From Sale Conditions to Call-Backs: Early Policy Exit in Bank-Sold Life Insurance, Studied on a Synthetic Indonesian Portfolio

Bank-sold life insurance policies tend to end earlier than agency policies, and the sale is often the reason. This preprint studies that question end to end on a fully synthetic portfolio of 32,000 policies issued between 2022 and 2025 through eight fictitious partner banks, an agency force and a digital channel, built from nine linked tables with known planted effects and documented leakage traps. RQ1 measures the gap in early exit between bank-sold and agency policies, shows that product mix does not explain it, assigns it with an exact model-based bookkeeping to the customer base, to observable sale conditions (a loan around the sale, selling in the last week of a quarter, a deposit-only customer buying a single premium, a premium large against the deposit balance) and to an unexplained channel residual, and measures how little of the variation lies with partners, branches and sellers using a multilevel logistic model. RQ2 predicts exit from signals available 60 days after issue, comparing an Explainable Boosting Machine with logistic regression and gradient boosting on a later cohort, with leakage-safe unit track records, ablations, calibration and temporal robustness. RQ3 builds shrunk scorecards with funnel-plot limits for branches and sellers, validates them against the planted truth, derives a call-back rule that weighs the value at stake rather than risk alone, and sizes the randomised pilot needed to measure the save probability. All data are synthetic: no data from IFG Life or any other insurer or bank were used, and no result is a finding about a real company. The archive contains the paper, the data with a data dictionary and ground-truth files, the code, a Jupyter notebook and the results, and regenerates every number in the paper with one script.