Know, Grow, and Protect Net Worth: Using ML for Asset Protection by Preventing Overdraft Fees

TL;DR

An ML-driven overdraft early warning system (ODEWS) that assesses a customer's risk of overdrafting within the next week using their banking and transaction data in the Mint app, resulting in a $3 million savings in overdraft fees for Mint customers compared to a control group.

Abstract

When a customer overdraws their bank account and their balance is negative they are assessed an overdraft fee. Americans pay approximately \$15 billion in unnecessary overdraft fees a year, often in \$35 increments; users of the Mint personal finance app pay approximately \$250 million in fees a year in particular. These overdraft fees are an excessive financial burden and lead to cascading overdraft fees trapping customers in financial hardship. To address this problem, we have created an ML-driven overdraft early warning system (ODEWS) that assesses a customer's risk of overdrafting within the next week using their banking and transaction data in the Mint app. At-risk customers are sent an alert so they can take steps to avoid the fee, ultimately changing their behavior and financial habits. The system deployed resulted in a \$3 million savings in overdraft fees for Mint customers compared to a control group. Moreover, the methodology outlined here is part of a greater effort to provide ML-driven personalized financial advice to help our members know, grow, and protect their net worth, ultimately, achieving their financial goals.

Figures (8)

• Figure 1: Overdraft Fees paid by Mint Customers per Bank from 09-2019--09-2020. On average Mint customers pay $250 million in Overdraft Fees to banks often in $35 increments. The share of overdraft fees collected by banks scales linearly with the number of customers. The goal of ODEWS is to provide a customer with an early warning to prevent paying unnecessary overdraft fees.
• Figure 2: Overdrafts per Week from 09-2019--09-2020. The Covid-19 pandemic saw mass shutdowns which led to a precipitous decline in the number of overdrafts due to a precipitous decrease in the number of transactions as well as changes in bank overdraft policies. As the pandemic progressed the number of overdrafts gradually increased. The rapid change in overdrafts due to changing customer behavior as well as changing bank policies makes the problem of preventing overdrafts well suited to an adaptable machine-learning based solution.
• Figure 3: The Mint app is used for tracking a customer's transactions and managing their finances. Customers can link their savings, checking, and investments accounts and see their transaction history.
• Figure 4: Temporal Cross Validation: when simulating making a prediction on 2020-06-14 the label is calculated between the dates 2020-06-07–2020-06-14 and features are calculated looking back six months from the date 2020-06-07. The test-set of the model is then calculated using a label from 2020-06-14–2020-06-21 and features looking back six months from 2020-06-14. In this way, train-test pairs are created every week over six months and used to train models taking into account temporal effects and serial correlations.
• Figure 5: Precision and Recall @k% for Chase Bank Model: The model informs who and how many people receive an email notification. There is considerable flexibility in how many notifications can be sent every week since there are no resource constraints. Models are selected trying to balance precision and recall. In this case the threshold of k% is set at 10% and precision@k% is 0.42 and recall@k% is 0.49, roughly balancing the precision@k% and recall@k%. The Chase Bank model has a 5x lift from the prior and 1.17x lift from business rules.