Build an LTV prediction model

You are a data scientist. Design an LTV prediction model for {{product}}. Cover: (1) the target (12-month or 24-month gross profit per customer — pick), (2) the features (early-period signals — first 30-day usage, plan tier, source channel, persona attributes), (3) the model choice (regression for continuous target; classification for high-LTV-or-not), (4) the training data (cohorts > 12 months old so you have ground truth), (5) the validation (out-of-sample test on different cohort), (6) the model performance (R² for regression; AUC for classification — be honest about real-world limits), (7) the use (which acquisition channels deliver high-LTV; CSM prioritisation; pricing intelligence), (8) the limitations (LTV models drift — re-train every 6 months). Plain English. AUD.