Doubly robust and interpretable evaluation of customer retention targeting in subscription businesses

Abstract Customer retention targeting requires estimates of incremental response, yet robust average-effect estimation does not guarantee reliable customer rankings. We develop a reproducible evaluation framework combining cross-fitted doubly robust scores, interpretable uplift ranking, covariate diagnostics, and resource-constrained decisions. An internet-service cohort of 5,517 IBM Telco sample records provides an observational illustration; technical-support subscription is a service-state proxy rather than a documented intervention. On 1,655 held-out records, the doubly robust top-20% contrast is 12.52 percentage points, with a conditional interval of [-5.43, 28.75], and paired comparisons do not establish superiority over either baseline. Thirty repeated partitions expose substantial ranking variability. Two hundred average-effect simulations verify conditional nuisance robustness. A further 600 simulations evaluate known conditional effects and decision loss under logistic, interaction, and weak-overlap mechanisms. Adding second-stage interactions reduces doubly robust conditional-effect error from 11.32 to 7.96 percentage points under the interaction mechanism, but increases it from 9.34 to 11.66 under weak overlap. These findings distinguish causal identification, nuisance robustness, and decision reliability. The framework supports scrutiny of variable timing, model specification, and targeting uncertainty without claiming demonstrated campaign effectiveness or financial returns.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11319085/v1
Primary Topic
Customer churn and segmentation
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Doubly robust and interpretable evaluation of customer retention targeting in subscription businesses

Xiaobo Xing
Research Square
Customer churn and segmentation
preprint

Doubly robust and interpretable evaluation of customer retention targeting in subscription businesses

Xiaobo Xing
preprint en

Abstract

Abstract Customer retention targeting requires estimates of incremental response, yet robust average-effect estimation does not guarantee reliable customer rankings. We develop a reproducible evaluation framework combining cross-fitted doubly robust scores, interpretable uplift ranking, covariate diagnostics, and resource-constrained decisions. An internet-service cohort of 5,517 IBM Telco sample records provides an observational illustration; technical-support subscription is a service-state proxy rather than a documented intervention. On 1,655 held-out records, the doubly robust top-20% contrast is 12.52 percentage points, with a conditional interval of [-5.43, 28.75], and paired comparisons do not establish superiority over either baseline. Thirty repeated partitions expose substantial ranking variability. Two hundred average-effect simulations verify conditional nuisance robustness. A further 600 simulations evaluate known conditional effects and decision loss under logistic, interaction, and weak-overlap mechanisms. Adding second-stage interactions reduces doubly robust conditional-effect error from 11.32 to 7.96 percentage points under the interaction mechanism, but increases it from 9.34 to 11.66 under weak overlap. These findings distinguish causal identification, nuisance robustness, and decision reliability. The framework supports scrutiny of variable timing, model specification, and targeting uncertainty without claiming demonstrated campaign effectiveness or financial returns.

Research Square
Heilongjiang University of Technology (CN)
Customer churn and segmentation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon