Is It True Attrition, or Is Data Capture the Challenge?

Sometimes what appears to be customer attrition may actually be another concern entirely, such as data capture and proper customer identification.  And as you add more channels to facilitate communication with a single customer, data capture and customer identification concerns are that much more prevalent.

Every organization has its own definition of attrition.  For the typical retailer, attrition may be defined as a customer who hasn’t made a purchase in the last 12 months.  Definitions can vary depending on business type and average customer purchase frequency.  Understanding attrition, and developing reliable indicators as to what or what does not precede attrition, can help a business from losing profitable customers and overall revenue.  

On average, most retailers experience 15% to 65% customer attrition within a 12-month period.  Losing customers on this scale has a negative impact on business, since acquisition (and reacquisition) costs so much more than retention.  Today, retailers need to learn as much as possible about customer attrition within their own organizations.  Questions often asked: how much inactivity is true attrition – and how much can be attributed to incomplete data capture?  How can retailers determine who stops buying and why?  

Attrition can only be eased when marketers understand true attrition.  Without having 100% data capture, across all channels, which feeds back to a single repository, it is difficult to determine who has truly stopped shopping, vs. who stopped using credit cards and/or providing phone numbers at the point of sale.  

Market research is one way to help marketers differentiate between who is still shopping but labeled "inactive" and who truly is no longer shopping.  With imperfect data capture, the mere fact that we do not see a customer in a particular period does not mean that she or he did not actually shop.  Market research provides the ability to go to those customers who have no purchase data in a given period and ask them whether or not they shopped.  Hence, all those customers who appear to be inactive on the database can be sorted into those who were truly inactive (true attritors) and those who shopped but were not picked up by the data capture process.  While it is not possible to perform market research on all inactive customers, it can be done on a sample of customers.  Then you can create a model from data that exists for all customers in order to predict – based on their transactional and demographic characteristics – which ones are more likely to be true attritors.

The first step to monitoring attrition is to determine overall attrition rates.  This is accomplished by tracking on a monthly basis the total number of inactive customers (whether real “inactives” or not).  The next step is to determine attrition rates by customer segment and ask, “What is the cause?”  Known factors that could affect attrition calculation could be loss of data capture (as a result of store compliance, state point-of-sale privacy laws, inability to identify and match a customer shopping through another channel, etc.); location status (store moved or closed, competition); merchandise issues; lack of advertising; store operational issues – or a combination of reasons.   

Determining the predictors of true attrition helps marketers develop models to help predict customers that have truly attrited, as well as those that are at risk of attriting.

Case in point: A national off-price department store needed to determine the impact of customer attrition on its business and thereby achieve a better understanding of:

• Overall attrition levels
• How much was attributable to natural attrition vs. data capture issues
• Among “true attritors,” which customer groups have the highest percentage of attrition?  Are some regions affected more than others?  What are some of the reasons as to why customers have stopped shopping?

The retailer selected two customer file samples – active shoppers and those thought to be inactive shoppers (no known transactions on the database for last 12 months).  Demographic data were appended to both files and quantitative research was conducted among “inactives” to identify those customers who are true attritors and reasons for their behavior.  Results indicated that what appeared to be attrition was primarily a data capture issue.  Most of the customers who had no available purchase data in the past 12 months actually did come into a store and buy something – they shopped but were not picked up by the data capture process.

This research was used to define who was a true attritor and who looked like one but was not.  Customer transaction analysis then was performed using indices to identify key characteristics that differentiate an attritor from a non-attritor within the population of those surveyed.  The intent was to see if there were behavioral differences between the two groups that could potentially be used to predict actual behavior.  For example, 10% of those with characteristic “A” are non-attritors and 20% of those with characteristic “B” are attritors.  The resulting index: 10% divided by 20% equals 0.5.  Therefore, those with characteristic “B” are twice as likely to attrite as those with characteristic “A.”

From the analysis, the retailer learned that: Customers who spend more and have more visits during the analysis period are less likely subsequently to attrite; non-attritors are more likely to pay with cash or credit card (and less likely to pay by check); and customers who shop within specific merchandise categories are less likely to attrite.  Behavioral characteristics – including the average spend per customer, number of visits per customer, and top categories for spending – also differ between non-attritors and attritors.

From here the merchant developed a powerful yet simple predictive model to predict the likelihood of being or becoming a true attritor.  The model included four predictors – total categories within the retailer, month of visit, total returns, and total spending.  All active and inactive customer groups were scored.  Active customers were targeted to gauge if those customers with high value and low attrition scores would constitute a lower percentage of inactive customers at the end of a specified period of time.  Inactives were targeted across both low and high attrition scores, to determine which would be more likely to respond.

Positive test results would, for active customers, help determine when a retention trigger program, involving coupons or private sales, for example, should be introduced.  For inactive customers, scores were used to help identify those customers that are shopping vs. those that are true attritors, as well as those to whom a reactivation program should be introduced.

This particular retailer was able to use this model to derive new understanding of its customer base.  It also realized that managing attrition must coincide with managing data capture – as true attrition, while a serious business concern, only can be adequately measured if accurate customer data capture, recognition, identification, matching, and value determination precedes it.  At the end of the day, the retailer successfully combined market research with data analysis to gain valuable insight that prompted appropriate marketing intervention.

Charles Noland is vice president, data mining and modeling, for San Antonio-based direct marketing services provider Harte-Hanks.