Publication: Federated learning for customer digital on-boarding
Abstract
Rekabetçi finans piyasasında Finans Kuruluşlarının (FK) dijital müşteri olma sürecini en iyi şekilde optimize etmeleri, müşteri deneyimini geliştirirken aynı zamanda müşteriyle ilk temas anında edindikleri sınırlı bilgiyi en verimli şekilde değerlendirmeleri gerekmektedir. Bu süreçte bankalar müşteri tanıma, çapraz satış & üst satış, müşteri temsilcisi eşleştirme gibi birçok alandan yapay zeka modelleri kullanmaktadır. Verinin az veya dengesiz olduğu ve aynı zamanda hassas olduğu için üçüncü kişilerle paylaşılmasının mümkün olmadığı durumlarda kullanılan yapay zeka yöntemlerinden birisi de federe öğrenmedir. Federe öğrenme verinin kendisini paylaşmadan geri bildirim mekanizmaları ile birden fazla paydaşın ortak bir modeli eğitmesi yaklaşımıdır. Bu tez çalışmasında, banka verileri üzerinde federe öğrenme yöntemleri kullanılarak dijital müşteri olma süreci için müşteri skorlama modeli geliştirilmiştir. Federe, bölünmüş, yerel ve merkezi sinir ağı modellerinin sonuçları paylaşılmıştır. Sonuçlar, federe öğrenmenin daha başarılı olduğunu, küçük ölçekli kurumların federe öğrenmeden daha fazla fayda sağlayacağını ve federe öğrenme metodolojisinin finansal hizmetler alanında uygulanabilir bir yöntem olduğunu göstermektedir.
Technological advancements and the pandemic have made digital processes crucial. During the pandemic, most sectors have experienced rapid digitalization. Also, banking regulations have facilitated digital account opening, called digital on-boarding. Formally, digital on-boarding refers to the process of becoming a customer that enables customers to open a bank account remotely and digitally. In the field of competitive finance market, financial institutions need to optimize digital on-boarding process in a sufficiently feasible way. While improving the customer experience, they also need to utilize this first moment of contact with the customer most efficiently. In this process, banks use artificial intelligence for many optimization problems such as customer recognition, cross sell & up-sell, and customer agent matching. Federated learning is a machine learning model that enables learning only through feedback mechanisms without sharing data. Federated learning is preferred in areas where data privacy and protection is of vital importance. In this study, a federated learning model is established using a bank dataset. In the computational experiments, performance of central, local, federated and split neural networks are assessed. Results show that federated learning is statistically better than others and can be exploited in financial services by bringing a remarkable advantage to financial institutions of all sizes, but especially to the smaller ones.
Technological advancements and the pandemic have made digital processes crucial. During the pandemic, most sectors have experienced rapid digitalization. Also, banking regulations have facilitated digital account opening, called digital on-boarding. Formally, digital on-boarding refers to the process of becoming a customer that enables customers to open a bank account remotely and digitally. In the field of competitive finance market, financial institutions need to optimize digital on-boarding process in a sufficiently feasible way. While improving the customer experience, they also need to utilize this first moment of contact with the customer most efficiently. In this process, banks use artificial intelligence for many optimization problems such as customer recognition, cross sell & up-sell, and customer agent matching. Federated learning is a machine learning model that enables learning only through feedback mechanisms without sharing data. Federated learning is preferred in areas where data privacy and protection is of vital importance. In this study, a federated learning model is established using a bank dataset. In the computational experiments, performance of central, local, federated and split neural networks are assessed. Results show that federated learning is statistically better than others and can be exploited in financial services by bringing a remarkable advantage to financial institutions of all sizes, but especially to the smaller ones.