Scaling dbt on Amazon Redshift: how KOHO cut transformation runtime 70% without rewriting a single model

KOHO, a Canadian challenger bank serving 2.5 million customers, ran its entire analytics estate — BI, ML, and dbt transformations — on a single Amazon Redshift cluster. As data volume grew to 200 TB, 15,000 datasets, and 30 million events per day, the nightly dbt job stretched past 6.5 hours, spilling into business hours and putting data mart SLAs at risk from workload contention on shared compute.

In this session, KOHO's Principal Data Architect shares how the team re-architected from a centralized warehouse to a Hub & Spoke model and, ultimately, a data mesh — using Amazon Redshift Serverless, Amazon Redshift data sharing, and the new datashare-aware dbt-redshift adapter. By moving dbt onto its own right-sized Redshift Serverless workgroup while preserving a single source of truth, KOHO cut its daily dbt runtime by ~70% (6.5h → 2h) in a config-only migration — with no changes to dbt models and no downtime. Not one analyst had to rewrite a line of SQL.

Attendees will learn how to decompose a Redshift transformation monolith, when to apply data sharing versus consolidation, and how the dbt Redshift adapter's data sharing support turns cross-warehouse reads from a workaround into a simple configuration — unlocking intra-day data freshness, independent per-team scaling, and the resiliency headroom to grow into a regulated bank.