Academic AI seldom reaches mkt: Wadhwani AI CEO Shekar Sivasubramanian

Shekar Sivasubramanian of Wadhwani AI says many AI pilots fail to scale because solutions lack real-world fit. He urges deeper work on data, languages and social sector deployment

A wide gulf persists between what is developed in laboratories and what eventually reaches users as market-ready products, often resulting in startups launching solutions that lack product–market fit and fail to solve real problems, said Shekar Sivasubramanian, chief executive officer of Wadhwani AI.

 

The institute, which focuses on AI solutions for the social sector — primarily agriculture, education, and health — works with 14 states on 25 AI tools that reach nearly 150 million people.

 

“Right now, AI is 95 per cent of the conversation in the social sector, but only about 5 per cent is about actual solutions. Most of the world today runs a pilot with an AI solution and walks away,” he said.

 

AI, he argued, is often treated as a cosmetic feature. “If you just add AI as a ‘nice-to-have’, which most people do, then it is reduced to exactly that — it only looks good. AI should be approached directly, at the core, in a way that proves its value. The main story here is patience,” he said.

 

Investment trends in India, Sivasubramanian believes, are also shaped by an ‘us-too’ mindset, with venture capital flowing into startups working on modern computing, large language models (LLMs) or generative AI.

 

“The only place where it makes sense to invest right now is when your underlying data is different. Anyone with internet access can scrape information, and you may not produce anything spectacularly new from that,” he said.

 

For startups — especially those operating in the social sector — success depends on breaking large problems into smaller, solvable pieces.

 

Scaling solutions to more than 20-30 Indian languages will demand heavy contextualisation, requiring startups to gather datasets, clean them, and organise them for usability, he said.

 

Government programmes such as the IndiaAI Mission, he suggested, should prioritise multilingual challenges in rural India and deploy credible solutions at scale.

 

Large schemes like IndiaAI should also focus on extracting and digitising available information nationwide, especially in rural regions.

 

“One crucial step is to cast the net wide, both geographically and across time. Don’t just look at today — go back, maybe 1,000 years. Look at everything written on paper, digitise it, make it machine-readable. Then you’re strategically placing yourself to extract every piece of information this country has ever produced, long before 1947,” Sivasubramanian said, adding that much of this data is still inaccessible to LLMs.