With over 30.2 million registered companies in 2018, small and medium-sized businesses make up 99.9% of all US businesses. They’re a powerful driving force of the economy, accounting for 1.9 million net new jobs in 2017 and nearly 50% of total US employment. Their sustained growth is critical to future economic developments. Yet, ironically enough, our banking system is unable to meet their funding needs.
Small businesses have smaller capital needs. Taking into account the risk factor, the long review process and the amount of work needed to be done, small business loans generally don’t pay off in the traditional lending system. As a result, banks largely avoid lending to small businesses, turning their focus to well-established SMBs, bigger commercial loans, and consumer products. On the other hand, small businesses rarely have the capital and the structure needed to carry them through a months-long loan review process, at the end of which they’re likely to be declined.
But these challenges have given way to opportunities, thanks to the advancements in the field of Big Data and Artificial Intelligence. The alternative lending market, driven by financial technology (FinTech) startups, is helping bridge this gap and providing an expanding number of small businesses with the necessary growth capital. Currently, machine learning has taken on a key role in reshaping the lending industry and consequently small businesses as well. Here’s how machine learning models can be used to create new opportunities and vastly improve small business lending.
Expanding the pool of creditworthy borrowers
The major issue with lending to small businesses thus far is that traditional lending systems examine a very narrow range of data points during the application process. More often than not, a small business is not going to be creditworthy as a result of being judged solely by criteria such as Paydex scores. This has proven ineffective in this day and age when various factors need to be considered in order to paint a more accurate, nuanced image of creditworthiness.
That’s where machine learning comes in, helping review extensive datasets and a significantly larger number of scenarios which aren’t even considered in the traditional system. By using machine learning models, lenders can review all kinds of relevant information, from Yelp scores to real-time shipping trends, and obtain enhanced real-time data. That way, small businesses can be reviewed more fairly and given a chance, while lenders can efficiently collect data beyond basic credit metrics in order to create advanced risk profiles.
Using predictive models to improve credit underwriting
The extensive time it takes to review loan applications is another major complaint among entrepreneurs, whose companies, as we’ve mentioned, rarely have the cushion to wait through the process. Even in the traditional system, small business lenders need to collect data from a variety of sources, including both the company data and the owners’ personal data in the review. Then these datasets need to be cleaned, important variables extracted, after which the lenders can determine financial eligibility and set the pricing.
This is not only a lengthy process but it also requires constantly getting back to the applicant to collect missing information. In practice, lenders also often end up with inconsistent data that demands further analysis.
Underwriting automation is making the biggest impact on the lending industry, by helping lenders enormously speed up the process while simultaneously factoring in a larger number of variables.
Small business lenders can use machine learning algorithms to establish predictive models to analyse historical loan data and streamline the underwriting process. For example, a model can be trained to calculate the probability of a certain loan defaulting within a given timeframe. This prediction will be used to effectively determine financial eligibility and set pricing.
Ultimately, these high-tech models can help reduce the total amount of time it takes from the moment businesses apply for a loan to the time they actually gain access to capital. Along with being a key solution to a major problem for applicants, this also means improved profitability for lenders as it would introduce significant cost efficiencies.
Improving on-going loan monitoring
Once a loan is approved, the amount of customer information expands rapidly. Without an efficient loan monitoring system at hand, lenders risk missing out on key information that could help them identify signs of loan trouble at an early stage. Machine learning provides lenders with an advanced system that would help them track and analyze large amounts of customer data. With this kind of advanced and thorough monitoring, they’re able to pinpoint issues and step in before a borrower defaults, which ultimately also makes SMB lending more profitable.
Assessing fraud risk
Machine learning provides an effective solution for fraud detection across numerous industries. It’s proving to be especially important for SMB lending, where fraud is one of the biggest risks. Predictive models are trained to assess fraud risk by analyzing historical incidences of fraud, and the more data used for training, the more accurate predictions. The idea is to analyze varied data that helps uncover relationships between fraud risks and application characteristics. This data includes factors such as application information, credit bureau data about the owners, third-party information from fraud data providers, collected verification information throughout the application process, etc.
The model presents the results as a score or percentage probability of fraud, which lenders then use to immediately assess an application or flag it for manual review. This is not only an effective way of detecting fraud, but it’s also extremely important for accelerating the process, since only a fraction of applications would end up requiring intervention by a loan officer.
Optimizing workflows
All in all, machine learning models can help optimize workflows and streamline practices throughout every step of the process. For example, models can be trained to recommend the next optimal actions to loan officers to help them save time and work through their queues efficiently. By using predicted factors and calculating the probability of closing a deal, these models will help officers prioritize workflows so that they can focus on processing. They can be implemented at various stages and in all kinds of ways, which ultimately results in cost efficiencies and improved customer experiences.
In conclusion, it’s safe to say that the lending industry is being changed from the core by the digital marketplaces that are harnessing the power of machine learning. Because the benefit is mutual for the lenders and the borrowers, there are certainly exciting developments to expect as well as new ways of implementing machine learning in the industry.