Credit unions at various stages of analytics maturity rely on predictive analytics models to serve members and solve business problems.
Whether built by their own data teams, custom designed by partners or purchased off the shelf from analytics vendors, the models delivered measurable results across several endeavors. The insights they generated helped credit unions stem the tide of attrition, cross-sell products and manage a wide range of risks.
Everything was working beautifully and then, COVID-19.
COVID-19 and its economic domino effect are likely to skew any “insights” a pre-COVID-19 predictive model delivers today. Right now, an attrition model may identify a set of members as at risk of leaving the credit union, when in actuality, those members are simply bracing for the financial impact of the crisis. Likewise, a collateral evaluation model may determine the value of a member’s assets is adequate to secure a loan when, in fact, falling housing prices could significantly change the calculation overnight.
This is not to say credit unions should stop using their predictive models. Rather, data teams should investigate using them in different ways. Here are three key ways credit unions can continue to rely on their models to make decisions throughout various stages of the COVID-19 crisis.
Retool models for the current environment
Pre-pandemic models can still provide a great deal of value during the current crisis. Applying them to an expanded set of use cases keeps the credit union’s analytics muscles in shape.
The trick is understanding how a particular model works. Decision makers need to have a clear line of sight into the data a model factors into its analysis, and most importantly, how each data field is weighted. This knowledge can help teams better respond to the insights the model delivers. Understanding a model’s actual formula may also lead to conversations with a data scientist around retooling the model for the current environment.
A pre-pandemic attrition model, for instance, may score the number of days a member is delinquent higher than the number of recent debit card transactions. Depending on a particular member segment’s COVID-19-related hardships, that calculation may need to be adjusted to paint the real picture of the segment’s current needs.
Data teams can also leave the model as is, without adjusting the weighting, and simply frame the results in a different way. A pre-pandemic attrition model will still identify at-risk members, but the “risk” needs to be redefined. The segments the model places at highest risk of attrition may, in fact, be at highest risk of default.
Adapt data activation strategies to emerging needs
Credit unions value analytics models because they set plans in motion. In today’s environment, the action steps a model advises will need to be adjusted to ensure the credit union is offering the most vulnerable members the most meaningful solutions.
If, for example, a credit union team was using a model to drive its outbound sales effort, they may now be able to use it to identify the precise financial challenges a member or set of members is facing. If the model recognizes a sudden stop in transactions, but also observes that deposits are steady, that segment may be hunkering down in preparation for a job loss. The team may find it appropriate to reach out to these members to see how it can help, or to offer one of several tiered relief programs based on a member’s individual circumstances.
Use model outputs to forecast different scenarios
Most credit unions are used to using asset liability management (ALM) models to set strategy and comply with regulatory guidelines. Many will be relying on these models heavily during the current crisis to strike a balance between member assistance and cooperative health. The outputs of predictive analytics models can be used in conjunction with forecasting models to enhance the accuracy of their predictions.
Say, for instance, an attrition model generates a list of members based on a variety of recent behavior or transaction changes. Credit union leaders can then analyze the product mix of that segment. That analysis will show leaders what potential losses may look like should the members default on loans, leave the credit union altogether or take other revenue-altering actions.
Now, in addition to a list of members who may be experiencing hardships, the credit union also has a view on what that hardship may mean in aggregate to the health of the credit union itself. This can help set priorities in terms of where relief programs are focused and to whom they are targeted — getting the most meaningful assistance to those members who have the highest potential to cause the greatest loss to the credit union.
In the credit union movement especially, predictive models are about delivering what’s best for each member as an individual while also sustaining the soundness of the credit union. That purpose remains in place today. In fact, if credit union leaders are bold enough to put their predictive analytics models to a new test, those models could become more powerful than ever.
