Design a temporal classification problem

I want to solve a customer attrition problem for an insurance client (I cant share the data due to confidentiality). I have access to 3 years data for customer attrition, claim pattern, customer service calls, demographic etc. Now customer attrition can happen anytime during an year. There are 2 ways I thought of setting up the problem. 1. Fix a reference data e.g. 1 Nov'16. Dependent variable is customer churn (yes/no) in Nov'16-Jan'17. Independent variables duration can be taken from Nov'15-Oct'16 (past 1 yr) & time based features such as transaction in last 3 months can be created. (I use this approach mostly). Only challenge here is that if there is a seasonality in attrition (high in Oct) i can't capture that with this model and hence i will have to re calibrate it later for Oct. 2. Consider year 2016, create multiple rows per customer e.g whether he attrited in Jan-Mar'16, Feb-Apr'16 upto Oct-Dec'16 and create features from past. Append these row wise (10 row per customer per year). My doubt here is: Wouldn't we violate the i.i.d. condition here? Also there will be seasonality in features as for every row per customer, time based features would be from different duration. Can someone please give some intuition/mathematical reason behind which one is better (or if there is anything better). Please suggest some papers/books if possible. I don't want to use survival as client wants just a classification model.

Thanks a lot in advance.