ELT is a methodology and means Extraction Load and (then) Transform vs. ETL which is Extraction Transform and then Load.

Each method has its own nuances. ELT has been around since the mid 1990.   However cloud databases have popularized ELT methodology as a differentiator.  Utilizing the ELT method means that data is loaded into the cloud database as it and only the data required data is transformed.

Knowing what data transformation needs to be done is the important key and not if  ELT is used or ETL.

A new and upcoming language for ELT is DBT  which is open source and used by ELT cloud database platforms like snowflake.

Where ELT potentially fails

As the company grows - Requires a higher level of governance as user can run their own transformations and consequently report different numbers.

Compliance - Could be a possible high impact issue as the data needs to be loaded before it can be used. For example - a data provider contract could specify that data can not be saved (in to your systems) BTW most single use data is in this category.  By saving it - how do you prove that you did not use it again. there are other examples as well like PII data)

When cloud databases are utilized and ELT is their standard.

Future proof - as no pre-supposition on which data to transform and how to transform. Only the data that is required for that transformation is used and do no decisions have to be made re: which data to discard and when.

Velocity of data means there is not adequate time to perform the transformation before loading . I.e it may not be practical to utilize ETL.

The ELT methodology reduces the time before data is available for analysis. ELT uses the target systems processing power instead of using a tools processing power.