Retiring $800K in Licensing While Modernizing a 100TB Healthcare Warehouse

Cigna's data organization was a casualty of its own success. A 100TB on-prem warehouse running on Teradata and Hadoop had become the connective tissue for thousands of analysts, engineers, and operational reports, but the Teradata renewal was on the horizon, the licensing was burning more than $800K every year, and leadership had committed to a cloud-first future.

• Signing the next Teradata contract would have locked in the legacy architecture for years past the cloud mandate, and it was already the largest single line item in the data budget
• A 10,000-user data org had no centralized governance. Data contracts, SLAs, and lineage were tribal knowledge held inside individual teams
• Years of pipeline tooling had been built around legacy connectors that did not understand cloud-native infrastructure, so the migration could not be a lift-and-shift

The decision was not whether to modernize. It was whether to do it on a timeline the renewal cycle and the cloud mandate would actually allow.

Sequencing mattered more than tooling. The plan committed to retiring Teradata workloads on a path that would land before renewal, while keeping legacy reporting alive on the old stack until each domain was cleanly cut over.

• Built the new EDW on BigQuery and the data lake on GCS, decoupling storage from compute so the cost model scaled with actual usage instead of provisioned capacity, retiring the fixed Teradata burn entirely
• Pioneered enterprise dbt adoption in its earliest releases, piloting with five teams before rolling out broadly so the modeling layer had real-world feedback before it became the standard for thousands of analysts
• Built a custom metadata and governance platform on GCP covering catalog, lineage, data contracts, and SLA monitoring. Cigna leadership pushed build over buy because the 2018 governance market did not serve the scale or healthcare-specific requirements without as much customization as a build
• Wrote custom connectors to bridge legacy reporting and orchestration tools that had not been updated for cloud data infrastructure, since the migration could not wait for every upstream and downstream system to modernize on the same clock

The hardest part was not the technology. Parts of the data engineering and analytics organization had been building practices and code on the legacy stack for seven years, and a cloud migration meant retiring patterns with real engineering investment behind them. We worked the rollout team by team, picked up legacy logic deliberately rather than abandoning it, and let the early dbt pilots do the persuading once the velocity gains became visible.

The migration retired more than $800K in annual licensing the moment the legacy contract lapsed, and the new cloud cost structure scaled with actual analytical demand rather than peak provisioning.

• $800K+ in annual Teradata licensing eliminated, with cloud spend tracking real usage instead of fixed capacity
• Model development cycles cut by roughly 50% across the broader data engineering org once dbt and the new modeling standards rolled out from the pilot teams to the wider organization
• The custom governance platform gave 10,000 builders their first shared definitions of data contracts, SLAs, and lineage, ending tribal knowledge as the only path to understanding upstream dependencies
• Audit and compliance reviews shortened materially because lineage and contracts were now queryable surfaces rather than reconstructed by hand each time
• Production reporting stayed live throughout the cutover thanks to the custom connectors bridging legacy and cloud, so the business never had to choose between modernization and operational continuity

Behind the licensing number was the larger story. A 10,000-user organization moved off infrastructure that had been pulling them backward and onto a foundation they could finally build on top of.

Enterprise modernization is a stakeholder problem before it is an architecture problem.

• The hardest engineering work in a 10,000-user org is rarely the engineering. It is earning the right to retire patterns that real teams have been building on for years
• Adopting tools before they are standard becomes a velocity advantage, but only if you pilot small, learn in public, and let early adopters carry the rollout to the broader org
• Build versus buy is not an ideology. At enterprise scale with healthcare-specific requirements, off-the-shelf governance in 2018 would have required as much customization as a build, and the build path was the cheaper, faster, better-fit answer