AI Case Study: QCash

This is an excerpt from FinAncIal: Helping Financial Executives Prepare for an Artificial WorldGrab your copy here!

This case study covers the leading player in AI-based small-dollar lending for credit unions, QCash.

Stay tuned to the CU 2.0 blog while we feature more excerpts from our financial guide to AI!

Purveyors of short‑term credit and small‑dollar loans have gotten an understandably bad rap in the last couple decades. And it makes sense—before regulations, some interest rates on these loans went as high as 1,000%. Even today, short‑term loan interest rates climb over 600% in some states. In many ways, the reputation that short‑term credit earned was, well, earned. But it doesn’t have to be that way. And frankly, it shouldn’t be.

The fact is that in 2020, the apparent health of the economy and the unemployment rate belie the financial reality for most Americans. Nearly 80% of American workers live paycheck to paycheck. Almost 40% of Americans wouldn’t afford a $400 emergency expense. While the economy is certainly benefiting some, those benefits aren’t necessarily trickling down to everyone.

QCash entered the short‑term lending marketplace to deliver necessary services to credit union members. They used (and recommend) interest rates only slightly higher than normal personal loan interest rates, which kept things fair for borrowers and worthwhile for their credit union.

 

Understanding QCash

QCash was founded in 2003, when the Washington State Employee’s Credit Union CEO noticed that their members were paying millions of dollars in fees per year at nearby payday loan centers. That presented a few problems:

The high interest rates at payday loan centers were jeopardizing borrowers’ long‑term financial viability.

Credit unions were failing to provide members with critical small loan services.

Credit unions were missing out on potential loan interest revenue.

To provide a safer alternative, they developed a system for underwriting and funding quick loans.

Their system relied on traditional measures of creditworthiness, such as FICO scores. It proved clunky and unprofitable for years. Loans took too long to underwrite and fund, and the whole program was costing them money. Using FICO scores, many creditworthy people didn’t get approved for loans for which they should have qualified. Other times, people with no business borrowing got approved for loans they struggled to repay.

QCash knew they had to do away with that. They were missing too many opportunities to help their members. In 2015, QCash realized that they had the data to expand their loan offerings even further. Analytics powered by machine learning allowed QCash to look beyond FICO scores for determining creditworthiness. Instead, they used broader data, including less‑tangible statistics based on the relationship between the member and the credit union. Now, a member’s history with the credit union, their account types, and their banking habits all contribute to QCash’s automated underwriting algorithm.

Those underwriting models are fast, too. They’re faster than traditional methods of measuring a borrower’s creditworthiness, largely because they don’t need to pull any reports. The speed that they underwrite—and fund—loans is particularly useful for members who can’t wait for their cash. From application to full funding, QCash can deliver loans in 60 seconds.

When you consider how high‑risk short‑term credit is, their loan loss rate is particularly impressive. They stay in the single digits, with an average loss of 6–8%. They manage all this without resorting to predatory lending techniques or outrageously high interest rates.

 

How QCash Works

Okay, so QCash can underwrite and fund a profitable loan quickly by leveraging aspects of AI. But how did they do it?

Today, QCash uses Microsoft machine learning platform, Azure. QCash eventually settled on a model that was easy to explain to regulators—uncomplicated inputs and outputs kept them out of unnecessary trouble.

In fact, demonstrating algorithms, equations, inputs, and outputs has been an issue for regulators in the past. Yet, as machine learning increases in popularity and efficacy, people who work with the technology are getting better about showing their work rather than showing only their answers. So long as knowledge about and usage of machine learning increases, companies that rely on aspects of AI can rest assured that presenting, explaining, and defending their technology will get easier.

Fortunately for QCash, their simple machine learning model has been extremely effective so far. After a few years of trial, error, and feature engineering, their machine learning dropped delinquent payments from 7% to just over 5% over a 65‑day period. That’s an impressive 20% improvement over baseline.

 

How QCash Could Have Improved Their Process

QCash’s employees all started as expert IT and finance staff. They were in business, but they were not businesspeople. They knew how to build their product, but they struggled to sell it. Thus, the road to adoption was very slow—typical of community financial institutions. Still, running a data‑driven enterprise means that they have numbers to back up their claims. As they collected data, they proved that their service works.

On the technical side of things, QCash wishes they knew which direction the machine learning landscape was headed. As early adopters of the technology, they tried several platforms before finding a setup that they like. For example, although they currently use Python and Microsoft Azure, they didn’t start there. The whole landscape was much less standardized then than it is now. But as AI technologies become more commonplace, some platforms are clear frontrunners, and wading into AI projects is less of a crapshoot.

As it stands, though, there’s one bit of the early process that they don’t regret: timing. They could have waited a couple years to get started, but they didn’t wait until the technology was further ahead. Why? Because they knew how important data is to machine learning. They bought into the need for data early, and it’s paid off: they can show proof of their model predicting with more accuracy than the credit bureau.

 

What’s Next for QCash

One of QCash’s next moves to increase their explainability for underwriting, particularly with more advanced machine learning models.

As QCash grows, they’re also investigating ways to improve collections. On their short‑term road map, they’re working on a collections model for non‑real‑estate loans. Their hope is that they’ll be able to apply a better collections model for their expanding loan portfolio.

Finally, QCash aims to expand their use of analytics. They believe that more data will improve their accuracy. This goal—and this understanding of the primacy of data—aligns with the goals of every AI industry player. Plus, analytics will reinforce their underwriting capabilities.

One thing that QCash wants financial institutions to know is that machine learning isn’t scary. It still has a way to go as far as simplicity and accessibility, but that’s changing quickly. As far as the near future is concerned, they expect machine learning to take a more prominent role in the financial sector.

 

Want to Learn More About AI-Based Fintechs Working with Credit Unions?

CU 2.0 researches, works with, and speaks to countless fintechs. We’re happy to help you learn more about the technologies and providers that can take your credit union to the next level.

Contact us to learn more about our Quarterly Fintech Calls. We’ll call you to discuss a few of the fintechs, products, and strategies that have caught our eye recently. It’s short, it’s personal, and it’s free to credit union leaders.

But don’t forget to read the book! Check out FinAncIal: Helping Financial Executives Prepare for an Artificial World!