That's really not an ai issue in my view. I see that as not considering all the variables prior to designing the application. They are likely using other inputs from the crm to determine the best message to use. So they need to step back and look at how a lead status could be used in the mix.
I have always placed a lower weight on lead statuses in the crm as they are typically derived from a user at the dealership. I see it as the dealership's opinion of how they view the customer at that date and time. There are other variables that are less ambiguous also consider most of the statuses are not used correctly either.
If I was thinking about this from their perspective I would try to leverage variables that are devoid of human error as possible, recent, and most important - impact the end outcome the most. For example knowing the car they looked at was green is less important than they had a trade-in.
The other approach is to completely ignore all the prior data entirely and instead optimize on responses.
