Realize business value from big data with Infosys data analytics solutions.

« October 2013 | Main | March 2014 »

December 3, 2013

Criticality of Predicting Customer Churn

My personal experience of various chrun prediction solutions, and the way the problem is being looked at across various industries is what i am making an attempt to narrate in this blog post. One may actually read it as scratch notes gathered in the process of understanding Churn Prediction process and lifecycle. This is by no means a comprehensive and exhaustive list, however can be good check points if you are embarking on this road or not able to reap benefits with your investments in predicting churn.

Industry trends show that annually there's over 20-40% churn especially in Telecommunication industry. The reason, why enterprises are serious about handling churn, and take appropriate measures to reduce / prevent churn, is that's it's much costlier affair to acquire new customers as opposed to making efforts to retain existing customers. Certainly there's a brand, referral, social and image management aspect also involved which can dent enterprise reputation and profitability if not handled effectively.

Most important reasons for doing churn analysis:
1. Reduce marketing costs - maximize profits
2. Reduce churn thru predictive models
3. Segment market into alike clusters - identify the customers that will generate most profits
4. Understand customers & their behaviors
5. Adversely impacts the profitability of organization
6. Cost of acquiring new customer is much higher than retaining existing customer
7. Reduce the loss of referrals via the existing customers, if they churn out
8. For making highly targeted and cost effective marketing strategies

Key reasons for customers to Churn:
1. There are competitors / other companies offering similar products and services
2. Have a better pricing model and options
3. Bankruptcy by companies
4. Social media influence and sentiment - word of mouth in social circles
5. Better customer connects and touch points to address the concerns


Figure 1 - The Churn Prediction Landscape

Data Sources for conducting advanced analytical techniques:
1. Billing systems
2. Customer Care - customer profiling
3. Call detail records (CDR's) - usage pattern, behavior
4. Customer social media influence and network - sentiments, blogs / posts
5. Product/Service portfolio
6. Network Service, Costs

Opportunities for increasing profitability:
1. Personalized offers based on customer behavior patterns, social network influence, customer usage patterns
2. Location-based data for location based advertising
3. Optimal Campaigns for up-sell, cross-sell, acquiring new customers, influential or viral marketing
4. Combining real-time network feeds, back office data/logs, network inventory, capacity planning, and monitoring service quality, along with subscriber information can provide opportunities to increase wallet share per customer and satisfaction




Figure 2 - The Opportunities Churn Prediction provides

Varying and at times conflicting Goals & Objectives :
Increase : Profitability, Wallet Share, Cross-sell and up-sell, Satisfaction
Reduce : Churn, Cost (Marketing, Campaign)
Bottlenecks : Processing increased volumes in real-time, Integrating various information sources with real-time customer context, Gathering all customer touch points

Traditional vs Modern Predictive Techniques:
1. Traditional - Fixed set of rules to identify a churn. Likelihood that by time it's detected, customer has decided to Churn and effectiveness of campaigns / offers or customer care interaction is low
2. Predictive - Predicting before potential has decided to churn. Can provide better alternates to influence the potential in not making a churn decision, and allows for opportunities to act
3. Live Prediction - Even prediction can only tell you whether a potential will churn or not, still at large and critical piece is to predict when the potential will churn or how long is the potential going to stay. Altogether a different set of possibilities, opportunities open up in this case as you can have targeted offerings, campaigns, product launches

Points to ponder:
1. Is highly targeted and cost effective marketing campaigns a good enough goal of Churn Analysis - What about campaigns in customer context, at right time and in right format (the way customers want or like it)
2. Is General linear model or Logistic regression model good to predict churn - maybe not as distributions may not be normal in data set, spikes in datasets on various events will not fit the linear or regression models well
3. What if you are not restricted to the data distribution but consider data itself to model, and find patterns within the data using data mining techniques - now you are talking some sense
4. How do you measure model effectiveness - Lift (ratio between results obtained with and without the predictive model)
5. Questions that will define your roadmap, and plan to tackle churn includes
 a) What behaviors are likely to cause churn? [Service provider behaviors, Customer Behavior & Competitor Behavior]
 b) What pro-active measures and steps can be taken to prevent churn?
 c) Most effective would be to identify, record and leverage the measures implemented to tackle Churn (by causes) & the effectiveness of those measures

In summary, while it's critical to evaluate and analyze the churn on regular basis, it's equally important to holistically consider all factors that contribute towards churn and get a better understanding of customer behavior to churn. The end result of this exercise results in being able to better plan and utilize resources to prevent churn, & continue to increase profitability with a happy customer base.

Subsequent article am planning to address some of the statistical modeling churn techniques, and which ones are better in which situations. Thought, experiences and ideas are welcome!!

Subscribe to this blog's feed

Infosys on Twitter