Are the credit models you are using to make lending decisions more than 2 or 3 years old? If so, you are likely making less than optimal credit decisions. You may be turning down a customer who is a good risk — while taking on customers who are more apt to default on their obligations.
Every year a model isn’t updated, its accuracy decreases. The economy changes. The consumer’s or business’s financial situation changes. Updating your models, using the most current data and attributes available, you can have confidence that you are making good credit decisions.
To make the most accurate credit decisions possible, many businesses are now turning to data-driven decisioning models that are powered by artificial intelligence (AI) within machine learning engines. While the standard regression model works well in some industries, the lift in predictive value from using AI data models can be very important in other industries, such as retail, fraud and marketing. These models use sophisticated algorithms to predict the customer’s future ability to repay their obligation, which means a much more accurate decision than traditional models.
Starting with High Quality Data
While data has always been at the core of credit decisions, models using machine learning are even more dependent on data. These models can be very accurate, but their accuracy depends on having the necessary data to understand what happened in the past and present behavior to make a prediction for what will happen in the future. The more data provided, the higher the accuracy of the decision. Here are three things to consider when building your data-driven decisioning model:
Clean Data – As innovation spurs business and technology to run faster and more efficient, the quality of the data underneath all of that innovation becomes even more important. Machine learning becomes smarter the more data it consumes. This means the accuracy of the credit decisions made by the model is largely dependent on the quality of the data provided. Data from third-party sources often contains mistakes, missing fields, and duplicate information, which results in less accurate credit decisions.
Correct Data Points – The accuracy of the results depends on considering the right criteria in the form of data points in the model. When you use machine learning and AI algorithms, they can predict which specific data points will help increase the performance of the model for the specific customer and the specific type of credit decision. Often, data points that you may not consider are the ones that can make a big impact on the accuracy of the decision.
Real-Time Data – In the past, there was often significant lag time between collecting and being able to use the data. By using real-time data with machine learning models, you can get a clear picture of the most current view possible and see changes in the different data points as they occur. This lets you make a much more accurate prediction of what will happen, with the consumer or business, than was previously possible with a traditional credit decisioning process.
Using Alternative Data to Get the Full Picture
Often, additional data — typically referred to as alternative data — that is not readily available from traditional data providers is used to enhance the accuracy and predictive ability of a model. While the model can seem complete without this information, the model may provide suboptimal results without it. Machine learning models can predict the situations and exact type of alternative data a model needs to produce an accurate decision. Experian offers a wide variety of alternative data that clients can use to improve decision models.
For example, a business owner may be taking out short-term loans to increase her cash flow, which makes her a much higher credit risk than she appears to be without this data. Weather information is also a common type of alternative data; a business located in Tornado Alley may need higher cash reserves to be a good credit risk. On the other hand, businesses located in an area impacted by a recent weather event, such as a hurricane, may be a good credit risk even with a lower score because both their business and local economy is recovering.
Regularly Evaluating Your Data Model
You must build in governance and make sure you are evaluating how the model is working on a regular basis, like having an annual checkup with your healthcare providers. Once you begin using a data model, you can’t simply set it and forget it. Ask the following questions to periodically evaluate your models:
- Are there changes in the outcome of the models? You need to verify that your attributes are still predicting the same outcomes as intended, as well as capturing the same data. For example, say you have an attribute in your model that counts the number of credit lines open for a small business. If the attribute changes and those types of credit lines are no longer reported by the data provider, that number can go from three or four to zero, without there being a change in the number of credit lines open by the business. Because the data that goes into your model has changed, your model is not accurate unless you update the attribute.
- Is your model stable? You need to make sure that degradation hasn’t reached a point where the predictive value is no longer accurate. For example, scores before the 2008 recession have a different meaning than afterward, due to the changes in the financial system.
The future of your business depends on making accurate credit decisions. Instead of using outdated models, use the latest technology and methods available by using machine learning data-driven models. It’s simple. It’s quick. And most importantly, data-driven models are accurate.