BY DAVID ROSS
By David Ross, VP, Advisors Plus Predictive Analytics at PSCU
I will never forget the time I received an escalation call from a member whose card had been declined for fraud prevention while checking into a hotel inParis. She was furious about what transpired, and her main argument was that we should have known it was really her using the card. Why? Because two months prior, she had purchased airline tickets on that same card. And the airline in question? It was not United, American or Delta – it was Air France. She exclaimed,“Is it really that unusual for someone to buy an Air France ticket and then two months later try to check into a hotel in Paris? You knew that it was me!” It was difficult to disagree with her, as it should have been easy to predict the airline and hotel transactions were related. That was her expectation as a member 10 years ago in 2008 when that conversation took place.
My team and I have seen member expectations continue to grow in the decade since.After years spent partnering with credit unions across the country, my team has found the best approach to knowing and serving members better is through the use of Predictive Analytics.
Predictive Analytics, as the name suggests, can help credit unions determine the probability that something is going to occur in the future. This is very different from traditional reporting, which details what has already transpired in the past, and is often referred to as Descriptive Analytics. To be clear, a credit union needs solid Descriptive Analytics as a foundation before progressing on to the predictive space. It is crucial to understand items such as how a portfolio has performed year over year, where members are transacting, how often a card is used, what interchange was received, how much interest income was generated, and what the total outstanding balance is. A credit union should regularly review reports and dashboards to evaluate KPIs, trending and other key metrics around the aforementioned items. If not available internally, a credit union’s card processor might have this information available. Digesting it is important not just to measure how well the portfolio has been managed, but also to identify potential areas of opportunity. For instance, a graph showing a steeper trend line for fraud losses versus peer groups could mean that a more conservative risk approach should be considered. But while valuable, such an analysis is not predictive in nature – it remains based on Descriptive Analytics.
Predictive Analytics employs a mathematical approach to forecast outcomes
Predictive Analytics by comparison is built on statistical data modeling, and probably the best-known predictive model in the credit union world is the credit score.Just as the credit score is used to assess the probability that a loan will be repaid, other predictive models can be built for a variety of use cases. For instance, a predictive model can be built to identify members who are likely to attrite. Similarly, a predictive model could be built to determine which members might be on the verge of becoming detractors, allowing the credit union to intercede before there is any impact to net promoter score (NPS). These are two powerful examples that leverage data to know the member better, and it is imperative to know the member better because that is the expectation today.
Cost and complexity are obstacles that need to be overcome
Moving from Descriptive to Predictive Analytics can present a challenge for many credit unions. To begin, detailed granular-level information is required. Even when a credit union has captured this type of data in a data warehouse, it still needs to deploy a statistical modeling platform with enough capacity and computational power to work with that data. It is also possible that changes to the data warehouse design will be required to support Predictive Analytics. Once these items have been addressed, there is the need to have a human resource familiar with the techniques required for predictive analysis. Quite often this requires hiring a data scientist, a position that currently commands a six-figure salary in the marketplace. Plus, in 2018, multiple publications have estimated a shortage of over 100,000 data scientists in the U.S., making the talent that much more difficult to find. Significant resources and costsare clearly involved, making the path to Predictive Analytics somewhat difficult to traverse.
Partnering with a subject-matter expert can be a more efficient approach
As an alternative, a credit union might seek out a vendor or CUSO (Credit Union Service Organization) to assist with the transition to Predictive Analytics. This is a very practical approach, but credit unions should carefully evaluate with whom they partner, as there are several pitfalls that might be encountered. One important question to ask is how will the potential partner get the data needed to build the predictive models? If data needs to be sent outside the credit union, how frequently does it need to be transmitted, and is PCI/PII compliance in place?Another question to ask is who is responsible for the analytics, as some vendors provide a data platform only, leaving the analysis to credit union staff. Be careful to understand the actual capabilities the partner is offering – are they providing reports (Descriptive) or are they building models (Predictive)? Finally, ask what happens with the final output. Going back to the attrition model example, would the credit union simply receive a list of members likely to close their accounts? Would it be a scored list? What is the next step that should be taken? Taking this proactive approach will help ensure the selected partner is best aligned to meet the credit union’s needs.
Once a credit union fully embraces Predictive Analytics, it will have insight into the membership that facilitates providing a level of service that exceeds even today’s high level of expectations. Members want to feel as if their credit union truly knows them at the individual level. With Predictive Analytics, credit unions can leverage their data to accomplish just that. Credit unions can know why the member has not activated a card, why a previously active card went dormant, or even when and where the member is likely to be traveling – something that personally would have come in handy back in 2008.
David Ross has spent the past 19 years working in a wide variety of data science and analytic roles. Most recently, he was responsible for global fraud analytics strategy and execution at Citi, a position that required working with countries around the world to implement best-in-class analytics tools. David is currently leading the initiative to develop models for Predictive Analytics at PSCU.