- H2O Driverless AI builds repayment, prepayment and default models from your own loan, payment and customer data.
- Propensity models predict whether a customer will apply for a mortgage or refinance within a given time window, so your offer arrives first.
- Every score comes with reason codes and sensitivity analysis, so credit risk and model risk teams can see how each input moves the result.
- A borrower likely to refinance gets a retention offer; a borrower likely to fall behind gets an early conversation about payment options.
Repayment behavior moves with every rate change
When rates fall, your best borrowers refinance elsewhere and the future interest you planned on disappears. When budgets tighten, other customers fall behind, and a late payment becomes a default before anyone calls.
Portfolio averages hide both. A prepayment curve for the whole book cannot tell you which borrower is about to leave, and a collections queue sorted by days past due cannot tell you who would repay with a small change to their plan. Every model you use for credit decisions also has to pass review: your risk team and your regulator will ask what drives each prediction and how it behaves when conditions change.
How AI debt repayment modeling works
Your credit analytics team builds repayment models in H2O Driverless AI on the loan and payment history you already hold, and your lending, retention and collections teams use the scores to reach each borrower with the right option at the right time.
Step 1: start from loan and payment history
Join loan terms, payment history, balances and rates to customer relationships and channel activity. For securities, add pool-level loan characteristics and the rate paths behind past prepayment speeds.
Step 2: predict repayment, prepayment and default
H2O Driverless AI trains classification models for outcomes such as default or early payoff, and regression models for prepayment rates on pools of mortgage loans. It compares interpretable models such as logistic regression with neural network and tree models, and keeps the one that scores best on your validation data.
Step 3: predict who will apply or refinance
Propensity models estimate whether a customer will apply for a mortgage, or refinance an existing one, over the next month, quarter or year.
Step 4: explain and stress each model
Explainable AI and ML interpretability show reason codes for each prediction, and sensitivity analysis shows how the output moves as each input changes, for your model risk team to review before the model is used.
Step 5: act with the customer in mind
Scores feed the lending, retention and collections systems you already run, rather than replacing them.
Prepayment risk beyond individual loans
The same modeling approach covers pools of mortgage loans behind fixed-income assets such as mortgage-backed securities, where a regression model on the pool's rate path and loan characteristics predicts prepayment speed for the security as a whole, not only for a single borrower.