Incremental model

An incremental model is a table-building strategy, most associated with transformation tools like dbt, where each run processes only the rows that are new or have changed since the last run and merges them into the existing table, rather than dropping and rebuilding the whole table from the full source data every time.

The logic typically filters the source on a timestamp or a change-tracking column to find new rows, then either appends them or updates matching rows by key, depending on whether the table needs a full history or a current-state snapshot. This differs from a full-refresh model, which is simpler to reason about but becomes impractically slow and expensive as source data grows into billions of rows.

Incremental models matter because they keep transformation run times and compute cost proportional to how much data actually changed, not to the size of the whole table, which is essential once a warehouse handles high volumes. The main pitfalls are logic that misses late-arriving records because the incremental filter window is too narrow, and drift between the incremental table and a full rebuild that only a periodic full-refresh run will catch and correct.