How AI Is Changing India’s IT Services Industry

In the past, IT companies charged for inputs such as the time and people involved in a project. “We need to dramatically change from being input-based to becoming more output- and outcome-based; even cannibalise our revenues to create completely new businesses,” Vijayakumar said.

Another way to look at AI replacing people is the emerging trend in the tech services industry of ‘services as software’. Tech and professional services once delivered by humans can now be delivered by software, or at least that’s the sales pitch from AI vendors. The opportunity, and the challenge, for IT companies, therefore, is to build those new services around AI-led software and platforms. “The change that AI is ushering in is very different,” Vijayakumar said.

In January, Infosys announced four specialised small language models (SLMs) for industry-specific use, as part of its Topaz AI suite of applications. These include a model for banking customers and one for IT operations, built on top of an ‘AI stack’ from Nvidia, the AI industry’s dominant semiconductor chipmaker.

Infosys also announced the extension of an existing partnership with Microsoft to help customers adopt GenAI and Azure cloud infrastructure and technologies. It also announced the launch of a Google Cloud centre of excellence, “to foster enterprise AI innovation”, according to the company’s January 16 press release on its third-quarter earnings this fiscal. Many more such industry-specific solutions will likely be a big part of how IT companies will build their AI services capabilities. 

The LLMs from companies such as OpenAI and Meta are the ones that are currently driving the global GenAI frenzy. OpenAI’s ChatGPT AI chatbot now has more than 400 million weekly active users worldwide.

Many Indian startups are building their own solutions on top of Meta’s open-source LLM, Llama. More recently, Chinese startup DeepSeek has created much excitement with its eponymous LLM and chatbot, released in January, that it said were developed for about $5.6 million, which is a fraction of the cost that it took OpenAI to develop GPT, the underlying model that powers its chatbot.

Infosys’s partnerships with Microsoft, Google and Nvidia are examples of how India’s IT companies are likely to be influenced by the rise of AI. The capex that these companies have announced, as well as plans at Meta and some of the more deeply funded startups such as Anthropic, adds up to many hundreds of billions of dollars.

This in turn will drive investments at India’s IT companies, although on a much smaller scale, to develop new services around those products and platforms, Parekh pointed out. And the trend extends to other emerging technologies such as Quantum computing, he said.

Infosys is already using two “early quantum platforms” (a commercial quantum computer is likely several years away), he said, to offer solutions to customers in drug discovery and telecom network optimisation, for example. Infosys is also adding AI features and capabilities to its core banking product, Finacle, an insurance product McCamish, which became part of Infosys originally via an acquisition in 2009, and an e-commerce product called Equinox, Parekh added.

HCL Technologies reported that its software unit, built around the $1.8 billion acquisition of some products from IBM in 2019, had an annual recurring revenue (ARR) of about $1 billion at the end of its third quarter that ended December 31, 2024. Similarly, Tata Consultancy Services, India’s biggest IT company, has a massive portfolio of products and platforms, including its BaNCS core banking software, ignio in enterprise IT operations, and Optumera in the retail vertical.

Such software intellectual property by itself doesn’t account for a big share of revenue; it’s value is in the higher-margin services it enables. With AI too, IT companies will have to find ways to offer valuable services to their biggest customers to keep growing. What CEOs such as Vijayakumar and Parekh are expecting is that the industry will see tough changes in the type of services that will be valuable, how customers expect to pay for those services and how IT companies themselves will deliver those services.

At a high level, the expectation is that AI will bring significantly more productivity, both within IT companies and in the IT and business operations of their customers, according to Parekh. This will result in improved “time to value”, Vijayakumar said. And, that value is to be delivered with half the number of people in comparison with how things work today.

Also read: AI can turbocharge Digital Public Infrastructure and unleash benefits at scale: Nandan Nilekani

Nasscom projects that the IT sector will end the current fiscal year as a “net hirer”, adding 126,000 net new hires, taking the industry’s workforce to 5.8 million by the end of this month, versus 5.67 million for FY24. Including some $19 billion in hardware sales and about $16 billion in software sales, the sector is expected to end the current fiscal year at $282.6 billion, which is about a 5 percent growth over FY24.

In the past, large companies depended on what they call “transformation” contracts, but the nature of the transformation itself is changing fast, says Yugal Joshi, a partner at the research and advisory company Everest Group. “Earlier, transformation generally meant you have to deploy new technology, change a lot of stuff in the operating model. But now the transformation could be simply a process change, or it could be just changing the way you are running your technology,” he says. 

He offers a simple example: “It may not be building anything, frankly. It may not be implementing new software or newer platforms. It could be like, okay, I have my run spend as X, and I want to reduce it by 30 percent. That itself is a transformation.” By run spend, he means operational expenses in a business.

In the past, much of this cost reduction would have been delivered by moving the work offshore, typically to India. Today a lot of it is around technology-enabled solutions, Joshi says. Of course, the offshoring-based cost savings continue, and still accounts for the biggest chunk of revenues for the sector, but “there is a limit to those things,” he adds. “With AI and GenAI in the picture, clients believe that now is the right time to push their technology services providers to build those assets,” which can make meaningful improvements in efficiency and effectiveness in hitherto hard-to-transform areas, he says.

Consider another simple example: Earlier, a transformation initiative would typically start with a questionnaire and a survey to determine the current state of the client’s IT systems. Creating the questionnaire itself is a simpler task with GenAI. And IT companies are codifying such processes at multiple levels of complexity because they have the experience from many projects.

Add AI to this and the potential is immense, but the jury is out on whether potential meets reality. The early evidence, however, is promising, he says. The broader message is that the time of only human-driven tech services is gone.

For now, AI definitely has limitations. “Research consistently shows that 75 percent of consumers still prefer to engage with human agents when dealing with complex issues,” wrote Wayne Butterfield, a partner at the sourcing and advisory provider ISG, in a recent post. For example, Verizon Communications deployed AI agents to handle 60 percent of routine customer queries, significantly reducing wait times. But when it came to billing disputes or technical issues requiring nuanced judgement, 60 percent of cases still escalated to human agents.

At Walmart, an AI-based customer service solution to manage returns and refunds during peak shopping seasons successfully processed more than 70 percent of cases, halving handling times. But during a widespread product defect issue involving a popular electronics item, the AI system failed to recognise customer frustration and distress. “The reality is that, while AI excels at handling structured, repetitive interactions, it struggles with the nuance, empathy and problem-solving capabilities that define high-value customer interactions,” Butterfield writes.

“Attempts to present AI as desirable, inevitable, and as a more stable concept than it actually is follow well-worn historical patterns,” write David Gray Widder and Mar Hicks in a November 2024 paper at Harvard Kennedy School’s Ash Center for Democratic Governance and Innovation. Widder is a postdoctoral fellow at the Digital Life Initiative at Cornell Tech, while Hicks is an associate professor of data science at the University of Virginia’s School of Data Science. Vendors tend to present a new technology as inevitable, a strategy to achieve buy-in and boost sales, they point out.

From automobiles and railroads to electricity and computers, this has been the case, but they all required major infrastructure investments—roads, tracks, electrical grids, and workflow changes—to become functional and dominant. “None were inevitable, though they may appear so in retrospect,” they write.

And building and deploying AI assistants—agentic AI being the latest tech being touted as inevitable—is no easy task and has several layers of complexity, from technology to legal liability, says Sidu Ponnappa, founder and CEO of Realfast, an AI software startup in Bengaluru and Singapore. Ponnappa and his team are initially focusing on AI agents that can support teams within IT services companies that deploy Salesforce software for their clients.

“If you need to get work done from an agent in an enterprise context, it has to go through the same checks, balances and processes that humans doing the work go through, and this is non-trivial,” says Ponnappa.

Use cases such as software coding tend to be the easiest ones to crack because one can test and verify the results literally on a computer, he says. But in more subjective areas with multiple stakeholders whose perspectives matters, it’s not easy to check if the work done by an AI agent, semi autonomously, is done well, and accurately.

Then there is the question of liability. When a human executive signs a contract, the liability is clear, but what are the rules of the game when something done by an AI agent derails a project. For such reasons, many projects don’t see the light of day, he says. One of the ways in which Ponnappa and his team deal with this is to never expose the AI agents they develop to any of their customers’ end-customers. This way everyone knows what to expect, and they are better equipped to avoid mishaps in the outside world.

Such nitty-gritties of enterprise IT will likely mean that the promise of the benefits of AI will take much longer than being promised today. As the adage goes, people tend to overestimate what’s possible in the short term but underestimate the long-term potential. In the near term, it could very well be almost business as usual. With one difference: The pace of change is definitely an accelerated one. Parekh of Infosys pointed out that “if you look back over the last few years, at one stage we were 20 percent digital. Now we are over 60 percent.” He’s referring to the share of revenues that can be attributed to services related to digital technologies. “It’ll be a similar story” with cloud computing, for example, and AI, he adds. 

What India’s IT companies ought to be doing is to double down on becoming ever more relevant to their customers, Vijayakumar of HCL Technologies said, chasing not revenue but “value”. “Today AI is still a small part of our revenue, but it’s [also] a large part because it’s part of every discussion. Parekh adds. “And as we grow, it’s going to become more and more relevant.”