Using Text Mining and Machine Learning to Enhance the Credit Risk Assessment Process
Advances in technology have instigated a substantial shift in consumer expectations. Today’s financial services customers demand access to a range of services, real-time updates and a seamless customer experience. At Open FinTech Forum, I will provide some insight into Spotcap’s approach to credit risk assessment using text mining and machine learning.
A recent survey by Oracle found, that although customers are generally satisfied with basic banking services, their satisfaction drops when attempting more complex transactions such as securing a loan. We have observed the same sentiment across the business community. This is why, at Spotcap, we’ve turned tradition on its head and created a more efficient take on business loans.
We undertake cash flow based, rather than credit-score based underwriting, and use technology to speed up the process. Combining tried and tested credit assessment principles with innovative technology such as our automated data scraping services, machine learning credit models, and skilled human analysts enables us to offer a more efficient take on business loans.
Machine learning credit algorithms
Our risk assessment utilizes numerous sources but relies heavily on three main sources - borrower profile, bank account, and business profile - and is supported by a set of machine learning credit algorithms. This approach allows us to accurately and fairly assess how a business is performing today, and make a prediction about its future performance.
Whilst we feed our models with hundreds of data points sourced from credit bureaus, tax agencies, business records and the applicants themselves, it is bank account transactional data that often paints the most accurate picture.
Spotcap’s Bank Account Model incorporates more than 200 numerical variables. Business bank account data, when structured correctly, is one of the strongest sources of predictive information for short-term lending and risk mitigation. We construct the raw data found in a bank account into a form of variables enabling us to derive meaningful insights.
We have also developed bank account text mining tools to identify key negative factors such as payment reversals, late fees and collections transactions.However, this requires a supervised approach to minimize the risk of false positives.
The more data you feed into your machine learning models, the more accurate will be your results. But it’s not only about quantity, it’s primarily the quality of data that matters. Well specified machine learning models can help lenders make faster and more informed decisions. However, even the most powerful machine learning algorithm will fail if applied to data with measurement error. The better your understanding of your data, the more accurate and insightful your results. Our underwriters and data scientists continuously add new knowledge and risk drivers to our models to get even more precise outcomes.
It’s all about automating the right parts of your analysis and remembering that human interaction is important at every stage of the model life cycle because we’re dealing with real people and real businesses, which are by nature complex. Human expertise combined with advanced technology enables us to make accurate, yet flexible credit decisions within one day.
Sign up to receive updates on Open FinTech Forum: