In our last few articles, we discussed various aspects of the idea of SMART Banking. We have shared our thoughts on customer segmentation, wallet share maximization and customer lifetime value (CLV) estimation. We have introduced the concept of CLV mileposts or ‘triggers’ as influential events that indicate a change in a customer’s financial situation. We have also shared some thoughts on how to manage CLV mileposts dynamically. In this paper we will expand the CLV discussion; CLV to the prediction of ‘time-to-event’. Time-to-event was a concept first introduced in our article (link to Smart Banking: Customer lifetime value)
Time-to-event is a measure of the duration of time required on average to progress from some initial state to a future state. The future state being dependent on what events occur along the way. Time-to-event is a very powerful tool for maximizing wallet share and increasing customer lifetime value. If we know WHEN to expect a change, we can prepare for it and profit from it. What happens to us each day determines how we react to our world tomorrow. We are the sum of our experiences.
Restoring Customer Service
Traditionally, “walk-in” customers provided the opportunity for the bank to maintain and build upon an existing customer relationship. Cross-sell and up-sell offers were made across the table. However, the popularity and convenience of online banking has several important implications for a traditional brick-and-mortar bank or bank branch. Today, customers have less motivation to visit a branch. Customers are often encouraged and offered incentives to do their banking online, never to have the need to walk into the bank. Directing customers to online services gives banks the opportunity to make customized offers to customers. Customers now ‘shop’ for the best offer for their new financial needs and often make decisions without giving their established bank the opportunity to win their business. Often, customer transactions (the data) is increasingly becoming the single most important touchpoint the bank has with its customers.
Predictive analytics is fast becoming an important catalyst for traditional customer awareness, relationship management and building. The need to analyze data to determine the timing of customer inquiries is fast becoming more than a need; it is a requirement. Time-to-event reduces marketing costs. A customized offer can be created at the point of need thereby significantly increasing the chances of an offer being accepted.
Building Solutions Online
Predictive analytics provides the tools that help banks stay connected to customers proactively. Data provides an unbiased and quantifiable resource with key insights that help banks make informed decisions. Transactional data is used to increase wallet share, and systematically and methodically maximize customer lifetime value. Realistic goals can be set and achieved based on quantified insights garnered from transactional data. A roadmap for achieving goals can be created, taking the guesswork out of presenting an offer to a customer. With predictive analytics, the roadmap is completely laid out in advance, in an easy to follow solution letting banks focus on strategy and fine-tuning the direction ahead of time.
1, 2, 3… GO!!!
Predicting time-to-event is characterized by three (3) distinct steps.
The first step is the identification of the influential events that are to be predicted. Influential events are those that offer the most variability between people. If an event is characterized the same way for everyone is it really of interest? Most likely the answer is no.
If an event is the same for everyone it gives us no ‘information’ about the differences between people.
We need to be careful not to get trapped into thinking that common events are of little value. Common events provide an overall direction and point the way; like stepping stones through the forest. The events we pick to measure are those ‘stones’ in the path that offer the most information at the individual customer level.
Predictive analytics is therefore described in this example as the science of determining significant differences amongst commonalities. We need to have a solid grasp of not only the techniques to be used, but also the nature of the information that the data provides.
The second step to predicting time-to-event is identifying the proper ‘tool’ for the job.
Predicting time-to-event involves the use of a toolbox of procedures within an area of statistics called survival analysis. Survival analysis is a very common practice for analysis of clinic trials and medical studies when determining the efficacy of a new drug or surgical procedure. The name lends an optimistic tone. Many tools in survival analysis are available.
Three (3) general areas of survival analysis, depending on the inherent level of information the data possesses, include:
- Parametric Survival Analysis
- Non-parametric Survival Analysis
- Proportional Hazards Models
Variations of the three types of models include many possible scenarios of censorship and truncation. As an example, the data may suggest the model to be a right-censored, non-parametric survival model or a left-truncated proportional hazards model, etc. The list of possibilities of modelling techniques can become quickly overwhelming. The importance of having a large statistical toolbox and a broad knowledge of procedures becomes increasingly important in order to find answers and make reliable projections. Additional specialized skills in modelling probability distributions to the data is also required for making credible predictions.
The third step in predicting time-to-event is modeling the effects of customer characteristics and intermediate events.
Intermediate events can influence time-to-event in a consistently or they may change dynamically over time. For example, if parenthood is an influential factor, we may conclude that having 2 children reduces time-to-event by three (3) days. The interaction between affecting factors may also be significant. Therefore, if parenthood and age are both important factors, we may conclude that the median time-to-event is reduced by three (3) days for a Gen-X customer but changes if the customer is a Gen-Y individual. We could even have factors that change in magnitude over time. All possible factors and their interactions need to be considered during the analysis. Depending on the findings, a multitude of modelling techniques may be employed.
In future papers we will begin demonstrating the power of predictive analytics in delivering solutions as they relate specifically to SMART Banking and maximizing wallet share. Stay tuned to see how predictive analytics provides tangible solutions that could benefit your bank.