in the case ralph wall suggested that churn might depend on customer age. perhaps a different logistic regression would be appropriate for different customer age groups? try splitting the sample in 0-6 months, 7-13 months and 14 and running separate models.