The automatic data type detection step uses a sampling or preview of each source column to determine the appropriate type. If this sample doesn't reflect the full or future set of values for the column the data type selected may be incorrect.
Automatic type detection is not used with structured relational database systems such as SQL Server.
If enabled, this step only applies to unstructured data sources such as flat files and Excel workbooks.
As per the Choosing columns and column names recipe, it's important to avoid dependencies on columns not required by the query.
As per configuration guidance in Chapter 1 (Configuring Power BI Development Tools) you can avoid automatic type detection via the Data Load options. As this is a Current File option only and since the setting is enabled by default you currently need to disable this automatic type detection for each new file.