understanding customer churn

Understanding Customer Churn – Defection, not Defectors

Dennis W. Gleiber, Ph.D.

Chief Research Scientist, The Olinger Group

Learning about defectors is not the same as learning about defection.  Both can be done, but only the latter speaks directly to solutions for the business problems associated with customer satisfaction, loyalty, and continued usage.

The very high costs associated with customer/client identification and acquisition are widely recognized today.  Most companies experience what is labeled “churn”, losing customers only to replace them with new customers.  Companies can often experience churn in excess of 50-90%, yet do little or nothing about it.  Such companies maintain volume and revenues by paying the costs necessary to constantly acquire new customers.  Some do not even know whether these new customers are in fact returning customers.  This will inevitably be a more costly proposition than retaining the same number of existing customers instead of losing them.  Retaining customers maintains volume and revenues at lower cost, increases customer lifetime value, and should create loyalty from limited experience and habit.  Not spending to recruit new customers saves money.  Continuing to recruit while retaining more of the existing customers means increased rates of growth, revenues, and profits.

Defectors

One widespread response to churn involves research that studies the lost customers, or defectors, after they are lost.  This is “too little, too late” and tantamount to “closing the barn door after the horse is out.”  Even if you know why customers defect, the best that can be done is to invest in their reacquisition.  This may cost less than initial acquisition, but certainly is more costly than retention (no lost revenues, no lost loyalty, no cost to identify or reacquire).  Every company, whether experiencing substantial churn or not, should be concerned about retention by anticipating defection.  This is done by combining knowledge of customer lifetime value, motivations to buy in the first place, the processes of building satisfaction and loyalty, and risk factors of ultimate defection.  Such a customer lifetime model is best obtained prospectively, but can also be pieced together from extant data from transactions, interactions, and survey research.

Some companies believe they can avoid the planning, effort, and upfront expense of such prospective modeling by waiting until the customer defects, and then studying the defectors.  Such a process, based on reviewing events associated with defection and surveying defectors, is better than doing nothing, but is plagued by glaring and potentially fatal design flaws.

First, getting information from defectors is likely to be more difficult than getting information from customers.  Defectors have already acted on their negative orientation(s) toward your company.  Why should they be willing to provide you with free or low-cost information about their experiences with and evaluations of you and your offerings?   Defectors may be hard to contact (contact information may change) and reluctant to talk.  They have already determined that they can survive without you and what your company offers or they have moved on and have begun to form a relationship with an alternative provider or source for what they want or need.

Second, information provided by defectors may be biased, unreliable, or invalid.  Their ill will is likely to provide a biased description of their experiences and orientation to your company, products, and customer service.  People generalize and stereotype.  If unaffected enough to defect, many defectors are likely to project their current evaluation or orientation onto many or all other and earlier aspects of their relationship with your company.  People do a much better job reporting current events and orientations than those more distant in time, making delayed responses from defectors less reliable.  We already know their current orientations are negative, thus introducing bias.  It is also easier for people to simplify an otherwise more complex experience.  Thus, more emphasis is likely to be placed on a proximate cause, making the defection seem like a single or simpler event when it is likely a longer, more complex one.

Loyalty is a kind of diffuse support, a residual psychological attachment that can offset short term dissatisfaction.  Short term dissatisfactions can erode loyalty until any negative event or experience is sufficient to induce defection.  In such cases any proximate cause is more like “the straw that broke the camel’s back” than the true explanation for the gross change in relationship, the defection.  Thus, the defector’s responses may be well intentioned and truthful as far as they go, but nonetheless an invalid representation of the true process that generates the defection.  The more time that elapses between defection and surveying defectors, the less reliable any, and perhaps all, responses become.  Half of the twelve contaminating factors Campbell and Stanley (1963) identify in their seminal work on research design clearly contribute to this unreliability.

Third, knowing only about defectors neither explains defection, nor provides information about retrieval or future retention, because it tells only half (perhaps even less than half) of the story.  The whole story of defection requires information about those who do not defect, the continuing customers.  This is a simple matter of comparative design (see for example Przeworski and Teune 1970).  Describing only defectors is no substitute for knowing whether, and if so, how and why defectors differ from continuing customers.  Defectors all experience A, B, and C, so we will change to -A, -B, and –C to bring them back or stop others from leaving.  However, continuing customers all experienced A and B too, but did not defect.  Can we expect the changes we made to stem defections or increase them?  Do the changes drive continuing customers away, bring back lost customers, or reduce defections among continuing customers?  Because we look at only defectors we do not know how defectors compare to continuing customers and cannot answer any of the previous questions.  Understanding defection must always derive from a comparative model that includes defectors and continuing customers, not just defectors.   Many models are available, but all depend on direct comparison for valid insights.

Defection, Not Defectors

The general goals of any defection study are to describe attributes and interactions of customers and defectors and use them to explain and predict the decision to end a customer relationship.  The business benefits of this knowledge include:

  1. Identification of preferred customers, those who have longer expected durations with, and therefore greater lifetime value for, the company.
  2. Modeling risk of defection to identify potential defectors in advance and take customer specific remedial actions that will increase retention.
  3. Modeling potential to, strategies for, and relative value of recovering lost customers and keeping them once recovered.

Event History and Duration

One such method is the event history approach (see for example Allison 1984).  It involves the combination of objective and survey data that simultaneously compares continuing and cancelling customers.  It includes models for estimating risk, predicting duration, and predicting defection.  It is most valuable when psychographic data are incorporated into a model based on objective data regularly collected and maintained in the course of business to serve as the basis for continuous identification of potential defectors before it happens.

Service Industry Case Study

For example, The Olinger Group recently analyzed data for a Fortune 500 national service provider to help them understand their extremely high churn rate.  One analysis of current active customers using a right censored duration model indicated substantial differences in expected duration antecedent to and independent of experience with the company.  Where customers are located and the nature of their initial contact with the service company, objective information routinely collected but until then unanalyzed, provides significant predictive power about customer lifecycle going untapped.  An expansion of this model to include other characteristics of the customers and the nature, timing, content, and disposition of their contacts and interactions with the company increased the explanatory power and precision of the model.  Inclusion of cancelled customers further increased the power of the duration model by removing the right censored restriction and allowing for additional direct modeling of the cancellation decision.

We modeled a right censored duration of time since sale with three sets of dummy variables – state location, recruitment channel, and use of add-on services – using data from 5,531 current customers to estimate customer life expectancy and probability of defection.  (The same model could be estimated for all customers at any point in time in the past so as to include more defectors.)    Findings were substantively meaningful and at times both consistent with and contradictory to conventional wisdom.

The type of initial contact with customers matters, accounting for three additional years of average customer lifetime (more than two times longer than others).  When an “inquiry” customer first buys, she is likely to remain a customer 3.98 times longer than a “neighborhood” customer, whereas a “telemarketing” recruit is likely to remain a customer 1.22 times longer than a “neighborhood” customer.  This indicates a substantial advantage to signing and keeping customers who initiate inquiry compared to all others.  Unfortunately the company had no such initiative in place.  An “inquiry” customer has an expected duration 2.35 times longer than a “telemarketing” customer.  This relationship remains after adjusting for the other factors tested.

The addition of the set of state dummy variables shows AL, GA, and TN to be very similar to FL in terms of expected duration of customers, while VA customers continue nine months longer than those in the other four states.  This is different from the anecdotal view the company holds that FL is uniquely different from the other areas.  The company’s “neighborhood” customer remains just a few days longer than a year, while a similar “neighborhood” customer in Florida remains about one month less.  This one year duration-average customer life is consistent with what we have heard from the company about its very high rate of churn.  Defection occurs on average at about the same time of year in southeastern states including Florida, even though it was conventionally viewed as very different from other states without the fixed seasonality experienced elsewhere.

These results are robust.  When another variable is added to represent the customers who use add-on services, the relative impacts of “method of initial contact” and “state” do not change.  In addition, using “type 2” service has an independent impact increasing expected duration approximately one year over and above the effects of mode of initial contact and state.  Controlling for “do not solicit,” the duration of the residual comparison group is reduced by about two months.  These customers who express the desire to not be solicited show an expected duration of about three months longer (the net increase of one month is substantively negligible).  This suggests that those who opt-out of solicitation are not different enough to cause loss for the company, and there is a marginal advantage to identifying them.  However, failing to respect their wish to limit contact with the company may have ramifications.

The explanatory power of the models indicates that there is substantial variation in duration (30-80%) left to be explored.  Results are consistent with what is already known about the market.  In addition, non-trivial, and perhaps non-obvious, differences in expected customer life in VA, number of services used, and how a customer is initially contacted were identified.

Gaming Industry Case Study

Casino6, in a regional market with five other casinos, was experiencing what they believed to be a substantial churn of first time players. Casino6 wanted to understand this churn process and develop mitigation strategies.  The Olinger Group conducted telephone interviews with nearly 1,000 Casino6 third and fourth quarter first-time players, both continuing players and those identified by Casino6 as defectors.  Two objectives were to describe Casino6’s players and defectors and identify drivers of defection.

Casino6’s player can be locals or from more than 100 miles away.  Though Casino6 retains more than 80% of its first-time customers during the three month period, more than half of Casino6’s continuing players say they visit another area casino only once during the three months.  Casino6’s defectors visit area casinos more frequently than its continuing players.  Despite patronizing competitors, Casino6 defections do not appear to be directly related to visits to other casinos.  In fact, visiting another casino is inversely related to Casino6 defection.  However, there is a marked difference in the percent of defectors who have visited Casino3 (16% compared to 6-10% for other area casinos).  Notice how the description of defectors implies they may be pulled away, while the model of defections implies just the opposite return players also play around at certain competitors.  The significantly greater number of visits to area casinos by Casino6 defectors indicates that defection losses involve more frequent, and therefore possibly higher, value players.

When customers defect from Casino6, they do not quit playing in the area, but they visit area casinos about 25% less often.  This suggests that there may be a segment of players that visits many casinos and after defecting from one reduces their total number of visits proportionally.  All of the area casinos host Casino6 defectors.  Thus, it is not simply one other product that is pulling customers from Casino6.  A disproportionate share of Casino6 customers visits Casino4, objectively the premium gaming product in the area.  Relatively more defectors come from the group that visits Casino6 most often, suggesting there may be a boredom factor involved in defection that should be tested directly in future research.

Players and defectors are similar in the structure of their overall evaluation of Casino6.

Both depend on a combination of casino experiences and evaluation of entertainment value.  Defectors based their overall evaluation of Casino6 on fewer factors that account for more of the evaluation.  It is interesting that Payouts are not a significant factor in the overall evaluation of Casino6 by defectors because previous studies of defectors only identified this as the most important characteristic of defectors.

One way to think about defection, PULLED AWAY, is to assume that one or more visit(s) to another casino can result in a negative comparative evaluation leading the customer to defect from Casino6 in order to frequent the other casino.  This does not appear to be true.  There is a significant relationship between visiting each of the area casinos and continuing to play at Casino6.  Players play around and the defectors who say they have played elsewhere are very similar to continuing players.  In fact, we have identified an additional segment of the gaming population, one that does not defect from Casino6 and from whom it could obtain a greater share of wallet perhaps more easily than retrieving one, or few time(s) visitors.  Unfortunately, we do not know whether Casino6 defectors also defect elsewhere.

Another way to think about defections is to assume that more visits to Casino6 indicate habit or positive evaluation that in turn immunizes Casino6 from defection.  This seems to be wrong as well.  Relatively more defectors come from the group that visits Casino6 most often.  Twenty-five percent of customers visiting Casino6 four or more times in the last three months become defectors.  This may indicate another potential segment of players who play frequently but change venue, perhaps even cycling through two or more casinos over an extended period of time.  These players further distort the description of defectors only and offer another group for targeting a greater share of wallet, not retrieval, since they will be back eventually no matter what is done.

Defectors rate Casino6 slightly lower for Slot Payouts than do Players, but not significantly different.  Yet, those who rate Slot Payouts higher at Casino6 are more likely to defect.  This seemingly non-intuitive finding provides additional evidence of the limitations of describing defectors instead of modeling defection.  Players who visited Casino3 are the most likely to defect from Casino6 (16%).  Six to ten percent of those who visit the other four competitors defect from Casino6.  Casino6 players visit the other area casinos on average nearly 4 times every three months, especially Casino1, Casino2, and Casino4.  If a customer defects from Casino6, they do not quit playing in the area, but they visit area casinos about 25% less often.  This suggests that there may be a segment of players that visits many casinos and after defecting from one reduces their total number of visits proportionally.

Players and defectors evaluate Casino6 in similar ways.  Both segments display a two dimensional pattern based on Casino Experience that accounts for 50-60% of the evaluations and Entertainment Value that accounts for another 5-10% of the evaluations.  The most striking difference between players and defectors is the evaluation of the “variety of games offered.”  Defectors also differ in their 10% greater consistency or structure of orientations in evaluating Casino6.

While there are clear differences between continuing players and defectors, there is no clear pattern of just casino visits that anticipates Casino6 defections.  Defection is deterred by factors related to the casino experience, especially general appearance, parking, and quality of games.  However these are some of the same factors that determine overall satisfaction for continuing players and defectors.  The single most obvious difference between continuing players and defectors at Casino6 is the rating of the variety of games not satisfaction with payouts.  This is not a driver of defection but its role in the evaluation attitudes distinguishes the two groups and could easily be a focus of defectors only research findings.  The old saw about low payouts leading to defection does not seem to apply because higher rating of payouts at Casino6 is associated with defection and not the other way around.  In fact, evaluation of slot payouts is a driver of overall rating of Casino6 for continuing players but not for defectors.  Nearly half of the overall evaluation of Casino6 is independent of the fourteen factors tested.  Indeed some of the evaluation of casinos is likely to be random or serendipitous, linked to specific events or experiences.  These players’ and defectors’ ratings indicate that “casino experience” factors are four times more important than the “entertainment value” factors for defection.  Players know what to expect in terms of entertainment at the casino but want it to be in attractive pleasant surroundings.

Conclusions

Defectors are not very different from continuing players.  Changes and enhancements to the casino need not focus on one segment to the exclusion of the other, but constant change is likely to pay dividends by providing a “new” or “different” experience for those who are change seekers.  Marketing to players and defectors should emphasize the casino experience and feature enhancement, improvement, and up-grades.

Collecting and analyzing objective data about customers prospectively provides benefits that are often overlooked.  The conventional wisdom is not always empirically true and anecdotal evidence may or may not have the degree of generalizability assumed within a company’s culture.  There is much that we think we know about our customers, defection, and defectors that should be verified before committing resources to retention or retrieval.  Prior knowledge and attendant segmentation of customers allows for better sample selection in a survey component of the research to support retrieval.  Such targeted sampling reduces dependence on aggregate results and increases external validity.  Continuously modeling risk or likelihood of defection based on the flow of customer touch points in conjunction with survey based models.  The greatest value from risk models is obtained when its explanatory power is increased due to the added knowledge gained form a survey component and risk can be estimated directly from objective factors without surveying all customers all the time.  This means keeping accessible records and linking them to customer interactions, satisfaction, and supplemental survey databases.

References

Allison, Paul D.  1984  Event History Analysis Regression for Longitudinal Event Data Sage University Papers Series on Quantitative Applications in the Social Sciences 07-046  Beverly Hills and London: Sage Publications

Campbell, Donald t. and Julian Stanley 1963  Experimental and Quasi-Experimental Designs for ResearchBoston:  Houghton Mifflin Company

Przeworski, Adam and Henry Teune The Logic of Comparative Social Inquiry1970 New York:  Wiley-Interscience

Findings from the Field

Emily McRae, Research Analyst, The Olinger Group

A regional casino decided to examine the profitability of changing their dining options. The Olinger Group designed and administered a survey, online and by telephone, to reach both casino guests and area residents.  All respondents were over the age of 21, lived in a regional area zip code, had dined out at a sit-down, eat-in restaurant in the area at least three times in the last three months, and were either the primary decision-maker (44%) or shared equally (56%) in the decision about where to eat out.Telephone survey participants were reached through random digit dialing or the casino’s player’s club list.  Online survey respondents came from a web-based panel.  Telephone participants were administered the same survey as the online participants over the phone.  225 people completed the survey via telephone, 215 online.  The overall sample was is proportionately female – 76% of the respondents were women versus 24% men.Of greater interest is that the sample for this study does not provide evidence for the conventional wisdom about the existence of a gender based “digital divide.”  In fact, the respondents for this survey contradict the idea that males use the internet more than females.  Significantly more women responded to the online survey than men – 54% of women completed the survey online, while only 42% of men did so.  Male respondents were more likely to provide their opinions by telephone – 58% answered the survey by phone, in contrast to 46% of females.  Whether these results describe usage or access, a time when the internet was disproportionately the domain of men may be over.  For this sample, concern that online surveys will not reach women as easily as they reach men is not only unfounded, it is misdirected. Further research should examine whether the “digital divide” has indeed reversed, or if it has simply been neutralized.