How to Improve Customer Retention in FinTech
A guide to identifying at-risk users and optimizing loyalty mechanics with machine learning.
1The Importance of Customer Retention in FinTech
In the FinTech sector, retaining existing customers is more economical and sustainable than acquiring new ones. With high customer acquisition costs in digital banking, loyalty strategies are crucial. This article explains how machine learning-based personalization techniques can help identify at-risk users and optimize loyalty mechanics.
2Defining and Measuring Churn
In FinTech products, churn is typically defined as users who haven't made any transactions for 30 days. This practical definition indicates the user is no longer actively using the product. The debit card product discussed in the article uses this definition, and a binary classification model was developed based on user transaction activity.
3Feature Engineering Strategies
Dozens of features were collected for the model: user profile information (age, region, device type), activity metrics over time windows (transaction counts, turnovers), time intervals (time since last transaction), and calendar features (day of week, month). These features are critical for capturing changes in user behavior and making accurate churn predictions.
4Structure of the Pre-Churn Model
The pre-churn model aims to identify users for whom existing loyalty mechanics are no longer effective. The model predicts the probability of a user making a transaction within the next 30 days. If the predicted probability falls below a certain threshold, the user is placed in the pre-churn segment. Preventive measures can then be taken for these users.
5Model Training and Validation
Gradient boosting algorithms like XGBoost, LightGBM, or CatBoost were used for model training. The model takes user features as input and outputs the probability of a transaction. During validation, the model is tested to ensure it works as intended in production, and the threshold value is determined based on factors like segment size and loyalty budget.
6Intervention Optimization with Uplift Modeling
The uplift model predicts which users will actually respond to loyalty offers. This model ensures resources are used efficiently by focusing only on users sensitive to interventions. It enhances the effectiveness of loyalty strategies by offering different deals to different user groups.
7Personalized Loyalty Offers
Based on model outputs, personalized loyalty offers are prepared for users in the pre-churn segment. For example, special discounts or rewards can be offered to users who haven't made transactions for a while. These offers are designed to re-engage the user with the product.
8Results and Impact
This approach can significantly reduce churn rates for FinTech companies. Machine learning models enable more targeted and efficient application of loyalty mechanics, lowering customer acquisition costs and strengthening long-term customer relationships.
Original article: Towards Data Science — FinTech'te Müşteri Sadakati Nasıl Artırılır?