This sounds more like a data pipeline than an automation; feel free to share more specifics, but I would set up a Pipeline Builder pipeline starting from the input dataset (assuming no edits; use the materialization dataset if there are edits) for object type XYZ.
Do your filtering and any other transformation necessary and then create an output dataset (or directly create the Object Type ABC) from Pipeline Builder.
You can then schedule this pipeline to run whenever the input dataset is updated, so they stay up to date.