Blog

Practical Machine Learning: The Use Case in Retail Banking

May 4, 2018

Despite all of the buzz about machine learning and AI, it’s still relatively rare to see concrete examples of practical machine learning in action, much less driving real business outcomes.

This is not to say that progress isn’t being made. One of the most intriguing aspects of machine learning is that the algorithms are in a state of constant adjustment. In a retail setting, this means that a machine learning system becomes better at predicting customer behavior with every data point it collects; it lives in a perpetual state of trendspotting. When used to its potential, machine learning gives every single customer interaction the power to provide insight across an entire customer base.

Since the launch of Ada, our machine learning suite for real-world marketing applications, we’ve received a ton of interest from customers in putting it to use. Today, we’re sharing a new example of Ada in action.

In the notoriously fickle world of retail banking, one customer is working with us to develop five concrete models for machine learning that drive concrete business outcomes and deliver ROI. While these models rely on advanced technology, the applications are intuitive and the value is clear:

Contact Attrition Risk Model – Designed to predict the likelihood and timeframe in which a contact will close all bank accounts and cease doing business of any kind with the bank. This gives our customer the ability to adjust messages to these customers and make overall business changes accordingly.

Account Attrition Risk Model – Predicts if and when a client is on the verge of closing a given bank account, but would still maintain other banking accounts and services and continue doing business with the bank.

Next Best Product Model – This model can recommend the products or services (checking, savings, mortgage, etc.) a contact is most likely to use, which is incredibly valuable when it comes to personalizing marketing offers. The model can be programmed to get as granular as necessary, even getting down to recommending the specific type of debit card a customer should be marketed.

Loan Default Model – This model is able to identify which contacts are most likely to default on a loan or credit card payment, which has applications that go far beyond the marketing channel.

Clustering Model – This model looks at all available customer data to find hidden structures and relationships and develops contact groupings for deeper analysis. Those contact groupings are presented visually and described statistically through an unsupervised model report. With this model, a retail bank could finely tune its customer acquisition strategy around specific clusters of ‘look-alike’ customers and improve their chances in the race to win more customers. This model is also extremely valuable for customer retention.

The ultimate goal of machine learning is to produce revenue growth more efficiently and forecast more accurately. In the world of retail banking, it also has the power to enhance the overall customer experience with “just in time” communication and a personal touch, as well as provide more intelligence to the business overall.

One of the most exciting things about launching our machine learning suite is seeing how customers put it to use. The creative ways that predictive algorithms can be applied to business challenges are virtually endless, and we’ll continue to share the coolest examples on this blog.

If you’re interested in learning more about what the QuickPivot machine learning modules can do, we’d love to hear from you.