Use Machine Learning To Grow Your Lending Business
This post will quickly outline how your can financing business can use machine learning to ...
- Document and execute a consistent lending strategy.
- Discover which attributes most significantly affect deal performance.
- Reduce your susceptibility to fraud from both internal and external actors.
What Are We Doing?
In our experience, the most beneficial application of machine learning (ML) in financing organizations is the evaluation of incoming deals. In short, we want to know, before you execute a deal, if that deal looks and feels like deals that have made your business money in the past.
Who Does This Apply To?
Not all financial services companies can immediately benefit from a deal evaluation system like the one described above. In our experience, the following businesses can get the most benefit in the shortest span of time.
- Mortgage companies (banks, hard-money, sub-prime, jumbo, boutique, etc.)
- Auto finance companies (banks, hard-money, sub-prime, etc.)
- Personal and secured lenders
- Business finance companies
These business all have something in common. Relative to other lenders, they are doing a high volume of numerically comparable deals.
How Does This Help?
This section will detail how a custom deal evaluation system can help you business.
Documentation and Execution
We find that a lot of businesses are doing well, but could be doing better. Oftentimes, they are doing well because they have one or more superstar deal makers that close a high volume of mostly successful deals.
Those people are essential, but sometimes their managers and executives wonder ...
- What is his/her criteria for a good deal?
- Is he/she hiding under-performers in with over-performers?
- If I got all my deal-makers in a room together, would they agree on what a good deal looks like?
Through analysis and interviews, we often discover unsettling answers to these questions.
Now, imagine you are in the future, and your business has access to some information like the following ...
| Credit Score | Deal Performance |
|---|---|
| < 600 | -28% |
| 600 - 650 | -11% |
| 650 - 700 | +0% |
| 700 - 750 | +9% |
| 750+ | +17% |
| Debt to Income Ratio | Deal Performance |
|---|---|
| < 10% | +13% |
| 10% - 30% | +3% |
| 30% - 50% | -4% |
| > 50% | -16% |
These figures can be derived for these attributes and many more across the breadth of data your organization already possesses. Using this information, you can document and execute a simple and repeatable lending process.
If we embed the results of our ML research into a web application, your lenders will always be a few clicks away from a straightforward prediction of deal performance.
Discover Important Attributes
Through our work with various lending organizations, we often find that they are asking for information from their customers and vendors that doesn't have any significant effect on deal performance.
For example, certain lending businesses rely and indices and scores compiled by various information vendors in their industry. We occasionally discover that these scores are statistically insignificant to their deals, saving money and admin time.
Conversely, we sometimes discover that often-ignored attributes, or new attributes derived from existing information, can be very important predictors of deal performance.
Reduce Susceptibility to Fraud
In the course of doing business, it is likely that some customers (and maybe some employees) have fraudulently reported something. Most likely, what they fraudulently reported is some attribute that is difficult to verify.
When we develop machine learning models for your lending data, we will implicitly discover which attributes have a significant effect on the deal performance. Attributes can be insignificant for one of two reasons ...
- The attribute genuinely doesn't affect the deal outcome (i.e. how many dogs the customer owns.)
- The attribute is fraudulently reported so frequently that it appears no different from statistical noise.
It goes without saying why it is helpful to know these attributes. Your susceptibility to fraud is related to how strongly you rely on bad data to make decisions, not how often applicants choose to lie.