Bottleneck Diagnosis in International Automotive Sales Funnels Using Gradient Boosting Trees: Evidence from Cross-Regional Team Efficiency Evaluation

Ziren Zhou

doi:10.70393/6a6374616d.333631

Authors

Ziren Zhou Chinese Academy of Social Sciences

DOI:

https://doi.org/10.70393/6a6374616d.333631

ARK:

https://n2t.net/ark:/40704/JCTAM.v3n1a02

Disciplines:

Applied Mathematics

Subjects:

Mathematical Modeling

References:

10

Keywords:

Car Sales Funnel, Gradient Boosting Tree, SHAP, Bottleneck Diagnosis, Team Efficiency Index, Cross-regional Comparison

Abstract

Against the backdrop of slowing growth in the Chinese and global automotive markets, declining lead quantity and quality, fragmented online and offline data, and distorted data entry via manual DMS (Data Management System) have made it difficult for automakers to identify sales funnel bottlenecks and implement refined operations promptly. This paper proposes a funnel bottleneck diagnosis and cross-regional team efficiency verification framework inspired by the Gradient Boosting Tree (GBT) concept: the funnel is divided into three key stages, each trained with a LightGBM classifier. Time-slice cross-validation and stratified sampling by region are employed, combined with SHAP parsing to construct a "bottleneck index." Simultaneously, a "team efficiency index" is defined, integrating indicators such as first-contact delay, 24-hour follow-up frequency, reach diversity, and stage conversion for comparison and statistical testing between teams and regions. Based on multi-regional and multi-team data applications, the results show that first-contact delay, follow-up discipline, and price transparency are high-impact factors across multiple stages of the process. After introducing interventions such as "30-minute SLA + automatic warning," early-stage conversion significantly improves, and the sales cycle tends to shorten. The Chinese market possesses inherent advantages in the breadth and speed of digital touchpoints, while mature overseas markets are more robust in terms of process discipline and distribution systems. Based on this, this paper presents a regionally differentiated design for indicator weights and operational priorities. The research contribution lies in embedding interpretable machine learning into the sales governance closed loop, providing an integrated methodology and a practical management measurement system that spans diagnosis, intervention, and validation.

Author Biography

Ziren Zhou, Chinese Academy of Social Sciences

M.S. in Business Administration, Chinese Academy of Social Sciences, Beijing, China.

References

[1] Li, Z., Ji, Q., Ling, X., et al. (2025). A comprehensive review of multi-agent reinforcement learning in video games. Authorea Preprints.

[2] Zhang, Z., Wang, J., Li, Z., et al. (2025). AnnCoder: A multi-agent-based code generation and optimization model. IEEE Transactions on Software Engineering, 15(2), 135-148.

[3] Yang, J., Hu, R., Wu, C., Jiang, G., Alkanhel, R. I., & Elmannai, H. (2024). Sensor-Infused Emperor Penguin Optimized Deep Maxout Network for Paralyzed Person Monitoring. IEEE Sensors Journal, 25(13), 25638-25646.

[4] Lu, J., Zhao, H., Zhai, H., et al. (2025). DeepSPG: Exploring deep semantic prior guidance for low-light image enhancement with multimodal learning. Proceedings of the 2025 International Conference on Multimedia Retrieval, 935-943.

[5] Zhao, H., Chen, Y., Dang, B., et al. (2024). Research on steel production scheduling optimization based on deep learning. Proceedings of the 2024 4th International Symposium on Artificial Intelligence and Intelligent Manufacturing, 813-816.

[6] Yang, W., Lin, Y., Xue, H., & Wang, J. (2025, April). Research on stock market sentiment analysis and prediction method based on convolutional neural network. In Proceedings of the 2025 International Conference on Machine Learning and Neural Networks (pp. 91-96).

[7] Hu, R., Jian, X., Wang, J., & Zhao, H. (2025, July). Construction of a prediction model for rehabilitation training effect based on machine learning. In Proceedings of the 2025 2nd International Conference on Image Processing, Intelligent Control and Computer Engineering (pp. 41-45).

[8] Yang, J., Wu, Y., Yuan, Y., et al. (2025). LLM-AE-MP: Web attack detection using a large language model with autoencoder and multilayer perceptron. Expert Systems with Applications, 274, 126982.

[9] Tan, C., Gao, F., Song, C., Xu, M., Li, Y., & Ma, H. (2024). Proposed Damage Detection and Isolation from Limited Experimental Data Based on a Deep Transfer Learning and an Ensemble Learning Classifier.

[10] Yuan, Y., Xue, H. (2025). Cross-media data fusion and intelligent analytics framework for comprehensive information extraction and value mining.