How We Do It

When our founder Douglas Merrill (ZestFinance CEO and former Google CIO) first encountered the challenge of scoring a subprime population, he immediately thought of Google’s high-dimensional machine learning search algorithms. This technology is optimized to handle missing and changing data because it consumes thousands – if not hundreds of thousands – of variables. He believed these same algorithms could work for consumer credit – and they can.

Traditional Underwriting Models are Limited

Everyone uses the same methods to underwrite. They either use logistic regression, decision trees, or a combination of both. Logistic regression can only handle 10 – 15 variables; additionally, all variables must be present (and correct). Decision trees require it be possible to divide all applicants into mutually exclusive categories.

Neither of these constraints really make sense. And when either is violated, the decision is incorrect. This isn’t surprising -- everyone knows these models are imperfect. However, they are not only imperfect, they yield incorrect answers, especially for the underbanked.

ZestFinance’s big data approach avoids these pitfalls.

All Data is Credit Data

ZestFinance uses machine learning techniques and large-scale data analysis to consume vast amounts of data and make more accurate credit decisions. When you analyze trends across thousands of signals, your ability to underwrite massively improves.

The ZestFinance decisioning infrastructure runs dozens of individual underwriting models in parallel and returns underwriting decisions in moments. Here’s how it works, at a high level:

The Math Matters

ZestFinance Co-Founder Shawn Budde has worked in the consumer credit space for 25 years, much of it at Capital One – arguably the premier underwriter in the subprime space. When he started working with Douglas, he was, in his own words, “skeptical that Google had anything to teach me about underwriting.” His skepticism was well-founded. Every few years, some company comes out with some form of “underwriting in a box,” and it never really works.

Now, Shawn is the strongest advocate for the Google-style technique because it outperforms all the best-in-class models he’s used throughout the past two decades.

The math matters in big data. If you get the math wrong, the data is useless or completely misleading. We get it right. ZestFinance has combined the best data modelers in the world with the best credit analysts in the field to build a big data model that dramatically improves underwriting quality.

To learn more about how our underwriting model can dramatically improve your credit scoring, get in touch with our big data experts at service@zestfinance.com.