Citi Global Wealth has spent the past 18 months moving client and market data out of separate regional data centers and onto a single cloud platform built with Snowflake and Amazon Web Services. Some reports that used to take weeks now take days, according to Computer Weekly.
Prateek Goel, managing director and global head of data and analytics technology at Citi Global Wealth, described the project at Snowflake World Tour Singapore 2026. Before the move, the wealth arm kept its data in Singapore, Hong Kong, the United Arab Emirates, Zurich and the US, and each location held its own set.
“When we talk about having an AI ambition, and we do not have data together in one place, it is just not possible,” Goel said.
Wealth clients’ portfolios shift with the markets, so the bank needs accurate, current data on both. Goel said access was a problem too, and that there was “an element of doubt” each time a person handled the data on its way through.
Store it once
Citi had been running Teradata, SAS, Hadoop and Oracle, but “what we needed was a cloud-enabled solution,” Goel said. The bank set one rule at the start: data is stored once, globally, and every use case draws on that same copy.
The architecture follows the medallion pattern. A bronze layer holds a data model for each source system, a silver layer unifies them under a common taxonomy that Citi calls its unified data model, and a gold layer holds models built for specific uses. Goel called the silver layer the most critical part, and said business stakeholders, not just technologists, helped write its definitions. Governance covers lineage, data quality checks and reconciliation.
Where Citi measures the return
Goel listed five areas. Reporting and management information went from weeks to days. Analytics, which he said made the biggest difference, now uses data clean rooms to give business users direct access to data they used to wait weeks or months for. System integration has moved from end-of-day files to APIs and events, and the bank is looking at model context protocol (MCP) servers as it does more with AI. Risk, finance and compliance teams use a machine learning pipeline that covers feature development, model building and deployment, with drift monitoring in production.
The fifth area is context. “Data is a foundation layer, and the second layer which is required on top is the context,” Goel said. “We’ve been very successful in building the foundation data layer. Our journey for building the context is starting now, and this is where the biggest impact in AI can be visible.”
He told The Edge Singapore the same thing more bluntly: “We can talk about AI and how impactful it will be, but if we do not have a common understanding of data, it makes the answer junk.”
Goel said work that took months now takes weeks and that thousands of hours have been saved.
WebProNews is an iEntry Publication