傳統精準行銷多以購買機率或行為傾向作為客戶篩選依據，然而此類方法無法區分原本即會購買的自然轉化客群，與受行銷影響而改變行為的客群，可能造成行銷資源錯置。行銷增益模型 (Uplift Modeling) 雖可估計行銷處置所帶來的增量效果，但現有方法多以單一增益值 (Uplift Score) 作為決策依據，難以完整呈現個體在接受與未接受行銷處置時的潛在行為組合，尤其可能忽略因行銷處置而降低購買意願的 Sleeping Dogs 客群，進而在收益與損失不對稱的情境下產生決策偏差。
為更完整描述個體行為的異質性，本研究提出一套具結構化約束的因果行為機率映射框架，將個體特徵至四類潛在因果行為機率的轉換，拆解為潛在行為反應參數的估計，以及由該參數映射至因果行為機率分布的兩階段過程。相較於直接估計四類機率，本研究透過結構化參數降低模型輸出的自由度，使四類機率之間的關係符合預先設定的因果行為結構。基於此框架，本研究建構端到端深度學習模型，並透過具已知潛在行為機率真值的模擬資料，以及金融機構外匯行銷隨機實驗資料進行驗證。研究結果顯示，本研究模型能更穩定還原個體層級的因果行為機率，並提升關鍵客群的辨識與排序能力；在考量行銷收益、接觸成本與負向反應損失的情境下，亦能取得較穩健的決策績效。整體而言，本研究將行銷增益分析由單一增益值的估計，延伸至個體潛在因果行為分布的還原，提供兼具因果結構、機率解釋性與決策彈性的精準行銷方法。

Traditional precision marketing often targets customers based on purchase probabilities or behavioral propensities, but cannot distinguish natural converters from customers whose behavior is changed by marketing interventions, potentially leading to inefficient resource allocation. Although uplift modeling estimates incremental treatment effects, reliance on a single uplift score may overlook heterogeneous potential outcomes and produce suboptimal decisions under asymmetric gains and losses.
This study proposes a structurally constrained framework for estimating causal behavior probabilities. Customer characteristics are first mapped to latent response parameters, which are then transformed into probabilities for four causal behavior types. This parameterization reduces model flexibility while preserving predefined structural relationships. An end-to-end deep learning model is evaluated using simulated data with known ground-truth probabilities and randomized foreign exchange marketing data from a financial institution. Results show that the proposed model more accurately recovers individual-level causal behavior probabilities and provides more reliable customer rankings across different decision scenarios. Overall, this study extends uplift analysis from the estimation of a single uplift score to the recovery of complete causal behavior probability distributions.

Ascarza, E. (2018). Retention Futility: Targeting High-Risk Customers Might be Ineffective. Journal of Marketing Research, 55(1), pp. 80-98. doi:https://doi.org/10.1509/jmr.16.0163
Chou, P., Chuang, H. H. C., Chou, Y. C., & Liang, T. P. (2022). Predictive analytics for customer repurchase: Interdisciplinary integration of buy till you die modeling and machine learning. European Journal of Operational Research, 296(2), pp. 635-651. doi:https://doi.org/10.1016/j.ejor.2021.04.021.
Goldenberg, D. (2023). Leveraging Causal Uplift Modeling for Budget Constrained Benefits Allocation. Long Beach, CA: CIML Workshop at KDD 2023.
Gubela, R. M., & Lessmann, S. (2021). Uplift modeling with value-driven evaluation metrics. Decision support systems, 150. doi:https://doi.org/10.1016/j.dss.2021.113648
Gubela, R. M., Lessmann, S., & Jaroszewicz, S. (2020). Response transformation and profit decomposition for revenue uplift modeling. European Journal of Operational Research, 283(2), pp. 647-661. doi:https://doi.org/10.1016/j.ejor.2019.11.030
Gubela, R. M., Lessmann, S., & Stöcker, B. (2024). Multiple treatment modeling for Target Marketing campaigns: A large-scale Benchmark Study. Information Systems Frontiers, 26(3), pp. 875-898. doi:https://doi.org/10.1007/s10796-022-10283-4
Gutierrez, P., & Gérardy, J. Y. (2017). Causal inference and uplift modelling: A review of the literature. International conference on predictive applications and APIs (pp. 1-13). Sydney: PMLR.
Hansotia, B., & Rukstales, B. (2002). Incremental value modeling. Journal of Interactive Marketing, 16(3), pp. 35-46. doi:https://doi.org/10.1002/dir.10035
Haupt, J., & Lessmann, S. (2022). Targeting customers under response-dependent costs. European Journal of Operational Research, 297(1), pp. 369-379. doi:https://doi.org/10.48550/arXiv.2003.06271
Holland, P. W. (1986). Statistics and causal inference. Journal of the American statistical Association, 81(396), pp. 945-960. doi:https://doi.org/10.2307/2289064
Huang, T. W., & Ascarza, E. (2024). Doing more with less: Overcoming ineffective long-term targeting using short-term signals. Marketing Science, 43(4), pp. 863-884. doi:https://doi.org/10.1287/mksc.2022.0379
Kane, K., Lo, V. S., & Zheng, J. (2014). Mining for the truly responsive customers and prospects using true-lift modeling: Comparison of new and existing methods. Journal of Marketing Analytics, 2(4), pp. 218-238. doi:https://doi.org/10.1057/JMA.2014.18
Künzel, S. R., Sekhon, J. S., Bickel, P. J., & Yu, B. (2019). Metalearners for estimating heterogeneous treatment effects using machine learning. Proceedings of the National Academy of Sciences, 116(10), pp. 4156-4165. doi:https://doi.org/10.48550/arXiv.1706.03461
Leng, Y., & Dimmery, D. (2024). Calibration of heterogeneous treatment effects in randomized experiments. Information Systems Research, 35(4), pp. 1721-1742. doi:https://doi.org/10.1287/isre.2021.0343
Nie, X., & Wager, S. (2021). Quasi-oracle estimation of heterogeneous treatment effects. Biometrika, 108(2), pp. 299-319. doi:https://doi.org/10.48550/arXiv.1712.04912
Olaya, D., Coussement, K., & Verbeke, W. (2020). A survey and benchmarking study of multitreatment uplift modeling. Data Mining and Knowledge Discovery, 34(2), pp. 273-308. doi:https://doi.org/10.1007/s10618-019-00670-
Radcliffe, N. (2007). Using control groups to target on predicted lift: Building and assessing uplift model. Direct Marketing Analytics Journal, pp. 14-21.
Rößler, J., & Schoder, D. (2022). Bridging the gap: A systematic benchmarking of uplift modeling and heterogeneous treatment effects methods. Journal of Interactive Marketing, 57(4), pp. 629-650. doi:https://doi.org/10.1177/10949968221111083
Seidel, P., Beyer, H., Rößler, J., & Schoder, D. (2024). Who Let the Dogs Out? How Ignoring Certain Customer Types is Improving Treatment Assignment Policies. PACIS 2024 Proceedings. Ho Chi Minh City, Vietnam.
Wager, S., & Athey, S. (2018). Estimation and Inference of Heterogeneous Treatment Effects using Random Forests. Journal of the American Statistical Association, 113(523), pp. 1228-1242. doi:https://doi.org/10.48550/arXiv.1510.04342
Ye, Z., Zhang, Z., Zhang, D. J., Zhang, H., & Zhang, R. (2025). Deep learning-based causal inference for large-scale combinatorial experiments: Theory and empirical evidence. Management Science. doi:https://doi.org/10.1287/mnsc.2024.04625
Zhan, B., Liu, C., Li, Y., & Wu, C. (2024). Weighted doubly robust learning: An uplift modeling technique for estimating mixed treatments' effect. Decision Support Systems, 176. doi:https://doi.org/10.1016/j.dss.2023.114060
Zhang, W., Li, J., & Liu, L. (2021). A unified survey of treatment effect heterogeneity modelling and uplift modelling. ACM Computing Surveys (CSUR), 54(8), pp. 1-36. doi:https://doi.org/10.48550/arXiv.2007.12769
Zhao, Z., & Harinen, T. (2019). Uplift modeling for multiple treatments with cost optimization. 2019 IEEE International Conference on Data Science and Advanced Analytics (DSAA) (pp. 422-431). Washington, DC: IEEE. doi:https://doi.org/10.48550/arXiv.1908.05372