I created a model with BTYD/lifetimes using the p(active)/p(alive) and optimized the threshold for dividing into active and churned customers. The prediction overall compared to actual data (holdout period) is good.
I want to predict now p(active) for really future periods and divide into active and churned customers.
My problem is that I don't have a threshold anymore that is valid for the future periods. When using the threshold optimized by holdout period all customers will be classified as churners very soon (which is not realistic).
All literature I found stops after optimizing the threshold.