In dbt, an incremental model is built as a saved materialization of all rows of the source data, allowing for efficient processing and storage of only new or updated records. This approach optimizes compute costs and improves data accessibility by limiting the amount of data actively being changed and processed. By using dbt to create models and transformations of Snowflake data, users can aggregate, normalize, or sort their data without continuously updating their pipeline and resending data, ultimately enhancing downstream processes such as analytics and predictive modeling.