How to Keep Your Customers: A Deep Learning Solution for FinTech
Team
Federico JoseĢ Derby Elizondo, Juan Carlos Sarmiento, Michael Morrissey, Colin Patrick Gaffney
Faculty Mentor
Markus Pelger
The Client: Alkanza
Alkanza Inc. develops automated, or "robo-investment" advisers for banks.
The Problem to Solve
Alkanza is interested in improving its customer retention, particularly among customers with subscriptions to multiple products or plans.
Engineering a Solution
Tools and Methodologies
The team used convolutional neural networks (CNNs), logistical regression and recurrent neural networks (RNNs) to model customer behavior and potential solutions. Each week, the model predicts the conditional probability of a user dropping a subscription in the coming week. This enables us to determine which users are susceptible to leaving the platform and can help target strategies to encourage continued use.
Further insights
- Initial Deposits: There is a clear exponential decay relationship between initial investment and the probability of leaving the platform (a lower initial investment means a lower probability of leaving)
- Number of Plans: A customer's second and later plans have a greater probability of discontinuation than a customer's first plan.
- Target Return (what users desire to earn): Sophisticated investors are less likely to leave the platform.
Recommendations
- Deploy a weekly intervention (e.g. emails). Interventions decrease a user's probability of leaving.
- We recommend that Alkanza compares optimized interventions with random interventions.
Deliverables
- Created a complex prediction model using state-of-the-art neural networks.
- Understood the main drivers of what makes users leave the platform.
- Modeled the economic value of improved predictions.