If you need to include an interaction term, it shows that your model is wrong. Either your causal structure is wrong, or you're not in fact comparing the same treatment.
Interaction terms are just band aids that save you from having to correct your earlier mistakes in model specification.
OP, we already knew our model was wrong. All of yours are too.
That doesn't mean interaction terms cannot help you summarize the data.
Ever wondered if the correlation between wages and education depends upon where you were born? That is an interaction term, and it is interesting even without a causal interpretation.
Hell, you could even say that big Raj Chetty paper on intergenerational mobility is one giant interaction term.