Underwrite.ai was founded in 2015 by Marc Stein based upon his experience developing underwriting systems for companies including JP Morgan Chase (consumer lending), Freedom Debt Relief (debt settlement) and LeaseQ (equipment leasing) over the past 20 years.
The genesis of the idea which led to Underwrite.ai began with work to correlated DNA microarray data with the probability of prostate cancer. This required the development of machine learning techniques which allowed for very wide and shallow data sets, e.g. 25,000 genes and 25 sample cases.
This discovery led to the question, what other types of assets could this algorithm be applied to? Consumer lending and small business lending immediately proved to benefit significantly from this approach.
Machine Learning and Our Technology
Machine Learning is a field in computer science that evolved from the study of pattern recognition and artificial intelligence. The focus of machine learning today is to create computer algorithms that learn from data and can make accurate predictions of outcomes based upon the patterns deduced within the data.
Unlike traditional statistical modeling, the predictive models of machine learning are generated by the computer algorithm, as opposed to determinations made by statisticians based upon their interpretation of the results of linear regression and related techniques.
At Underwrite.ai we take portfolio data of cured loans (or other instruments) and classify the cured loans as either Good or Bad based upon factors such as status (Paid Off, Charged Off, Defaulted, Late, Collections) or profitability. We then train models based upon a wide array of algorithm types which then compete against each other for the greatest accuracy in predicting the outcome of loans. We are then able to feed in new application data and determine the probability that a given application will have a good outcome. We can then classify applications into tiers (or rates) based upon their probability of poor performance.
In a large body of peer reviewed scientific analysis (and validated further by our experience) it has been found that this methodology offers very significant improvements over the the standard techniques of underwriting using linear regression based scorecards.
Traditionally, this type of analysis was impractically expensive and too slow to be used effectively in realtime banking applications. Much of our work at Underwrite.ai has focused on optimizing the process through distributed processing in the cloud to allow for this technique to be applied in realtime at minimal cost. This has brought us to the point today where we can decision applications at a cost starting at $2 per which typically can reduce defaults to one fourth their baseline performance.