Business Development Companies: Opportunities In Machine Learning
BACKGROUND: The BDC Reporter was approached by Ian Foley, the CEO of acuteIQ, asking if we’d be interested in an article about the opportunities for Business Development Companies in harnessing the capabilities of machine learning to enhance loan and borrower identification. Given that much has not changed at many lenders when it comes to prospecting for new loan opportunities since we were pounding the streets (or driving around to endless meetings to be more exact) thirty years ago, this seemed an interesting new perspective. What’s more, the article is a welcome change-in the midst of earnings season-from our regular diet of earnings, under-performing loans and capital raises. So we offer up-under our Other Views rubric-this fascinating potential new approach that commercial lenders-including BDCs-may be adopting in the years ahead:
“The lower middle-market is often characterized as being inefficient for lenders, and therefore a good opportunity for new entrants like BDCs. However, getting to scale in the market often trips up new lenders … but, by applying machine learning, lenders can drive significant efficiencies and make this a profitable market segment.
The lower middle-market is defined as companies between $10 million and $100 million in annual revenue, with lending requirements of under $30 million. According the US Census Bureau, there are 32,000 companies that fit this criteria.
One of the key challenges lenders find in this market is sourcing high quality deal-flow. In particular, without sponsors to filter opportunities how does a small BDC team cover the ground outside their own network of contacts? Banks typically addressed this challenge using their branch network and a number of new lenders are investing heavily in online advertising; but BDC team’s with ten or less staff focusing on origination is going to have a hard time to cover even a niche-based investment approach. Another challenge is the due diligence process, which for private companies that all have different documentation management and financial accounting processes, means identifying positive or negative signals from the data is time consuming or open to human error.
One approach to address these challenges is to use machine learning, a subfield of Artificial Intelligence that is defined as the ability for machines to learn without being explicitly programmed. In the financial service industry, machine learning has been used for many years, including fraud detection (e.g. Visa uses machine learning to detect fraudulent credit card purchases) and more recently Robo Advisors (e.g. Charles Schwab’s Intelligent Portfolio). The lending industry has only started to use machine learning in the last five years, with companies like Kabbage using machine learning to quickly assess a borrower’s eligibility for funding.
Extending this to the lending process of origination and due diligence, has traditionally been hampered by the lack of a training data set. Getting an accurate training set is key for machine learning, since unless the underlying data is accurate, predicting trends from the data will be ineffectual or give lots of false positives. At acuteIQ, we have addressed this by spending the last 2 years building a database of 19 million companies, with over 500 million facts of information, with the ability to identify unique ‘signals’ of businesses that are in the market for a commercial loan.
To support the origination process, lenders have used acuteIQ’s platform, branded LenderIQ, to better understand the market to find untapped niches and then identify specific companies to start prospecting, which has resulted in an average of 4x increase in relevant deal volume. For example, a new alternative lender with over $100 million loan portfolio, used LenderIQ to identify three new industry segments to pursue (e.g. Turkey Farmers). Machine learning was used to identify the ideal customer profile across LenderIQ’s database and then as borrowers started to convert, machine learning was used to find more businesses with a ‘look-alike’ profile. For another client, LenderIQ was able to identify ‘signals’ of loan intent by finding trucking companies that had rigs with an average of 400,000 miles and then using this data to propose a customized equipment finance loan. Also, LenderIQ’s data has helped SBA lenders identify gaps in the market by mapping lender data by industry, loan type and duration to target new businesses that had 6 months left on a fixed loan.
Machine learning can optimize the due diligence process, too. Reviewing borrower loan documentation, identifying inconsistencies in financials or verifying compliance are all processes where automation can cut costs and speed up decision-making. For example, a bank used LenderIQ to cut the borrower review time by 20%, which ensured they were the first lender in a competitive process to respond with a loan offer. LenderIQ was able to ingest the structured and unstructured data, which was in the form of different formats (e.g. .PDF, .XLS), and then used the banks approval check-list to pull out the key data, which was then provided to a loan officer for review.
Machine learning is definitely not a panacea, and it will be sometime before there is a Robo Lender that can handle each step of the loan life cycle. But, in the meantime, lenders can take advantage of machine learning to more effectively tackle the lower-middle market and develop a competitive advantage. Artificial intelligence is one of the persistent thematic trends that will influence the lending industry over the next 20 years – BDC executives just need to work out when they want to get on board”.
Ian Foley is the CEO of acuteIQ, a technology platform that uses machine learning to help lenders source new business customers in the lower and middle-market. acuteIQ’s product is deployed in institutional and mid-market lenders, providing a 4x improvement in customer acquisition. He can be reached at firstname.lastname@example.org.