From 14-hour batches and poor documentation to AI-ready data: Mitti's dbt rebuild

Mitti has used dbt since 2020. But the project that grew up during that time had real problems: 14-hour daily batch execution, poor documentation, and questionable data quality, none of which are compatible with AI-powered BI.

We rebuilt the data model from the ground up, focused on quality, performance, and AI readiness. Results: 90% reduction in execution time, AI BI adoption exceeding 50% across the business, and over US$140K in infrastructure cost savings.

This session covers how we approached the rebuild, the decisions that drove the performance gains, and what it actually took to make dbt-powered AI BI work at scale inside a product company.