Churn Modeling Overview Mather Economics LLC May 24, 2016
© MATHER ECONOMICS2 Introduction to Mather Economics
Founded in 2002 by Mather Lindsay, PhD to bring leading practices in applied microeconomics to our business clients Clients in the United States, Canada, Europe, Asia Pacific, Australia with offices in Atlanta and Amsterdam 35 Employees and 10 academic affiliates Strong academic links to leading universities in the field of micro economics Introduction
Mather Economics Client Base 4 Work with 17 of the largest media companies; across four continents Many of the largest subscription brands (500+ newspapers) Data analyzed on 30M print/ digital subscribers weekly Manage over $4B in subscription revenue annually
Assist our clients by managing their revenue relationships at the household or engaged-user level through… Data Collection Analytics and Practical Application A-B Testing, Measure and Adjust Custom Reporting The Mather Approach
Acquire: Maximize start value/volume Retain Minimize churn Renew: Leverage willingness-to-pay Upgrade: Grow customer engagement Reacquire: Target offers based on history 6 Acquire Retain Renew Upgrade Reacquire Holistic Approach to the Audience Model
Why Churn Modeling? 7 Mather historically focused on renewal pricing Data-driven method of optimizing renewal rates at the subscriber level Has effect of reducing pricing-related churn vs. traditional flat increases, however: There still exists a churn problem Stops analyses showed majority of churn not related to price Validated through A/B testing How can Mather help our clients reduce non-pricing related churn? Develop a model using existing subscriber data to estimate churn propensities at the subscriber level Markets can then segment market based on churn risk and take preemptive action to save at-risk subs
© MATHER ECONOMICS8 Churn Modeling and Payment Path Analysis
Churn Model Objective & Specification 9 Objective The objective of churn modeling is to apply churn probabilities to individual print subscribers by identifying characteristics and attributes associated with a specific stoppage event. Specification To estimate the churn probabilities, we use a logit model, which is a discrete choice model used to predict a binary outcome A logit model gives us the ability to estimate the probability of a certain event occurring given a set of explanatory variables Data Requirements 2+ years of transactional data (starts, stops, reverts, upgrades, etc.) 2+ yrs of complaint data (complaint date, type, description) 2+ years of payment data (payment date, thru date, amount)
Churn Model Inputs 10 Variables Included Subscriber Characteristics (FOD, rate, term length, Ezpay, channel) Complaint History (# complaints, complaint type, etc.) Payment Segment (early payer, late payer, very late payer, etc.) Digital Engagement (time on sight, PVs, page breadth, etc.) Deviation Metrics ( tracks changes in behavior over time) Examples include: Making a payment one or two standard deviations outside of a sub’s normal payment window Changes in engagement on the website Deviations in behavior are significant predictors of churn and contribute greatly to the ability of markets to isolate timing of churn
Churn Output 11
12 While a negative relationship between tenure and churn seems obvious, the non-linear fashion of that relationship is revealing and highlights the importance of getting subs to retain during their first year, where churn risk declines so rapidly Churn Insights: Tenure
13 Churn risk and income have a negative relationship, meaning higher levels of income are associated with lower risk of churn Churn Insights: Income Higher Churn RiskLower Churn Risk
14 Interestingly, more complaints are associated with lower churn risk, but the type of complaint matters Churn Insights: Complaint Type Higher Churn RiskLower Churn Risk
15 Subs that deviate from their normal patters by at least one standard deviation are at a significantly higher risk of churn Churn Insights: Income Higher Churn RiskLower Churn Risk
16 An effective scoring process allows you to customize your approach by risk/ opportunity Example below shows two subscribers, one at high churn risk and one at low churn risk Churn Modeling Output: Profiles
17 Case Study: Applying Incentives to Churn Risk Accounts
Churn application pilot showed that significant churn reduction could be achieved with small-dollar incentives In the case of a $1 greeting card among high churn probability subscribers, churn was reduced by 2.7 percentage points over a control group 90 days post application 18 Case Study: Applying Incentives to Churn Risk Accounts
Tracking churn targets over time shows the “staying power” of the incentive is quite strong Reduction in churn vs control remains relatively stable at the 60, 90, and 120 day marks post application 19 Case Study: Applying Incentives to Churn Risk Accounts
20 Analyzing payment patterns of individual subscribers allows for custom communication, resulting in more effective stop save touchpoints Applying dynamic messaging, customized at the subscriber level, resulted in a 14% decline in non-pay stops versus a business-as-usual control group in one of our markets leveraging churn analytics Case Study: Payment Data-Driven Dynamic Messaging
21 In this case study, payment analysis showed that the last five retention touchpoints netted only 7% of total payments Recommend replacing some of these late touchpoints with less costly alternatives and move date of 2 nd invoice up to realize revenue earlier Case Study: Adjusting Touchpoints to Impact Expenses
Conclusions 22 Key Takeaways A well-defined churn model can successfully identify subscribers with higher propensities to churn First rule of economics holds true: INCENTIVES MATTER! Providing incentives to subscribers to make payment results in lower churn levels (across all risk levels) The most expensive incentives aren’t necessarily the most effective In one case study, a $1 greeting card resulted in comparable retention improvements versus costlier options Targeted, personalized messaging (almost zero marginal cost) based on payment analysis insights results in higher pay-through rates
© MATHER ECONOMICS23 Questions
Data is Oil… Publishers need gasoline “…Data is the new oil — it’s very valuable to the companies that have it, but only after it has been mined and processed. The analogy makes some sense, but it ignores the fact that people and companies don’t have the means to collect the data they need or the ability to process it once they have it. A lot of us just need gasoline.” Derrick Harris, Gigaom.com, March 4,