You should consider various factors when determining whether or not to use an extract. In some cases, you won't have an option (for example, OLAP requires a live connection and some cloud-based data sources require an extract). In other cases, you'll want to evaluate your options.

In general, use an extract when:

• You need better performance than you can get with the live connection.
• You need the data to be portable.
• You need to use functions that are not supported by the database data engine (for example, MEDIAN is not supported with a live connection to SQL Server).
• You want to share a packaged workbook. This is especially true if you want to share a packaged workbook with someone who uses the free Tableau Reader, which can only read packaged workbooks with data extracted.

In general, do not use an extract when you have any of the following use cases:

• You have sensitive data that should not be accessible by certain users, or you have no control over who will be able to access the extract. However, you may hide sensitive fields prior to creating the extract, in which case they are no longer part of the extract.
• You need to manage security based on login credentials. (However, if you are using Tableau Server, you may still use extracted connections hosted on Tableau Server that are secured by login. We'll consider sharing your work with Tableau Server in Chapter 12, Sharing Your Data Story).
• You need to see changes in the source data updated in real time.
• The volume of data makes the time required to build the extract impractical. The number of records that can be extracted in a reasonable amount of time will depend on factors such as the data types of fields, the number of fields, the speed of the data source, and network bandwidth. The hyper engine typically builds the new .hyper extracts much faster than the older .tde files were built.