Sriram Raghavan, General Manager, IBM Software, India and Software Innovation Lab
Only about 15 per cent of Indian enterprises surveyed have successfully taken at least one AI use case into scalable production, even as organisations move beyond experimentation and proofs-of-concept (POCs), according to an IBM executive.Citing a recent IBM survey conducted with the Ministry of Electronics and Information Technology (MeitY), Sriram Raghavan, General Manager, IBM Software, India and Software Innovation Lab, said the remaining 85 per cent of organisations are still working towards deploying their first AI use case at scale to realise measurable business outcomes.“This is no longer just about POCs. There are real proof points of moving into production. But is everybody moving to production? Is every industry moving into production? The rate and pace change,” Raghavan said.Scalable StrategyAccording to him, organisations that have successfully deployed AI at scale share three common traits: they focus on a handful of high-impact use cases instead of running numerous pilots, have stronger data readiness, and carefully manage AI deployment costs.“Peanut-buttering POCs across many areas tends not to work,” he said. “The reason is less to do with the technology itself and more to do with the fact that successfully getting business outcomes requires more.” He added that companies need to identify a business function—such as supply chain, procurement, human resources or IT—and scale AI within that domain before expanding further.Data readiness remains another major hurdle, particularly as enterprises adopt agentic AI. Organisations do not need to make their entire business AI-ready from the outset, but they must ensure the data supporting priority use cases is properly governed, connected and contextualised, Raghavan said.Cost HurdlesCost economics is the third major challenge preventing pilots from moving into production, he added. Many enterprises build successful demonstrations using the largest AI models but later find them too expensive to deploy at scale.“Sometimes, companies go all in without thinking through the unit economics of how much value they’re getting out of every unit of AI they’re deploying. It’s easy to build a demo or a POC early on that works well with the largest, most expensive model. But when you start to roll it out, you quickly realize it is not affordable because the value the use case is giving doesn’t justify the cost,” Raghavan said.He noted that adoption varies across industries, with banking, financial services and insurance (BFSI) leading the way due to more mature data architectures and stronger technology adoption. Other sectors are progressing at a slower pace, not because of reluctance but because they are still building the required data and operational foundations.Published on July 9, 2026
