In my field (cinema/theatrical analytics), the value of AI/ML/Data Science is quite easy to calculate as we mainly use models that produce results that are then checked by human experts. To produce the same amount of output of just a single model, we would need 10x headcount.

I do agree though that an estimation of the value of AI work should be the first step in a project - often it's better to ask "why do we need this" even before asking "can this be done?".