How WEEL Reached a Zero Percent Default Rate

WEEL is an online factoring company focused on the Brazil market. Two years ago, we set a seemingly impossible target: get to a 0% default rate without impacting growth. The default rate is the percentage of unpaid loans over the total amount lent out. To reach a 0% default rate, a lender needs to build a perfect decisioning system. Every loan needs to be repaid, and in factoring, there is no opportunity to hide. The duration of factoring loans are generally only thirty days, so the lender receives extremely quick feedback about the accuracy of the decisioning system.

Brazil is a country famous for high default rates, fraud, high credit risk, and numerous surprises. Despite the challenges, we believed that we could use advanced technologies, large data sets, and machine learning to build a world-class decisioning system that would enable us to reach our target of 0% default.

We are happy to report: mission accomplished!

What Did We Really Achieve?

If we analyze our unpaid loans versus our entire loan book over our entire lending history, our default rate is 0.5%. That is already an impressive achievement and certainly the lowest default rate ever achieved for any factoring company in Brazilian history. How then can we be so audacious to report a 0% default rate? The standard convention for computing default is to include late fees to offset defaults. According to that standard, our official default rate is actually -0.5%. For simplicity, we report 0%. Most people have never heard of a lender having a negative default rate!

How Did We Do It?

As with any significant achievement, the main ingredient in reaching our target was a lot of hard work and late nights at the office. Along the way, we also identified some other key learnings that made a big difference.

1. Partner for (Data) Success

Building a credit decisioning algorithm requires a lot of data. Pioneer fintech lenders like LendingClub or Prosper had to gather lending data the old fashioned way, by lending. Lending to acquire a data set for research is a very costly experiment. Today, fintech lending is more understood and even traditional players are willing to contribute data sets as strategic partners to help early-stage lenders refine their credit algorithms.

2. Trust the Machine

Modern lending is a data-driven science. An algorithm can always analyze more variables than a human. Machines don’t carry biases and they don’t have off days. It’s very tempting to challenge the algorithm-driven decision “just this one time.” Resist the temptation! Human error is a tremendous risk factor that can undermine even the best decisioning systems.

3. Look At Everything (even the metadata)

Good data scientists are like detectives. They need to analyze thousands of variables to identify the magic bullets that help drive the algorithm’s predictive strength. Sometimes the most important variables can be easily tossed aside. Metadata elements, which describe the dataset and the data history, are generally treated like trash to take out of the data set. With a bit of serendipity, we were surprised to find that metadata elements, like the pace of client data upload, were extremely relevant predictors of creditworthiness

4. Doomsday Might Happen

As data scientists, it is very easy to “fall in love” with our credit algorithms. We can test on gigantic datasets and prove that they work with 99% accuracy, but our tests generally assume normal circumstances. What happens if something unusual happens? In Brazil, “unusual” events are normal! Just a few months ago, truckers were on strike across the country and numerous sectors completely shut down. It’s important to stress test even the best algorithms against doomsday scenarios.

5. Don’t Rely on Collections

Lenders often take a view, “hey, if it doesn’t work it out, we’ll deal with it in collections.” This is a terrible mindset. By the time a loan reaches collections, the odds of recovery have diminished significantly. The decisioning team needs to take ownership to ensure that loans don’t reach collections. Applicants that are likely to cause collections headaches should be filtered out in the initial decisioning.

6. If there is a will…

WEEL means will in Portuguese. Above all else, we needed the will and stubbornness to believe that we could succeed. With a team of talented people focused together on a joint goal, even the impossible can be achieved, a 0% default rate in Brazil.