Financial customer analytics requires specialized machine learning pipelines that incorporate domain-specific understanding of customer behavior. Existing automated ML approaches often lack the capacity to effectively construct marketing-relevant features and that manual construction of predictive models demands specialized expertise that is difficult for many institutions to consistently secure and maintain. To address this gap, we propose an automated framework for generating end-to-end machine learning pipelines tailored to financial customer analytics tasks. The system processes raw customer datasets alongside natural language instructions, and autonomously performs data modality recognition, domain-aware feature engineering, model selection, and pipeline assembly. The framework autonomously performs domain-aware feature engineering by automatically computing key marketing indicators (RFM metrics, CLV, engagement scores)-capabilities absent in generic AutoML systems. Experimental validation showing 1.4% to 5.4% accuracy improvements over existing automated ML techniques while reducing development time by nearly sevenfold. Natural language interface enabling business stakeholders to configure pipelines without machine learning expertise.
Tian et al. (Tue,) studied this question.