Advanced Analytics
Explore complex analytical transformations and their impact on business metrics
Summary
The marketing team wanted to understand customer promotion behavior, specifically, which customers had ever used promotional offers. This meant adding a new 'is_promotion' column to the 'stg_payment' model, to flag transactions where the payment method was ‘coupon’.
Adding columns to foundational models like 'stg_payment' makes teams nervous. Even though it seemed like a simple addition, the teams worried: "Will this break existing dashboards? Are downstream models still working? Will our current reports show different numbers?"
The data team had to prove that adding 'is_promotion' was truly non-breaking change, that all the old stuff would still work, and the new metrics would just be a bonus. Instead of deploying and hoping nothing broke, they demonstrated that the change was additive-only, with zero impact on existing models and metrics.
The result? New promotional insights delivered with confidence, and a template for safely extending data models without stakeholder anxiety.
Problems
Prove schema additions won't break existing business logic
-
Addition anxiety: Adding columns to foundational models trigger fears that downstream logic might suddenly break.
-
Downstream uncertainty: Teams can't predict which models, dashboards, or reports might be affected by seemingly simple schema changes.
-
Safety verification challenge: Proving that new columns are truly "non-breaking" requires validating every dependent model and calculation.
Solutions
-
Visibility: Make schema changes crystal clear
Lineage diffs highlighted the new 'is_promotion' column addition, while confirming no existing columns were modified or removed. -
Verifiability: Demonstrate zero downstream impact
Breaking change analysis proved that downstream models like orders remained completely unaffected by the new promotional tracking column. -
Velocity: Build confidence in safe extensibility
By documenting non-breaking evidence, the team established a pattern for safely adding analytical capabilities without messing up what already works.
Trust, Verify, Ship
Cut dbt review time by 90% and ship accurate data fast
