Preventing debt repayment dropouts using predictive modeling

A debt management program only works if people stay in it. American Financial Solutions, a nonprofit whose program has helped over 450,000 people repay more than $9 billion in debt across 20 years, was losing clients to early dropout, and asked us to build a messaging strategy across its channels, counselor calls, email, text, and mail, that would keep people repaying.

We analyzed public datasets alongside records from thousands of AFS clients, applying k-means clustering and logistic regression to more than 100 characteristics per person. The model showed that beneath very different demographics, the psychological drivers of unsustainable financial behavior are strikingly similar, and it traced dropout risk to a specific origin: elements of the initial onboarding call were setting clients on a path to quit months later. The barriers it identified explained 80% of program dropouts.

We turned the model into an early-warning process for targeting at-risk clients, then redesigned the touchpoints themselves: counselor call scripts that surface each client's own motivations at enrollment, check-in texts, marketing collateral, and mail warnings, each change with its own ID, testing criteria, and rollout schedule.

The dropout rate fell 50% with the redesigned touchpoints in place. The model behind them explained 80% of dropouts before they happened, from the earliest moments of a client's relationship with the program, and the strengthened program is credited with billions of dollars in additional debt repaid.