How Platformization and AI Are Changing the Analytics Development Lifecycle: T-Bank’s Experience

Today, 17,000 people use the Data Platform at T-Bank, with more than 50,000 recurring processes running and 1,500–2,000 ad hoc notebooks being created every day. At this scale, it becomes clear that it is no longer feasible to work across a large number of fragmented tools or keep adding new ones. They start duplicating each other, solving the same problems in different ways while making tool selection more difficult for users.

In my talk, I will show exactly where the analytics development cycle breaks down. Why the quality of Data Governance, Data Quality, and the processes themselves depend so heavily on the tool and the team; why increasing process maturity forces changes in both tooling and ways of working; how we ended up in a situation where half of our critical ETL processes depend on tools that lack sufficient control and observability; and how AI affects all of this.

The key idea of the talk is to view the platform not as a collection of services, but as a unified ADLC (Analytics Development Lifecycle). This changes how we think about the Helicopter notebook platform, the Tedi ETL tool, and our solutions in Data Governance, Data Quality, and DataOps. Platformization remains important, but its main driver is no longer just tool standardization — it is AI and the agent-based approach. AI-native development and agents are beginning to define new requirements for platforms, interfaces, data quality, and approaches to development automation.

I will also show the metrics we use to measure analysts’ work in our tools and explain why DORA and SPACE cannot be applied to data development without adaptation. Using examples, we will look at how metrics differ for ETL and ad hoc tasks, and what actually indicates the reliability of data systems.