Credit Modeling in the Age of AI
Traditional credit models rely on a limited number of credit attributes and linear regression techniques which often result in limited predictive power and higher default rates.
Utilizing historical data coupled with underwriter feedback on previous model variance cases, we have developed a unified AI platform for decisioning and pricing loans across all channels, which has achieved significant improvement in accuracy.
Custom Non Linear Models
Our proprietary algorithms take a more sophisticated approach by evaluating thousands of applicant details, the past performance of your loans, and the expertise of your underwriters. We knew that in addition to being non-linear, our models needed to be dynamic so that they continue to learn after they are deployed. This learning happens through a continuous feedback loop from loan outcomes as well as underwriter evaluation.
Like humans, our models never stop learning, they continuously optimize themselves. This is the definition of modern artificial intelligence, not algorithms that think, but algorithms that learn and evolve.
Get started with Underwrite
Develop a custom model
Our platform processes portfolio data of cured loans—classified as Good or Bad based on factors such as status (Paid Off, Charged Off, Defaulted, Late, Collections) or profitability. We then train multiple types of machine learning models, which compete for the highest accuracy in predicting loan outcomes. We then combine those models to build the production model.
Process new application data
New application data is fed into your custom production model to determine the probability of good outcomes and assign a predictive score.
Beyond providing a highly accurate predictive score we can then chain together optional models for decisioning, pricing, denials and more.
Feedback and Retraining
Feedback is continously provided on the performance of loans after the fact and that feeds back into retraining the model. Our models continuously learn and evolve through a feedback loop, and adapt to macroeconomic trends in real-time.
Chained Models
Underwrite.ai is unique in that previous models were machine learning models where our approach develops a true artificial intelligence model which effectively replicates the process of the skilled human underwriter.
The platform itself is an ensemble of machine learning ensembles. Working with your historical data we can chain together the below optional models based on your business needs and preferences:
- Score: The first model scores all applications
- Decision: Based on the predictive score we divide into Approve and Deny/Refer buckets
- Pricing: Approved applications are assigned a price
- Explanation: An explanation is provided for denied applications
Explore our resources about the use of artificial intelligence in financial services
Start for Free Today
Get in touch to set up your custom model and experience underwrite.ai free of charge for 30 days.