A carregar...
Publicação
Churn Prediction in Online Newspaper Subscriptions
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.33 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
In today's connected world, people turn more to the web to be informed of the news. Newspapers in
an online environment profit from employing subscription models. Despite that Portugal remains one
of the countries with higher levels of trust in news, readers present a low propensity to subscribe.
Hence, online newspapers' existing customers are a valuable asset. Therefore, it is in the best interest
of such businesses to monitor these customers to identify potential churn signs down the line.
Customer churn prediction models aim to identify customers most prone to attrite, allowing
businesses that leverage them to improve their customer retention campaigns' efficiency and reduce
costs associated with churn. Two different research approaches, namely prediction power and
comprehensibility, have been at the core of churn prediction literature. Businesses need accurate
models to target customers' right subset. However, many models are black-boxes and present reduced
interpretability. On the other hand, understanding what drives customers to churn can support
managers in making better-informed decisions.
This project report presents the development of a plan to tackle churn prediction in a Portuguese
newspaper with an online subscription model using Machine Learning methods. The models'
performance was evaluated in two experiments. One experiment assessed the performance for all
types of subscriptions and another considered only non-recurring subscriptions. The results of the first
experiment were tempered by an unplanned marketing campaign that run simultaneously with the
experiment on top of the contrasting contexts in which the model was trained and evaluated. On the
other hand, the second experiment's results suggest that for non-recurring subscriptions, a phone call
from the call centre proved to be an adequate retention measure for probable churning subscribers.
Additionally, models' predictors were analysed and it was found that users with lower fidelity rates
and few subscriptions present a higher propensity to cancel their subscriptions. The same occurs with
users whose product is annual or longer-lasting. These findings shed light on how to minimize churn
and improve reader engagement. Based on the models' results, and predictors' analysis, the
newspaper decided to implement a re-engagement newsletter to keep users engaged and prevent
future churn.
Descrição
Project Work presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence
Palavras-chave
Churn Prediction Classification Data Mining Interpretability Online Subscriptions Machine Learning