Machine Learning Can Support ‘People Helping People’ Mission

Computer programming has come a long way since the late 1970s and early 80s, when high school students learned BASIC and wrote simple programs on Commodore 64s. Today, six-year-olds learn coding from LEGO kits purchased at Target and young entrepreneurs are earning millions writing apps and selling them to global fintech companies.

Until recently, however, one thing had remained constant for nearly 40 years in computer programming: Humans had to first write explicit rules and models.

AI, Machine Learning and Your Data

Artificial intelligence doesn’t require explicit rules. Instead, computers perceive their environments and make decisions that will increase the likelihood of achieving their goal.

Machine learning is a subset of AI. It gives computers the ability to learn – usually by providing statistical data – without being programmed every step of the way.

Credit unions are ripe with data, which means machine learning can bring a lot to the table to the credit union community. We’ve been hearing for years that credit unions should do more with their valuable data, and now we have a golden opportunity to leverage that data to improve our operational efficiencies and analytics in multiple areas.

Early adopters in financial services are already using machine learning to a competitive advantage. In just a couple of years, machine learning software will be widely adopted as a general practice, and within five years institutions that aren’t using it will find themselves at a competitive disadvantage.

If you’re in charge of your credit union’s five-year operational, financial or technology plan, that news may come as a surprise. Thankfully, you don’t have to develop the technology yourself. Forward-thinking vendors are already providing machine learning solutions and creating new ones. You have the opportunity to partner with a provider within the next few years, and apply machine learning in a way that caters to your specific needs and improves both your bottom line and member experience.

CU Direct has been exploring new ways to effectively leverage and incorporate machine learning at credit unions in our Innovation Lab.

The Benefits of Machine Learning Technology

For lending executives, the development of machine learning applications that assist with processing and quality control during the loan funding stage can have a dramatic positive affect on your lending process. Even if you’ve automated your lending process, when it’s time to fund a loan, someone still has to manually “stare and compare” key information on the loan docs to make sure it matches what is on your system. You know the drill: Make sure the name is spelled the same, the interest rate is correct, the loan amount is within your lending policy parameters, the assigned due date is correct, etc.

Machine learning can automate most of that process. Machine learning even has the ability to enhance optical character recognition (OCR) and automatic document recognition (ADR). For example, legacy OCR systems may not be able to distinguish a “5” from an “$.” With OCR, character recognition is black and white, but with machine learning, if you tell the computer it’s a $, it will retain that decision moving forward.

Further, machine learning can deliver a quality control confidence score, informing you with how much certainty each field in your loan package is correct. You have the ability to adjust the acceptable confidence score to increase loan processing efficiency. The interest rate will need a 100% confidence score, of course, but to speed things up you may decide a loan can be approved with a lower confidence score in the email address field.

The result is a system that can process more loans without having to hire additional staff. Your credit union won’t spend employee hours on quality control, so staff can dedicate more time to other areas and needs.

Other potential machine learning solutions, when built upon AI, can effectively analyze data to improve underwriting, loan portfolio management, marketing, cross-sales and the member experience.

Let’s look at underwriting as an example. Most lenders currently depend upon the analysis of a borrower’s credit history and other data like credit scores to determine creditworthiness. Machine learning can expand these metrics to include an applicant’s entire financial footprint – things like credit card spend, deposit account activity and other digital footprint data – to more accurately predict borrower performance.

I’m sure it comes as no surprise that Amazon is using its vast database of sales histories and customer product reviews to identify companies participating in its marketplace that can be offered small business loans. Sales through Amazon can suggest which companies are experiencing robust growth and may need capital to increase production if deliveries are delayed. And if a product is especially popular (or unpopular) among customers, it would contribute to the company’s potential for longevity and growth.

Similarly, to properly value a loan, significant data in addition to borrower creditworthiness must be considered, such as fluctuating collateral value, economic predictions and other factors. Machine learning can not only analyze all of these data sources together to produce a coherent decision, it can also learn from the data to make better predictions about future loan performance.

Machine learning can also identify new correlations among member subsets that can be used to improve marketing. For example, this technology can determine which owners are likely to buy a new car based on their current car model or that pet owners who spend more than $300 per month on supplies are more likely to take out and use a credit card (purely hypothetical examples).

The beauty of using machine learning in this way is that it combines the digital ability to analyze large collections of data. The human brain simply processes information thousands of times slower than a computer. Machine learning algorithms can crunch millions of fields of data for days on end to uncover correlations and insights that we may never discover on our own. With these answers, we can further refine the questions to produce even better results. Imagine how much time that would take one of your in-house data analysts – time that would be taken away from other strategic initiatives.

Preparing for Machine Learning

What can credit union executives do to prepare for these powerful new tools? The most important step is to clean up your data. The old adage “junk in, junk out” still applies. Machine learning may be able to infer a more intelligent answer, but it does not think abstractly. It must have clean data from which to work.

The second step is to prepare for the cloud. Cloud solutions make machine learning ventures more affordable and effective. Credit unions don’t currently have to host their systems in the cloud, but they should become cloud-ready to take advantage of hybrid systems and/or leverage APIs to take advantage of emerging solutions.

The benefits of this technology speak primarily to the bottom line, but they also support the credit union philosophy. For example, when applied to underwriting, machine learning can be used to expand credit to borrowers who lack a credit history or whose credit score may not accurately reflect propensity to repay. Freeing up staff from mundane quality control tasks allows them to spend more time assisting members, ensuring quality member service and experience.

Simply put, machine learning can free up your team to provide even better service to your members.

Brian Hamilton is Vice President, Innovation for CU Direct. He can be reached at brian.hamilton@cudirect.com.