India’s AI Future Hinges on Data Quality, Not Models: EY Partner

Artificial intelligence is evolving at an extraordinary pace, with much of the global conversation centered on large language models and breakthroughs in generative AI. However, a recent conversation on the “CAIO Connect” podcast with Alexy Thomas, who leads the data practice at EY Technology Consulting, highlights a critical but often overlooked truth: the real foundation of AI success lies in data.

With more than 25 years of experience in the data and analytics space, Thomas emphasized that while AI models may capture headlines, it is the quality, governance, and connectivity of data that will ultimately determine how effective AI systems become. His insights provide a timely perspective for India, a country that possesses vast amounts of data but still faces challenges in unlocking its full potential for AI innovation.

“India is data rich. However, one of the problems that we need to tackle fairly quickly is our data is within silos,” Thomas said. According to him, India has valuable datasets spread across enterprises, government systems, and publicly available sources on the internet. However, these datasets often exist in isolation, limiting their usefulness for AI applications.

The solution, he argues, is not to move all information into a single massive repository such as a data lake or warehouse. Instead, the focus should be on creating connected data ecosystems where information is organized using clear taxonomies and shared standards. When data is structured and labeled consistently, organizations can exchange and interpret it more easily. This interoperability would allow startups, data scientists, and enterprises to access and use data seamlessly, accelerating innovation and unlocking the true value of India’s data resources.

Another key theme Thomas emphasized is the importance of AI-ready data. Not all data is suitable for training or powering AI systems, and organizations must ensure that their datasets meet several critical standards. First, data must be trustworthy, meaning its origin, provenance, and lineage are clearly documented. Second, the quality of data is essential; AI systems perform only as well as the information they are trained on. Third, strong data governance must be in place, ensuring that ownership and accountability are clearly defined if issues arise. Fourth, AI should increasingly rely on structured enterprise data, which often contains high-value insights for business applications. Finally, Thomas highlighted the need to minimize bias in datasets, since biased data can lead to flawed AI decisions and reinforce existing inequalities.