Machine learning models can help distill complex datasets into accurate decisions, driving better small business lending decisions.

According to the SBA, there are over 30 million small and medium-sized businesses (“SMB”) in the US and these companies drive nearly 50% of total U.S. employment.  Continued small business growth is critical to the economy, and SMB financing is an important driver of this.  

Small businesses need access to financing for a variety of reasons, including working capital fluctuations, inventory requirements, major purchases (equipment, real estate, etc.), revenue seasonality and growth investment opportunities.  In today’s market, financing is offered by both banks and non-bank lenders, with banks generally working with larger, more established SMBs and non-banks often working with the smaller and less established SMBs that receive less attention from banks.

The fundamental challenge of small business lending is simple: there’s often a lot of work to be done for a relatively small and risky loan.  As a result, many banks avoid small business lending and instead focus on consumer products (e.g. personal loans, auto loans, mortgages, etc.) or larger commercial loans (e.g. $5M+ loans to large profitable companies).  However, this challenge also creates an opportunity – lenders that identify ways to efficiently serve SMBs while maintaining high-quality underwriting can build highly profitable and scalable businesses.  

Machine learning – the ability for computer programs to learn from historical data without needing to be explicitly programmed – is playing a key role in helping small business lenders tackle this challenge, driving better decisions, lower costs and improved customer experiences.  This post explores a few ways that SMB lenders can embrace machine learning to improve their businesses.

1. Improved Credit Underwriting

In small business lending, one of the biggest opportunities for driving efficiency is underwriting automation.  Small business lenders typically need to collect data from a variety of data sources (including both company data and personal data about the business owners), clean these datasets, extract the important variables, determine financing eligibility and set pricing.  In practice, this often requires a back-and-forth process with the applicant to collect the required information and, unlike consumer loans which largely rely on uniform credit bureau data, collecting business information often results in inconsistent data that needs further analysis.

Whether it’s an unsecured loan, line of credit or asset-backed loan, machine learning can help SMB lenders improve their underwriting and credit assessment practices.  Machine learning is used to automatically analyze historical loan data and train models that make predictions, such as the probability that the loan will default within a given time period.  This prediction can be used to determine eligibility and to set risk-based pricing.

Machine learning has enabled many lenders to go beyond just looking at basic credit metrics and allowed them to create advanced risk profiles for prospective borrowers. Particularly for lenders with a high volume of historical loan performance data, machine learning can effectively create credit predictive risk models that better assess applicants and lead to lower default rates on loan portfolios.

2. Fraud Detection

Fraud is a key risk in small business lending, and machine learning can help.  Predictive models can be trained to assess fraud risk by analyzing historical incidences of fraud to uncover hidden relationships between application characteristics and fraud risks.

Most fraud models we’ve trained end up including a wide variety of factors and, given enough data, can be extremely accurate.  The data used to train a fraud model is typically a combination of stated application information, credit bureau data (about the business owners), verification information collected during the process and third-party information from fraud data providers.  The result of a model is typically a score or percentage probability of fraud and, armed with these predictions, the lender can immediately approve, deny or flag the application for manual review.  This results in only a fraction of applications requiring human intervention by a loan officer.

The biggest obstacle to using machine learning for fraud detection is data. Training an accurate prediction model typically requires (at least) hundreds of verified fraud cases, which may not be available (note: a “verified fraud case” can be either a historical incident of fraud or previously confirmed fraudulent application that was rejected).  In those cases, we suggest starting with off-the-shelf fraud scores (provided by providers such as LexisNexis or TransUnion) and training an in-house model over time, with the goal of eventually relying exclusively on the in-house model, which will predict the lender’s specific fraud risks better than generic scores.

3. More Effective Customer Acquisition

Given the small size and relatively short duration of small business loans, efficient customer acquisition is very important for SMB lenders.  For small business lenders in the US today, customer acquisition costs are a major expense and companies are seeking ways to make their customer acquisition more effective, both by increasing efficiency in reaching out to new prospects and more effectively retaining existing customers.

To combat this issue, machine learning models can be used to trained to help lenders optimize marketing spend. For example, sending physical mail advertising loans to potential applicants is a common marketing strategy and lenders can track responses to these mailers and develop a collection of response rate data over time. Applying machine learning to this data allows lenders to determine which potential customers are most likely to respond (i.e. the model will predict the “probability of response”) and prioritize their contact lists to send mail to the highest potential leads.  Similarly, lenders can use machine learning to determine marketing channels, optimize marketing messages and identify cross-sell opportunities to existing customers – all resulting in marketing lower costs and improved customer acquisition.

4. Workflow Optimization

Small business lending often involves significant back and forth between the lender and borrower, and loan officers need to carefully manage this process to close the transaction. This usually involves manually processing applications in a simple queue, which can result in sub-optimal allocation of the loan officer’s time.

As SMB lenders gather data, they can use automated machine learning to train models that recommend the “next best action” to loan officers. This lets them work through their queues dynamically, based on the opportunities that are most in need of attention. The prioritization can be based on a variety of predicted factors, including the likelihood that the deal will close, expected profitability of the loan, time-based factors and more.

Using machine learning to prioritize workflows lets loan officers focus on processing, which accelerates response rates, reduces costs and improves customer experiences. Further, by centralizing aspects of the decision-making, small business lenders can ensure that staff are always focused on the right applications, creating more consistent outcomes.


Small business lending is defined by the constant struggle between efficiency and loan-level profitability, and machine learning is being leveraged to increase automation, improve decisions and drive better customer experiences. Fueled by large historical data sets, predictive machine learning models provide opportunities for small business lenders to create sustainable competitive advantages that enable profitable, long-term growth.

About DigiFi

DigiFi is a technology company that helps businesses make better automated decisions.

Our platform lets businesses easily use automated machine learning and rules management to optimize critical decisions with no coding or technical expertise required. Repetitive work that used to take hours can now be completed in minutes, letting your team focus on what matters most.