Your dbt tests passed. Why did the metric still move?

Retention metrics often break in production due to inconsistent definitions, trial conversions, and account-level complexity. At Frame.io (Adobe), we saw the same metric product different results across Segment, Amplitude, and data warehouse models, leading to confusion across product, growth, and leadership teams. 

In this talk, I'll show how we reframed retention as a modeling problem and used dbt to build a centralized, reliable metrics layer. We redesigned cohort and activity models, standardized metrics definitions, and implemented dbt test to catch inconsistencies early.   

What attendees will learn:  

• How retention metrics break in real production systems 
• Common failure modes: cohort instability, activity definition, identity, and timing  
• How to design stable cohort and retention models in dbt  
• How to standardize metric definitions across teams and tools  
• How to use dbt tests to enforce metric correctness and prevent drift