Turn an expensive nightly full-refresh model into an incremental dbt model

The nightly job that rebuilds the whole orders table now takes ninety minutes, and the warehouse bill has three people asking questions.

Intermediate to advancedRead 5 minDoing: half a day to build and verify, plus a few nights watching the runtime and the late-arrival check before fully trusting it8 steps

▶ Start

What you'll have at the end

An incremental model that inserts only new and changed rows, cutting the nightly run from ninety minutes to under ten

Before you start

You need

A dbt model currently built as a full table rebuild from a raw source that carries a column marking when a row last changed and a column that uniquely identifies one row.

Not covered

Orders the source system deletes outright, which a filter that only looks for changed rows never catches on its own, and reshaping what a single row of this model represents: this only changes how the existing rows get built.

Leans on

8 steps

Where to go next

Give dbt a dev target

For a sandbox target to build and test this conversion against before it ever touches what the dashboard reads.

Checked 25 Aug 2026

Part of the dbt (Data Build Tool) cookbook