Episode Details

Saurabh Gupta is CEO of The Modern Data Company, and we spoke about why so many AI initiatives struggle to scale despite massive investment in the technology. His perspective comes from nearly 30 years in data: designing the World Bank’s open data platform, spending 12 years leading statistical data at the IMF, serving as Chief Data Officer for Washington, DC, and later working across more than 20 major enterprise data initiatives at Thoughtworks. Across those environments, he kept seeing the same pattern: “people are not focusing on outcomes, people are focusing on technologies.”

Saurabh explains why “bad data leads to bad AI,” and why adding more compute cannot fix a weak data foundation. His approach starts with right-to-left thinking: “bring only the minimum data that you need to solve a problem,” then expand as new problems emerge. He describes one manufacturer planning seven to eight quarters of foundational work before his team delivered the first use case and supporting platform in less than one quarter. Instead of stitching together 10–12 specialized tools, the method combines ingestion, quality, governance, orchestration, transformation and cataloging while keeping context attached to the data itself. Complex customer problems, he says, can often move from months to roughly four or five weeks.

For listeners building with AI, the practical lesson is simple: start with the outcome, minimize the data and infrastructure required, prove value quickly, and only then expand.

Key takeaways

• Start with the business outcome before choosing technologies or infrastructure.
• Bring only minimum necessary data, then expand as adjacent problems emerge.
• Replace 10–12 disconnected tools with a unified ingestion-to-cataloging layer.
• Package context, governance, lineage, and transformations with each data product.
• Track unused pipelines and shut them down to stop wasted compute.
• Aim to prove complex use cases in four to five weeks, not months.