Modern customer behavior is perceived to be much more complicated than it was twenty years ago. That may or may not be true.
Twenty five years ago, you created several hundred RFM segments and you called it good. Today, your three hundred RFM segments would overlay against twenty marketing channels, getting you ever closer to an infinite number of segment combinations. Is it any wonder folks want you to have a "one-to-one" relationship with a customer? It's become hard to segment the customer in a meaningful way.
It is perfectly reasonable to simplify the number of dimensions you deal with. I create simple models ... in the case above, a model for a product line called "Widgets", and a model for a product line called "Wudgets". The predictions for each model are reduced to five segments ... sort of like when you were in school ... A/B/C/D/F.
Then, I measure future "Widget" spend across the A/B/C/D/F grades for "Widgets", and A/B/C/D/F grades for "Wudgets". In other words, let's pretend that the customer is a prior "Widgets" buyer, but has never ever purchased "Wudgets". We'd look across the "A" row for Widgets, and then align with the "F" column for Wudgets ... the customer is worth $44.38 to the Widget merchandise category.
Say you have an email marketing campaign for Widgets ... you know that the customer must be worth at least $30 of future spend to generate a productive Widget email campaign. Even if the customer has not purchased Widgets in the past ("F"), a recent Wudget purchase might move the customer into A/B territory for Wudgets ... and when that happens, the customer has sufficient value to email the Widget campaign.
Better yet, you can build Wudget purchase intelligence into the Widget model ... and by doing so, you eliminate the need for the grid altogether ... you trigger/target based on the Widget grade.