Why a Rev alternative has to be honest about accuracy
Automatic transcription is very good and still occasionally confidently wrong, usually on names, jargon and crosstalk. On a vertical clip that costs you fifteen seconds of typing. On a deposition it costs considerably more, which is why Rev sells human passes and why that business exists.
So the honest split is not quality versus cost. It is what happens when the model is wrong, and who is accountable for catching it.