← Toutes les actus
dbt Labs Blog17 août 2026
Model for the token, not the table
Résumé
Modeling Gong call transcripts directly in the warehouse with dbt, instead of hitting Gong's API on every AI call, cut token costs by 20x. The takeaway: for AI workloads, model the data at the token level, not just the table level.
We were burning through Gong's API to feed AI. Modeling the transcripts in the warehouse with dbt cut token costs 20x.
