How AI is transforming India’s BPM industry

India’s business process management (BPM) sector is undergoing a structural reset, moving from cost-led headcount delivery to artificial intelligence (AI)-orchestrated intelligent operations that promise measurable business outcomes rather than just completed tickets. In the third episode of the Economic Times x Nasscom podcast series, a panel of industry leaders traced what this shift means for workflows, workforce skills, the evolving relationship between BPM providers and Global Capability Centres (GCCs), and India's positioning in the global services economy.

For decades, India’s BPM competitiveness rested on labour arbitrage, operational scale and the ability to execute large volumes of repeatable processes reliably. But as enterprises embed AI, automation and data-driven decision tools into core operations, the value proposition is changing: from narrow task completion to “intelligent service orchestration” that combines domain expertise, human judgement and AI capabilities to drive business performance.

Sreeram Pagalla, Global Vice President – CX Operations, GCC and Transformation at Concentrix, framed the early phase as mostly about “arbitrage”, taking advantage of the cost differences between countries with simple, repetitive processes. He identified the last five years, accelerated by generative AI, as the turning point, where the differentiator is now the ability to “orchestrate an engine” that delivers outcomes such as improved customer retention, faster resolution times or higher sales conversion, rather than merely answering emails or processing invoices. In this model, much of the “decision-making” is done by AI systems, shifting the human operator’s role to judging whether AI decisions are appropriate and whether underlying processes are correctly configured.

Mythily Ramesh, Co-founder and Managing Director at NextWealth, drew a sharp distinction between traditional BPO-BPM workflows, which are “linear” and “deterministic”, and AI-driven work, which is “probabilistic” and requires human-in-the-loop governance to ensure outputs are accurate and trustworthy.

“When you come to the AI sort of work that comes in, it is probabilistic. So, by its very nature, because it is probabilistic, you definitely need the human in the loop. It’s not like this is a standard way of doing things, but you need to have the human in the loop to ensure that the output is accurate, and the results that the AI is giving is trustworthy,” Ramesh explained.

She explained that “instruction sets” are replacing static SOPs as the primary operational guide, needing frequent updates, “every two, three days”, as AI systems and training data evolve. Measurement also changes: instead of simple accuracy or productivity metrics, operations now track inter-rater reliability (such as Cohen’s Kappa), agreement on golden data sets and closed-loop feedback from real-world performance to continuously refine models, instructions and training.

“So, let’s say you get sort of an email or chat in terms of a customer inquiry. The AI system is now deciding whether it goes to the chat bot, whether it goes to the specialist, or who does it, you know, is there an escalation for it?” Ramesh added.

Compliance, in this probabilistic environment, becomes more central, not just tougher. Ramesh emphasised that core compliance goals now map to emerging AI governance frameworks: systems must be trustworthy, reliable, responsible and explainable, with human validation of outputs, especially for sensitive personal, financial or health data, serving as both a quality control and a compliance mechanism to mitigate risk and maintain accountability.

Why this AI wave is different from earlier automation phases
Sangeeta Gupta, Senior Vice-President and Chief Strategy Officer at Nasscom, acknowledged that the BPM sector has undergone “multiple shifts” over two decades, but argued that the current AI moment is qualitatively different: an “exponential shift” in what is technologically possible, driven by rapid advances in AI models and capabilities. Unlike consumer apps, enterprise BPM operations are constrained by existing processes, compliance requirements and risk management, meaning AI must be integrated in a “very, very structured way” with deliberate choices about where it adds value and where human oversight is essential.

Gupta described BPM firms as becoming “almost the consulting arm for the enterprise”, helping clients map current workflows, design AI-enabled alternatives and decide where human-in-the-loop controls are critical, moving beyond siloed improvements to “cross-functional orchestration”. She also highlighted a business-model shift: from FTE or labour-arbitrage contracts to "intelligence arbitrage”, pricing or structuring deals around higher-value outcomes delivered by combining AI and human expertise.

“I think this phase is different, but I would say we’re still in the midst of that shift. It isn’t as if the entire shift has happened when the industry, the enterprises, they're all kind of right now in the middle of this shift and, you know, making sure that this shift gets the best outcomes for their customers, their employees and their stakeholders," Gupta surmised.

GCCs and BPM providers: convergence as the new tool of transition
With India now home to more than 2,400 GCCs, the panel noted increasing integration between GCCs and BPM providers. Pagalla explained that many older captive setups had become inefficient, weighed down by legacy systems, rising costs and compliance challenges. BPM partners, he argued, are stepping in as transformation orchestrators, bringing deep domain knowledge and overhauling legacy GCC operations end-to-end, redesigning workflows, deploying AI and rebuilding teams, rather than simply supplying additional headcount.

“Now look at this integration where companies like us, right, come together and say that we have deep domain knowledge. You don’t have to figure out a solution. We bring you that solution because I work in your domain extensively with hundreds of customers that deployed this and done this time and again,” Pagalla said.

From “back office” to “AI factory”: Jevons’ Paradox and India’s opportunity
Mythily contextualised India’s opportunity against its history as the “world’s back office”, acknowledging a near-term tension: as AI automates more work, headcounts in certain BPO tasks will fall. Citing Jevons’ Paradox—where greater efficiency initially reduces input use but ultimately expands total demand—she argued that, over time, the opposite dynamic would kick in: a short-term reduction from 100 to 40 people could eventually grow to a 400-person requirement as new AI-enabled workloads scale.

Her central proposition was that India could become the “AI factory of the world”, particularly for human- and expert-in-the-loop work: validating, testing and training AI models, and handling the most complex 25-30% of cases that automation cannot resolve. She outlined a hub-and-spoke vision, with major cities serving as AI development centres and smaller towns operating as large-scale spokes for human-in-the-loop operations.

“So, in summary, I think there is a huge opportunity, but to leverage that opportunity, our narrative has to be different. The BPO industry and organisations have to change their entire approach of how they are, because there is no existing process,” Ramesh said.

India’s global positioning and the 2030 outlook
Gupta placed these shifts within India’s broader technology ecosystem, highlighting that the country’s unique combination of tech services firms, BPM companies, a rapidly expanding GCC base, startups and major global tech presences.

“So, I think there is no other country that has an ecosystem that is scalable across every different segment. It’s not like one segment only that we are very strong,” she said.

In that context, BPM firms were positioned as critical integrators: because they sit at the front end of customer relationships and have run core processes for years, they understand where data lives, what breaks and what fixes are needed, enabling them to stitch together agentic AI transformations and partner with specialised AI vendors to deliver them faster.

“Hence, somewhere at least, we always struggled with how BPM is assumed to be just plain back office services. I think it’s no more back office. It is a lot, lot more than that,” Gupta said.

Looking to 2030, Gupta argued that the industry’s defining pillars are shifting from cost, quality and efficiency to domain intelligence and expertise. She outlined three interlocking factors: deep domain expertise in specific workflows (finance and accounting, logistics, procurement outsourcing), intelligence that reimagines workflows with AI and agentic systems, and expertise in learning—organisations will need “learning velocity” to keep employees on a continuous learning pathway.

“I think the 2030 industry has to be about domain intelligence and expertise, right? Where do you have the right domain?… How are you the expert with the domain around all of that?” Gupta said.

She also suggested that by 2030, rigid labels such as “IT services”, “BPM” and “GCC” might blur, with the next competitive advantage coming from orchestrating AI, people and processes together.

How roles within BPM are evolving under AI augmentation
Mythily emphasised that role evolution is already underway: as automation handles more routine work, the tasks reaching human workers are increasingly complex and judgement-driven. She cited medical coding as an example: where AI has automated 50-60% of the work, what remains for humans are the most difficult cases, shifting skill requirements towards subject-matter experts who can handle nuanced, domain-specific problems.

On skills, she drew a distinction between mechanical engineers and mechanics: the industry no longer needs only high-level engineers; it needs mechanics who can quickly understand context, respond rapidly, grasp the larger purpose of their work and feed feedback back to clients to improve AI systems, including work such as reinforcement learning from human feedback (RLHF), multi-turn conversation evaluation and human evaluation of AI outputs.

“Today you don’t need mechanical engineers, you need mechanics and there is a very subtle difference in that… he should be able to understand what is the context of what the client is giving, how can I respond very fast, what is the bigger purpose for which the work that I’m doing is meant to be,” Ramesh said.

In the closing segment, Gupta argued that AI must be a bold, company-wide agenda owned directly by CXOs, not delegated solely to the CIO, CFO or a newly created AI officer. Leaders cannot ask employees to become AI-savvy without undertaking their own AI learning journeys; they need to understand tools, solutions and how AI can make them more efficient, even if they do not code.