Together, artificial intelligence, machine learning, and deep neural networks are a boon to businesses awash in data.
Machine learning has flipped the script on traditional lending, allowing for more accurate and faster decisions by shifting traditional decision-making from analysis of individuals to analysis of trends and patterns.
The result for lenders? More repeat business, and lower operational costs.
These outcomes matter in a world where technology-enabled financial service is shaking traditionalists to the core. Echoing internet pioneer and fintech investor Mark Andreessen’s pronouncement that “software is eating the world,” Goldman Sach says fintech is poised to gobble a good third of the annual revenue of traditional financial-service companies.
Reflecting this transformation, the global digital-lending platform market expected to approach $20 billion by 2026 for a compound annual growth rate of 19.6% through the seven years prior.
In the wake of mission-critical benefits such as faster decision making, happier customers, and overall cost savings, machine learning also confers on lenders a range of ancillary leverage points including:
- Higher processing efficiency
- Streamlined compliance
- Efficient analysis of data in large volumes
- Enhanced accuracy
But how does machine learning work? — and how do lenders actually use it? Let’s start on the road to answering those questions by defining some terms.
Machine Learning and Artificial Intelligence
Machine learning is a subset of artificial intelligence, which is a functionality (some call it a “device,” others a “process”) that takes into account aspects of its environment to make decisions or predictions, mimicking human cognition.
Machine learning supports artificial intelligence. It does so by using algorithms and statistical models to perform many specific if-this-then-that type tasks virtually at once, drawing on patterns and inferences rather than explicit case-by-case instructions. In other words, machine learning takes relevant inputs (AKA “training data”) and constructs mathematical models that bring “thinking” to nuanced processes requiring multiple inputs from vast datasets — such as determining a would-be borrower’s suitability for a loan of a specific size, type, and duration
A deep neural network is a method of machine learning. It functions in the realm of artificial intelligence like a dating app for data. Here’s the idea. A neural network sits between multiple datasets coming in (inputs) and decisions or predictions based on that data and the underlying training data (outputs). Its function is to apply the correct formula to the type of data in question.
Cars, Motorcycles, and the Cost of Bad Decisions
For example, if the task at hand is to categorize, in separate groupings, types of cars and types of motorcycles based on photographs of these vehicles, the neural network is what makes sure car pictures are scanned for the physical characteristics of cars rather than motorcycles, and vice versa. Typically neural networks process many input types at the same time (not just two, as in the example above) — which is one reason these brainy systems are so “deep.”
Together, artificial intelligence, machine learning, and deep neural networks are a boon to businesses awash in data — such as lenders — because they help these enterprises identify, sort, and make accurate decisions based on multiple data points from multiple datasets, rapidly and often simultaneously.
And of course, this is beneficial to businesses that are looking to distinguish themselves from competitors, digital and not, with the speed and (above all) accuracy of their loan decisions. As far back as 2015, research firm Javelin Strategy found that “false declines” — loans not granted for due to faulty data interpretation — impacts as much as 15% of US consumers and costs lenders a cool $118 billion a year.
Machine Learning Takes the Guesswork Out of Lending
So how are artificial intelligence, machine learning, and related concepts changing how lenders make loans? Well, let’s answer that by comparing traditional lending practices with more up-to-date approaches.
Traditionally, the credit-worthiness of prospective borrowers was determined by scorecards. This approach, characterized by elevated levels of interpretability, and guided by a combination of economic theory and subjective business intuition, has several advantages, including accuracy and ease of oversight. On the downside, scorecard methodologies simply can’t handle big-data inputs.
That was OK when lenders were content to sort through a limited number of data sources for information on loan applicants — things like loan applications, the lender’s internal databases, and credit-bureau scores. But now there’s a flood of additional data sources on prospective borrowers, including social networks, mobile devices, payment systems, and web activity.
Using Machine Learning to meet the Challenges of Big Data
“These sources are highly relevant to lenders determined to gauge the credit-worthiness of would-be borrowers with a higher degree of accuracy, but the information they want would remain locked away in datasets too vast and unwieldy to get at without help from machine learning,” says Boris Teplitskiy, head of risk at TurnKey Lender, a financial technology company that specializes in loan-servicing software that prominently features artificial intelligence. “The lending industry has a big-data problem — and machine learning, quite simply, is the solution.”
And there’s more innovation in store for lenders as fintech continues to crush barriers. Augmenting Andreessen’s 2011 view that “software is eating the world,” his colleague Angela Strange, a general partner at venture-capital firm Andreessen Horowitz, recently added that fintech is doing the same, as innovative tech companies transform how companies devise and distribute financial products and services.
One effect of the fintech revolution, writes Strange, is that “People who were previously hard to gauge now become new customers.”
Are you currently using machine learning in your lending processes? We would love to hear how!