基于电话销售语音数据的客户购车意向甄别模型  

Customer Purchase Intention Identification Model Based on Telesales Voice Data

作　　者：白雪峰 熊子瑜 王汉生[1] Xuefeng Bai;Ziyu Xiong;Hansheng Wang(Guanghua School of Management,Peking University)

机构地区：[1]北京大学光华管理学院

出　　处：《经济管理学刊》2024年第2期189-216,共28页Quarterly Journal of Economics and Management

基　　金：国家自然科学基金项目(12271012)对本文研究的资助。

摘　　要：电话销售在汽车营销中扮演着关键角色,但其面临着市场上客户意向差异显著的挑战。这一差异大大增加了电话销售人员的工作难度,降低了营销效率。因此,通过客户的回答准确甄别出具有真实购车意向的重点客户,是提升电话销售效率和效果的关键。本文基于语音分析技术提出了一个客户购车意向甄别模型。该模型通过分析电话录音,从客户的情绪维度、态度维度以及认知维度来建立指标体系,并利用逻辑回归方法来构建分类模型。本文使用542条真实客户的语音数据进行验证,结果表明客户的情绪唤醒度和耐心程度与他们的购车意向存在显著的相关性。原模型、AIC模型和BIC模型外样本检验的AUC值均超过了92%,展现了出色的识别精度。此外,为衡量该模型在真实场景下的性能,将负例通过Bootstrap方法扩大,同时保持正例不变,发现在23.1%的覆盖率下,模型可实现90%的正例捕获率,进一步证明了其在实际应用中的有效性和可靠性。China’s automotive industry is undergoing a significant transformation towards electrification and intelligent mobility,emphasizing the need for effective marketing strategies to enhance brand perception and navigate the challenges in traditional sales channels.Within this evolving landscape,telesales play a crucial role in screening customer intentions,a process vital for boosting sales efficiency and reducing costs.However,the reliance on the subjective judgment of sales representatives in this multi-step process often leads to inefficiencies,including missed opportunities and resource wastage on low-intent customers,underscoring the need for a more precise approach in aligning marketing efforts with customer intentions in the automotive sector s dynamic environment.Our research introduces an innovative model aimed at identifying customers purchase intentions through the analysis of voice data from telesales conversations.Utilizing advanced voice analysis technology,this model scrutinizes various aspects of customer responses,including emotional,attitudinal,and cognitive dimensions,employing logistic regression techniques for accurate classification.The validation of this model involved an extensive dataset of 542 authentic customer voice recordings,which unveiled a significant link between factors such as emotional arousal,patience,and the likelihood of purchase intentions.Impressively,the model s precision,determined by AUC metrics in tests beyond the initial sample,surpassed the 92%mark.This level of accuracy underscores the model s effectiveness in pinpointing true purchasing intentions,setting it apart in the realm of customer intention analysis.The outcomes of our study underscore the model s practical value,especially in the context of telesales where distinguishing genuine buyers from a vast pool of contacts is crucial.By homing in on pivotal indicators like emotional arousal and patience,our model adeptly filters through the noise to identify those customers most inclined towards making a purchase.Th

关 键 词：汽车行业 电话销售 语音分析 逻辑回归 

分 类 号：C81[社会学—统计学] C93