Decision tree with missing values as a signal (missing not at random)

I am trying to build a decision tree (and later on a random forest), to predict future usage in a product. Some of the most important features are previous usage of this product, such as, the highest price this user was willing to pay in the past, lowest price etc. These features are relevant only to the customers who used the products in the past, but are missing for the customers who haven't. So we have some missing not at random, with the missing value indicating something about the customer (first timer). I want to be able to use these features, but it doesn't make sense to impute to the missing population. any idea how this can be handled?

Thanks