Relying on survival analysis for customer retention

Predictive analytics is playing a significant role in recognising at-risk and valuable customers to develop customer retention strategies.

By M. Hriday

Customer retention analytics has become a widespread application for business intelligence tools and solutions.

Companies have been looking at gaining an understanding of the variables that influence customer lapse. Specialists highlight that retention starts with the first customer contact and continues throughout the entire lifetime of a relationship.

One of the most useful applications of business intelligence is developing effective customer retention solutions and reducing costs by identifying the greatest number of customers likely to churn within a small percentage of a customer base.

Approach

Quite often the discussion and the point of contention happens to be the churn rate. Churn is a critical issue as it can be quite difficult to replace old customers with new ones especially if the industry is witnessing saturation. There are a couple of ways to modelling churn. This includes approaching churn as a binary outcome, and predicting which customers will stay and which ones would leave. The other approach is to evaluate the remaining lifetime.

Should one get away from modelling churn solely as a binary outcome “will they quit or not” question?

Michael Berry, founder & principal, Data-Miners certainly believes so.

Berry says this is the traditional approach, but it is a bit simplistic. 

According to him, if we ask “will the customer still be here tomorrow?”, the answer is nearly certainly yes. If we ask “will the customer still be here in 10 years?”, the answer is almost certainly no.

“So we set some arbitrary length of time and say “will the customer still be here at that arbitrary time?”.  The reason the problem is posed this way is that binary outcome problems are easy to model with logistic regression or decision trees.  A more interesting question is “How long will the customer last?”. We can approach this question using survival analysis,” said Berry, who is scheduled to speak at the forthcoming Business Analytics Summit to be held in San Jose (November 12-13).

Predictive analysis

Predictive analytics is about extracting information from data and using it to anticipate future trends and behaviour patterns. This is based on a variety of techniques from statistics and data mining. According to a specialist like Rosella, customer tendencies are captured using predictive models from past historical data. They are measured as customer metrics.

Predictive customer metrics are then used for determining business strategies and actions to take. For each business task, a set of business strategies are developed. Predictive rule-based reasoning determines customer-centric strategies. Based on the selections of strategies, actions are taken automatically or manually. The core lies in capturing relationships between explanatory variables and the predicted variables from the past occurrences and exploiting this information to predict future outcomes.

Functioning

Such analytics offers actionable projections and this form of business modelling predicts the individual behaviour of existing or prospective customers under certain conditions. The central element of predictive analytics is the predictor, a variable that can be measured for an individual or other entity to foresee future behaviour. Multiple predictors are combined into a predictive model which, when subjected to analysis, can be used to forecast future probabilities with an acceptable level of reliability.

This may seem like a reliable solution but one can still face several hurdles.

According to Duncan Houldsworth, global head of marketing analytics at Hall & Partners, the biggest challenges are usually always around data and data integration.

Houldsworth says even in today’s technologically advanced business environment, it is still a tussle to firstly access all the data sources needed to gather a holistic view of the customer and  then to combine the data in a meaningful way for either data mining or predictive analytics.

“Advances in database technology have helped significantly, but this is definitely a work in progress and requires continued attention,” added Houldsworth.

He said one should ensure that database architects (and database administrators) are integral members of the task force alongside predictive modelers and data miners.

“Just like building a house, all the specialists need to be part of the team to overcome the challenges,” recommended Houldsworth.

Maturity level

Houldsworth says modelling and predicting customer satisfaction is still a challenge, but it is progressing significantly well in some sectors and markets.

Predictive analytics is being used in retail, telecom, insurance, financial services and the pharmaceutical industry.

“I believe that the opportunity is to start viewing customer satisfaction measurement not as a means to an end in itself or a standalone piece of analysis, but a fully integrated part of the assessment to determine how well your marketing mix is functioning in attracting, retaining and maximising customer interactions with your business,” he said.

In this context, he says, the maturity level is quite low with customer satisfaction being kept separate from business intelligence activity around marketing and sales effectiveness.
 

 

Business Analytics Summit

Business Analytics News is scheduled to conduct the two-day Business Analytics Summit at San Jose in November (12-13) this year. The conference will feature  leading Business Analytics executives including ones from Monster Worldwide, JetBlue Airways, New York Times Company, Boire Filler Group and Data-Miners Inc.

For more information, click here: http://www.businessanalyticsnews.com/usa/agenda.shtml
 

Or contact: Ben Satchwell by email ben@businessanalyticsnews.com