice AI needs training data in the languages users actually speak, across the accents, dialects and code-switching patterns of real conversations. That data does not exist in any web corpus. It only accumulates inside actual deployments, and India has more of those deployments, running at greater scale and in more linguistic variety, than anywhere else.

Megh Singh was producing a version of it in 2017, a Hindi voice query on a railway station’s free Wi-Fi that trained nothing and went nowhere. The infrastructure to catch what came next has since been assembled without design: in call centres logging two decades of customer conversations, in a research lab collecting speech from 51,000 speakers across 400 districts, in payment and grievance systems processing hundreds of millions of voice interactions a year.

India was written off because the AI boom was about text, chips and data centres. The boom in voice AI is about something else entirely, and India has more of it than anywhere else.

The value of gold held by Indian households nearly doubled in under a year, rising from $2.8 trillion in March 2025 to $4.8 trillion by January 2026. It now stands at 175% of the combined value of household deposits and equity.

This might not have much of an impact on consumption, however. Households earning under INR 500,000 annually hold nearly 70% of all household gold stock, and those households are unlikely to change spending or saving behaviour based on price appreciation alone, Kotak estimates.

Kotak modeled 20,000 quick commerce order baskets and found that more than half are priced below INR 500, a fat left tail driven by the fact that urban household spending puts a hard cap on how much quick commerce can expand. Only 52% of the consumption basket is addressable by quick commerce at all, the bank concluded.

The model’s central finding is that basket value is more sensitive to supplier pricing — specifically discounts to MRP — than to the number of items per order. Hence, the path to better unit economics runs through harder supplier negotiations rather than nudging customers to buy more.

HDFC Bank’s stock has fallen roughly 25% year-to-date and is trading at a price-to-book ratio of around 2×, the lowest level recorded since at least 1999.

An elevated loan-to-deposit ratio carried over from its merger with HDFC Limited, unexplained compression in loan yields, and stalling CASA market share gains despite the bank adding thousands of branches since FY22 have all weighed on investor confidence. Deposit growth and asset quality remain ahead of peers, but management has issued guidance across so many metrics simultaneously — loan growth, LDR, NIM, CASA — that investors have no single benchmark against which to measure a good quarter.

E-commerce accounted for 6% of urban India’s FMCG sales, and up to 14% across all metros, in Q4 2025, with traditional retail doing the rest, and rural volume growth ran at 2.9% against 2.3% in urban areas, the eighth consecutive quarter of outperformance, per NielsenIQ. That gap is being won and lost at the wholesale and distribution layer, by companies that make their products the easiest to find, easiest to trust and easiest to replenish.

"Affordable premium" in rural India is a constrained idea for exactly this reason. The room that exists is to take a brand the consumer already trusts and move it up one price step, Parle-G to butter cookie to cashew cookie, soap bar to liquid dispenser, at the same INR 5, INR 10 or INR 20 they are already comfortable spending. One or two new products, not ten, because the upgrade has to travel through the same shops, carry the same familiar name and ask for roughly the same money.

HSBC estimates that 40 to 45% of Hindustan Unilever’s revenues come from what it calls slow-moving segments: tea, soaps, Horlicks and Glow and Lovely, all high-penetration categories with very little upgrade path left, and the company has not built a single new organic brand at scale in a decade. Britannia has had its non-biscuit portfolio stuck at roughly 22 to 25% of revenues for three to four years, and per HSBC, its share of e-commerce including quick commerce is still in the high-single digits. Balaji keeps building new distribution into underpenetrated markets; HUL and Britannia are, in different ways, companies whose existing distribution has run out of road.

Rural distribution means the stockist and the kirana; urban distribution means the telco bundle, the ranking algorithm, the society gate, the dark store, but whoever sits between a product and repeated usage in any of these forms ends up controlling the demand.

Home services platform Urban Company has built its business around a constraint that has no physical shelf. Home services are incident-driven, a broken fan, a blocked drain, a facial before a wedding, and every one of those situations is its own separate demand context. Advertising bathroom cleaning to someone who needs a massage does not work, and running targeted performance ads for appliance repair does not work either because the need does not exist until the appliance breaks, so the company has focused mostly on making sure people have heard of it, not on chasing them.

In value-commerce, who sits closest to the moment of demand has separated Meesho from its rivals more cleanly than product or price. Nearly 199 million people bought something on Meesho over the fiscal year that ended last March, orders grew 36%, and it has cemented its position as the largest free cash flow generator among scaled e-commerce platforms in India.

Its shoppers in smaller cities and towns had come to online shopping through WhatsApp and UPI, starting with INR 100 or INR 200 purchases, spending cautiously, and only moving to platforms with bigger or more expensive selections after several transactions and often more than a year of experience. Meesho got there first, was reliable enough that people came back, and the habit formed.

On Meesho, new products surface quickly, the feed feels fresh, and sellers send more inventory because the fees are simple enough to make the math work. Others run on legacy infrastructure in a way that new, unproven products end up sitting behind ones that have already built up reviews and search history.

Telecom operator Jio has 503 million mobile subscribers; rival Airtel has 349 million, and both, along with cable operators Tata Play and Dish TV, have bundled 20 to 25 streaming services into their existing plans, creating a distribution layer for OTT content that individual apps cannot match on their own.

In smaller cities and towns there are already more people watching OTT content through these bundles than through direct subscriptions, and Airtel’s bundle, which includes Netflix, JioHotstar, Zee5 and SonyLiv among 25 or more platforms, starts at INR 279 a month against a stated standalone value of INR 750. A streaming service that has spent years trying to convince people to sign up directly can be lapped by one that simply gets added to a phone plan the subscriber already pays for.

Payments company PhonePe, used by hundreds of millions of users, did not try to build a full travel booking product but has added a thin layer that lets users book flights or buses without leaving the payments app, enough to capture the transaction, not so much that it weighs down what the app is actually for.

Apartment society software startup Mygate started as a tool for managing visitor entry in gated housing societies, logging who comes and goes and sending residents a notification when a delivery arrives. It turned out that most of the people walking through those gates every day are delivery workers, from Swiggy, Zomato, Blinkit, Flipkart and Amazon. That has made Mygate’s actual user base, the residents of those societies, one of the more precisely targetable audiences in urban India: affluent households with cars, dual incomes and someone ordering delivery almost every day, which is why quick commerce and e-commerce companies have been buying ads on Mygate at scale.

Devotional content platform AppsForBharat found the same thing in a different context. It has built apps for devotional content, bhajans, temple livestreams, religious calendars, and its decision to build more temple and content partnerships in north India created a business concentrated in the north, because users came wherever the supply of relevant content already was. The distribution, not the product, created the geography.

A new air-conditioner brand can price 5% below Voltas and still lose, because someone spending INR 35,000 to INR 40,000 wants to know the company will send a technician if something goes wrong two years later, and that confidence comes from the brand being present in enough stores that it feels established.

LG Electronics India, the market leader across refrigerators, washing machines and inverter air conditioners, runs more than 1,000 service centres staffed by 13,300 certified engineers, supported by 777 brand shops and over 30,000 sub-dealers, and that infrastructure is what a consumer is really buying when they pick LG over a brand they have never had to call.

LG has also launched what it calls an essential series in tier-2 and smaller cities, refrigerators and washing machines at lower price points with basic-plus features, targeting first-time buyers who are aspirational but not yet ready for the flagship range. It is the consumer durables version of Parle-G to butter cookie.

Logistics platform Porter has built its business around a specific kind of customer: the trader or small business owner who needs to move goods across a city on short notice, without knowing in the morning how many vehicles they will need by afternoon. The value is not the truck but in the certainty of having one when the need arrives.

Food delivery platform Zomato’s bestselling categories, biryani and pizza, work because they can survive a 30-minute delivery, fill a meal on their own, and are ordered from enough restaurants that the platform can consistently fulfil the demand. Its rapid-delivery format Bolt failed because restaurants could not cook fast enough to meet the promise. Bistro, Zomato’s office-hours snacking service, works in locations where there are enough office workers nearby, enough delivery riders who are free during the day, and a small company-run kitchen already in a dark store. Take away any one of those and the economics seem to fall apart.

India still rewards whoever is easiest to find, easiest to trust and easiest to replenish. The product question usually comes after.

Food is India’s largest retail category at $600–640 billion, and e-commerce has barely touched it — just 1-2% penetration. Mobiles, at $40-45 billion, are a fraction of that size but nearly half already transact online, leaving the category most associated with Indian e-commerce the least room to grow.

UPI’s transaction volume growth has more than halved — from 82% in FY2023-24 to 42% in FY2024-25 — and is projected to slow further to 25% this fiscal year, prompting India’s parliamentary finance committee to question whether the world’s largest real-time payments network should continue to depend on government subsidies to stay solvent.

After nearly a decade of an unchallenged duopoly over India’s ride-hailing market, Uber and Ola now have fewer monthly active users combined than Rapido.

IRDAI hasn’t announced anything yet, but recent language in the Insurance Act has revived a power the regulator has always technically had — the authority to dictate how much intermediaries can earn. The regulator ran product-level caps until a few years ago, setting specific limits on what agents and brokers could be paid on each type of policy. It then relaxed the regime, capping overall expenses rather than individual commissions, which gave insurers more flexibility in how they distributed their spending. The new statutory language means the harder power is live again, and people close to the industry describe a “new urgency” from the government around getting more Indians insured.

Insurers across the industry paid out $11.7 billion in commissions in the financial year that ended last March, a figure that has been growing at a 16% compound annual rate over the past decade, and private life insurer commissions alone compounded at 25%. In India, commissions are high and hence premiums are high, so the policymakers want to squeeze one so the other falls.

A cap would compress what intermediaries earn, but it would also reshape who holds bargaining power between the insurer and the distributor. In India, where finding and convincing customers to buy insurance is genuinely difficult, the distributor has historically had the upper hand.

Most of what the Indian insurance industry sells as life insurance is really a savings product with an insurance wrapper — a way for households to park money, earn returns and get a tax break while carrying some sort of death cover (however thin). Term life, where you pay a small premium and your family gets a large payout if you die, and standalone health insurance that covers hospital bills together account for only about 6-8% of the volume mix, and the rest is credit life and group insurance.

Tens of millions of Indian families have no meaningful financial cushion against death, disability or serious illness. If policymakers want “more protection, not more savings wrappers,” the cost of getting those products to people is the obvious target.

But protection in India isn’t bought the way it is in wealthier countries where people go looking for coverage. It’s sold through agents and call centres where a human being walks a reluctant and skeptical customer through a product they’d never seek out on their own, explains the terms and handles the paperwork. A commission squeeze works in countries where insurance is a pull product.

If protection is sold, not bought, the commission debate stops being about cutting fat. The government may be about to defund the only distribution engine that can push protection into Indian households.

For the largest online insurance broker, call-centre operations — the teams that phone potential customers, explain policies and close sales — run to roughly 40% of core revenue. Performance marketing, the digital advertising that generates leads for those call centres, adds another 15% — and those are direct costs, cuttable if commission compression stays shallow. Indirect costs — offline marketing, salaries outside the call centre, corporate overhead — run to another 20%, and industry executives insist these “may not be quickly reduced.”

One scenario analysis by Bernstein that is making the rounds models three levels of commission cuts. A shallow cut produces a 6% revenue hit and roughly 10% profit hit in the first year; a moderate cut takes those figures to 18% and 33%; and a deep cut reaches 30% and 61% — even assuming the broker immediately slashes costs in response. The market seems to have priced in deferral or shallow cuts, not deeper ones.

In 2018, SEBI, the securities market regulator, imposed a sweeping reduction in the fees that mutual fund companies could charge investors. That in turn compressed what mutual fund distributors — the agents and platforms that sold funds to retail customers — could earn. It was a consistent, pan-industry action that left distributors no choice but to absorb the reset and adjust. The insurance commission pool is arguably the largest fee income pool for distributors in Indian financial services, and a consistent, pan-industry action from IRDAI would put insurance distributors in the same position.

There’s also the question of when commissions get paid, not just how much. Intermediaries currently get a large share of their commission upfront, in the first year of the policy, even though the insurer’s economics depend on the customer continuing to pay premiums for years. A deferral regime would shrink the upfront payout and push more into renewal years, tying the intermediary’s income to whether the customer actually keeps the policy. The agent or broker would then have reason to sell the right product to the right person rather than whatever pays the highest first-year commission.

The intermediary’s costs, though — staff, call centres, marketing, the work of actually finding and onboarding the customer — all get spent at the point of sale. If the income from that sale gets pushed into future years, the intermediary may owe taxes on earnings it hasn’t yet received, because the commission is recognised as income even if the cash hasn’t arrived. If the cash inflows is pushed out, the pressure isn’t just on margins but on working capital — the day-to-day cash a business needs to keep operating.

Some scenarios being modelled say that if cuts go deep enough, the call-centre model of assisted selling stops working entirely. For the largest online broker, that model is described as “critical” to selling health and term life — the exact products policymakers say they want more Indians to own. The distribution channel that has been reaching consumers who would never have bought protection otherwise can come to a halt if the economics changes.

Agent productivity in the traditional agency channel — the individual agents who sell policies door to door or through personal networks — is also heavily skewed. Most of the business comes from a small percentage of productive agents who are likely to stay committed even with lower payouts. The long unproductive tail that could become more disengaged wasn’t generating much volume to begin with. The deeper vulnerability is the call-centre model, where costs are fixed and high, and where the economics of selling protection to people who don’t want it depends on upfront commissions covering upfront acquisition costs.

Some investors and other industry executives say insurers might eventually get nudged to build their own web and call-centre sales engines “alongside” the intermediary ecosystem — doing the selling themselves rather than outsourcing it to brokers and agents. A commission policy dressed up as consumer protection could end up pushing the industry to bring distribution back in-house. If the economics for intermediaries shrink and protection still has to be actively sold to people who don’t want it, the insurer might be the only one left who can afford to do the selling.

If insurers build out direct channels, bargaining power shifts their way; if intermediaries figure out how to sell efficiently even with restructured payouts, it shifts back. The balance ahead of the draft regulations appears tilted in favour of insurers — especially large, disciplined ones who stand to gain from lower distribution costs if implementation is strict.

One model floating around PB Fintech, the parent company of Policybazaar, goes in a different direction — toward profit-sharing with insurers, curated product shelves where only vetted policies are offered to customers, and delegated authority where the platform takes on underwriting and claims work that insurers normally do themselves. Other markets call this a “managing general agent” or MGA, an intermediary that does more than sell, helping design and run parts of the insurance operation, paid like a partner rather than a salesperson. (Non-banking financial companies, or NBFCs, grew into something similar in lending — origination-and-servicing machines that found borrowers, assessed their creditworthiness and managed loans, work that banks couldn’t or wouldn’t do themselves, especially in smaller cities and rural areas.)

If commission levels become politically sensitive and regulators want every rupee paid to intermediaries justified, the survivors will likely be the ones who can argue performance, not volume.

India’s logistics market has consolidated to the point where two delivery companies dominate supply and two e-commerce giants dominate demand. That tight structure keeps pricing rational — but the same consolidation that creates the opportunity also creates the fragility, as Shadowfax’s business hinges almost entirely on the strategic decisions of a handful of clients.

Monthly active users on merchant apps for Paytm and PhonePe have stagnated since late 2023, sitting at roughly 14 million and 15 million respectively, even as their combined registered merchant base reached ~47 million and is still growing at 10–15% annually.

Assuming ~75% of the combined installed device base of ~23 million are active, and an 80-90% overlap between the two platforms’ merchant bases, device penetration among registered merchants is already at 32-35%, Bernstein estimates.

Further reading:
Indian Payments’ Most Durable Revenue Line?
PhonePe’s $15 Billion IPO and the Cost of Building on Government Rails

Krafton’s PUBG was the most popular mobile game in India before the government banned it, and its return as Battlegrounds Mobile India (BGMI) has not restored that dominance.

Indian developers account for fewer than seven of the top 100 mobile games by monthly active users in the country, according to India Dispatch’s analysis of Sensor Tower data.

The Indian-made games that do appear in the top 100 are Ludo King, Indian Bikes Driving 3D, Indian Vehicles Simulator, Indian Theft Aura Simulator, Ludo Supreme Gold King, WinZO, and Truck Masters — all board games or simulators themed around Indian roads and vehicles.

No Indian studio has a game in the top 100 in battle royale, puzzle, or runner categories, which remain dominated by Garena, King, Voodoo, and others.

Barring the Russia-Ukraine war, no geopolitical conflict in the past six years has dented the Nifty beyond a week — most recovered within two.

The report has been picked over enough by now, but I want to dwell on its India thesis for a moment. Indian IT’s export model was built on the cost arbitrage of Indian developers relative to their American counterparts, the report says, and the marginal cost of an AI coding agent has collapsed to the cost of electricity, so that value proposition is dead.

In this scenario, TCS, Infosys and Wipro see accelerating contract cancellations through 2027, procurement teams demand 30% renewal discounts, the rupee falls 18% against the dollar in four months, and the IMF begins preliminary discussions with New Delhi by the first quarter of 2028.

It’s a gut-wrenching scenario, but the argument is built on a model of the Indian IT industry that is roughly 15 years out of date.

India’s software and IT-enabled exports were roughly $205 billion in FY25, per the RBI’s annual survey of software export companies. The claim that this entire edifice was built on “one value proposition,” cheap developers, is a great misunderstanding of what these companies actually do for a living.

The technology stacks of the world’s largest enterprises are sprawling and non-monolithic, they take years to adapt to each new wave, and the work Indian IT companies do is overwhelmingly not greenfield coding. Custom application maintenance alone accounts for roughly 35% of a typical Indian IT company’s revenue, per HSBC’s service-line framework: incident management, service requests, change requests and problem resolution across architectures where SAP, Salesforce, Snowflake, Databricks, Guidewire and ServiceNow coexist in configurations unique to each client.

At an event this week, Anthropic shared a point that directly undercuts the “AI coding agent collapses the model overnight” narrative: the displacement risks that the firm itself highlighted are highest for job profiles involving data entry and project and technical documentation, not programming. There was no mention of the mean increase in coding productivity at the event, but the high end was pegged at 90%, mostly in recently established companies rather than those with long tenures and legacy codebases.

The gap between what AI can do to a two-year-old startup’s codebase and what it can do inside the accumulated infrastructure of a Fortune 500 enterprise is not a minor implementation detail.

These enterprise platforms require deterministic outputs, exact, repeatable, compliant, and AI is probabilistic, meaning its output can change every time the same question is asked. That mismatch is the core reason AI cannot wholesale replace the systems of record that form the backbone of enterprise IT. Enterprises will make AI behave more deterministically through context engineering, testing and governance, and Cognizant CEO Ravi Kumar flagged context engineering in his most recent earnings call as the key to making AI outcomes deterministic over probabilistic. Building and maintaining those guardrails is system integrator work, and not something a standalone coding agent handles at the cost of electricity.

AI does have a deflationary impact on the sector and it is not trivial. HSBC, in a note this week, raised its estimate of gross AI-led revenue deflation for Indian IT from 8-10% to 14-16%, broken down across 13 revenue sub-segments. Enterprise readiness is significantly lagging AI progress, HSBC also noted, echoing Infosys’s recent investor day, meaning the deflationary impact will likely take multiple years to fully materialise rather than arriving as a sudden cliff.

CLSA separately estimated overall effort reduction at 10-20% depending on service line exposure, application development and maintenance and BPO seeing the most deflation, cloud migration and consulting the least. A granular revenue headwind measured in percentage points, however, is a fundamentally different thing from an industry in freefall where the IMF shows up at the door.

These deflation numbers do not exist in a vacuum. Anthropic’s internal research finds that about 27% of AI-assisted work consists of tasks that would not have been done otherwise: scaling projects, building tools like interactive dashboards, exploratory work that would not be cost-effective manually. IDC projects GenAI to be net positive for IT services spend growth in 2026.

Bank of America argues the opportunity side is materially larger than the threat. About 62% of the value that AI generates for enterprises is expected to come from core business functions like operations, sales and marketing, and R&D, where IT services companies have an entirely new addressable market. IT as a support function accounts for only about 7% of total AI value generation, and that is the smaller bucket where deflation is concentrated.

Cognizant, at its recent investor day, put the current labour spend pools in core business processes, including R&D, sales and marketing, HR and customer service, at multiples of the IT services labour pool across every segment. The opportunity for providers of AI agents to help enterprises modernise those core processes is, on a spend-pool basis, significantly larger than the revenue at risk from coding automation.

The enterprises themselves are moving in this direction. Citizens Financial announced a multi-year initiative called “Reimagine” broken into 50 work streams, half of which are tech and AI-enabled. Bank of America has mapped all of its work into 3,700 processes and 55,000 individual activities as a prerequisite to deploying AI and expects technology spending to increase 5-7% in 2026 to more than $13 billion total, with over $4 billion in new initiatives. Goldman Sachs has identified six discrete work streams and assembled teams to underwrite all activities for efficiency opportunities.

These public announcements aren’t painting the picture of IT services contracts getting cut. They are evidence of the need for technology partners to deploy firm-wide transformation at a scale no AI coding agent can execute autonomously.

Anthropic itself has been building an extensive partnership network with system integrators to take its solutions to enterprise clients, and it’s not a strange — enterprise customers account for roughly 80% of Anthropic’s business and about 40% of OpenAI’s, which is why both are investing in SI partner ecosystems rather than going direct.

Accenture today struck a multi-year deal with Mistral AI to use the startup’s models to assist its own clients in leveraging AI, following a different partnership Accenture struck in December to train 30,000 professionals across financial services, life sciences and the public sector. Cognizant rolled out Claude to 350,000 employees in November. Deloitte rolled it out to 470,000 employees and began training 15,000 professionals in October. IBM announced a strategic partnership in October to infuse Claude into its software portfolio.

At the Anthropic event, the company described Claude Cowork making strong progress across data, legal, finance and banking, and emphasised that implementation is happening in partnership with firms like Accenture, Infosys, BCG and Deloitte to create offerings tailored to enterprise environments. Anthropic expects agentic architectures to evolve from single-agent workflows to multi-agent systems where an orchestrator coordinates specialised agents working in parallel across dedicated contexts, the kind of complexity that system integrators exist to manage.

The doomsday narrative on the Indian IT industry isn’t new. The claim that ERP would kill the application development and maintenance business turned out backwards, as SAP and Oracle ERP implementations became multi-year, multi-million-dollar projects. The claim that the shift to cloud and SaaS would cause significant volume shrinkage for IT services has played out, and software revenues including SaaS have grown at 3-4x the rate of IT services spend without destroying the sector. The claim that social, mobility, analytics and cloud selling required a heavy onsite presence that Indian IT did not possess also proved wrong, as cloud and data analytics are currently the fastest-growing service lines for Indian IT firms.

In the last major technology cycle, the shift to cloud, average organic constant-currency revenue growth for the top five Indian IT companies bottomed around FY16-FY18, re-accelerated from FY19, and the sector found itself re-rated. The narrative has always changed when earnings growth turned around.

Citrini’s broader macro argument is that IT services exports are the single largest contributor to India’s services trade surplus and finance its persistent goods trade deficit, so if AI destroys that revenue base, the surplus evaporates, the rupee crashes 18%, and the IMF shows up.

Citi’s economics team, analysing RBI data, offers a detailed counter. The 10-year CAGR for software services exports is 9.5%, against 3.4% for goods exports, and software services exports have never experienced a year-on-year decline in 24 years, not through the dot-com bust, the global financial crisis, or Covid.

Management and consultancy services exports, which include GCCs, have grown even faster at a 26.3% 10-year CAGR. The share of management and consultancy services in total services trade surplus has gone from about 13% pre-Covid to about 37% in FY25. Even in the most recent data, October 2025 through January 2026, after AI anxiety was already widespread, total services exports continued growing at roughly 12% year-on-year.

Citi does expect India to post a balance of payments deficit in both FY25 and FY26, the first consecutive deficits since at least 1991, driven by persistently low net FDI flows and a widening goods trade deficit. That is a real vulnerability and worth taking seriously on its own terms, but Citi’s analysis maps out exactly how bad the software export picture would have to get for the Citrini scenario to materialise, and the numbers do not support it.

In their base case of roughly 8% growth in software exports for FY27, the BoP returns to surplus at about $20 billion. Even if growth halves to 6%, the surplus holds at $15-17 billion. A scenario of flat software exports, zero growth, something that has never happened in 24 years, would reduce the surplus to $5-8 billion but still not tip India into deficit.

The Citrini scenario of a 5% year-on-year decline, accelerating cancellations and an 18% rupee depreciation requires something that has never occurred in the history of Indian software exports, sustained at a pace that outstrips every historical precedent including the worst global recessions of the past quarter century. It also requires ignoring that India is sitting on record foreign exchange reserves of $725.7 billion, enough to cover nearly 12 months of goods imports and about 96% of external debt, a buffer of a fundamentally different order from the early-1990s crisis that actually did bring the IMF to Delhi.

The on-site service share in total software exports has fallen to 9% in FY25 from 20% in FY16, meaning the sector’s immigration dependence is significantly lower than a decade ago. Geographical diversification is improving as the U.S. and Canada share has declined from 64% in FY13 to about 54%, replaced by Europe and Asia. The rising share of private limited companies in software exports, from about 35% in FY13 to about 60% in FY25, may partly explain the gap between listed company revenue growth and the resilience of aggregate software exports in RBI data, suggesting GCCs and unlisted entities are capturing a growing share of the export pie.

There is also an emerging structural opportunity that the Citrini narrative entirely ignores. HFS Research projects a category it calls Services-as-Software to grow from essentially zero today to $1.5 trillion, eating into share from both traditional software and traditional IT services. In this model, AI-driven autonomous delivery replaces seat-based SaaS pricing and time-and-material services billing, and pricing shifts to outcome-based and consumption-driven models. Salesforce has already moved from seat-based to consumption-based pricing, Accenture and Globant have made early moves, and more than 60% of enterprises plan to replace at least some of their professional services with AI-driven solutions by 2030, per an HFS survey of 1,005 major global enterprises. IT service companies proactive about building their own AI agents, and willing to cannibalise legacy revenues, can gain share from software companies rather than just lose it.

CLSA adds that GCC headwinds could turn into a tailwind as bloated GCC cost structures and the mainstreaming of AI push work back toward outsourcing, and that enterprises increasingly prefer to “build” over “buy” for AI-driven differentiation, increasing integration and customisation work for services firms rather than reducing it.

NSE IT now trades at roughly a 2% discount to the Nifty, against a 10-year average premium of about 15% and a 5-year average premium of about 29%. Indian IT has de-rated to approximately 17x FY27 estimated earnings. The sector’s problems are real: low- to mid-single-digit organic constant-currency revenue growth for large caps over the past three years, GCC competition for talent, IT services being crowded out in overall tech budgets. But a scenario that imagines IMF discussions with New Delhi because AI coding agents run at the cost of electricity is not a description of those problems.

The AI agents market is projected to reach $52 billion by 2030, growing at a 45% CAGR, per BCG. GenAI services software is growing at a 57% CAGR over 2025-29, per IDC. GenAI revenues already account for roughly 5-10% of total revenues for Indian IT companies and are growing fast: Accenture’s GenAI bookings reached $2.2 billion in the first quarter of FY26, or 10.5% of its overall order book, up from 0.4% in FY23, and TCS reported annualised AI services revenue of $1.8 billion, or 6% of overall revenue. CY26 organic growth guidance by Cognizant, Capgemini and EPAM is all coming in better than CY25, and multiple brokerages expect growth to exceed consensus in at least FY27 on cyclical tailwinds and a strong U.S. corporate earnings season.

In short: 24 years of software export data that has never posted a decline, $200 billion in annual revenue, partnerships with the very AI labs whose products are supposed to be the instrument of the sector’s destruction, possibly a new $1.5 trillion market category emerging at the intersection of services and software, and the largest U.S. corporates in the middle of mapping their entire workforces into process architectures that require technology partners to modernise. I think India’s IT is going to be fine.

Ather’s market cap today: $2.95 billion. Ather’s valuation in private markets three years ago: $426 million.

Introducing Charted, a new newsletter from India Dispatch that picks one chart and unpacks what it tells you about India’s economy, markets and policy. Four to five times a week.

An investment bank used these projections to successfully sell the deal to a bank’s wealthy clients.

The only Indian apps in the top 20 are PhonePe at #11 with 167 million daily active users and JioHotstar at #18 with 94 million.

Indian apps make up 35 of the top 100 and 123 of the top 300. In comparison, 44 US apps in the top 100, 74 in the top 300.

Introducing Charted, a new newsletter from India Dispatch that picks one chart and unpacks what it tells you about India’s economy, markets and policy. Four to five times a week.

Large-caps were hit hardest, where foreign ownership dropped to its lowest point in a decade, even as mid- and small-cap holdings remained comparatively resilient.

Introducing Charted, a new newsletter from India Dispatch that picks one chart and unpacks what it tells you about India’s economy, markets and policy. Four to five times a week.