How to Create and Use an Attrition Model

A continuous threat to the health of any company is the loss of customers.

That makes prospecting efforts all the more important, because newly acquired customers are worth more in the future. Profits rise because repeat sales rise, and most profits come from repeat sales.

Your catalog’s attrition can be mapped using many variables. Recency, or how long ago customers made purchases, is the most common. Total spending, average amount spent, method and type of purchase are common variables used to estimate attrition by customer segments. Demographics and psychographics also may be helpful predictors.

A common strategy for marketers is to re-contact all customers who made a purchase within a certain time frame, typically two to three years. This may vary based on seasonal demand, but customers who haven’t made a purchase after a certain length of time are dropped and assumed lost through attrition.

All Customers Aren’t Equal

While the popular method of dropping customers after a prescribed time period is simple and easy to implement, it assumes all customers are equal, which is rarely the case. The diagram “Re-Contact Profitability” shows the profitability of such efforts using customer segments based on recency and the amount of their first purchases. The dashed line shows the break-even point, below which the organization loses money when re-contacting customers.

Notice how an organization with results like those shown would likely make an overall profit contacting all customers for 24 months. However, some customers are in segments that are contacted at a loss. Other customers are in segments that could be contacted profitably if reached beyond 24 months.

In some cases, such as low initial orders, customers are unprofitable if contacted within a few months. It’s likely you’ll spend more money re-contacting them during the next 24 months than you made on the initial sales. In other cases, such as a higher initial sale, the highly profitable results more than make up for a loss in lower segments, but re-contacting is no more frequent and stops at the same time if no second purchases are made.

Many Variables

While it’s easy to graph and understand a segmentation strategy based on two variables, the reality often is more complex. Advanced statistical techniques are most useful when a multi-dimensional relationship best describes customer behavior.

Example: Segmenting based on recency, frequency, average order size, product category, or customer’s age or gender is impossible to graph in black and white on a flat sheet of paper. This doesn’t mean that segments must be so numerous and segmentation rules so complex it’s impossible to implement a marketing strategy using multi-dimensional data based on statistical segments. On the contrary, a good statistical analysis should simplify relationships that are otherwise too complex to logically determine.

Grouping techniques, such as CHAID or CART, are beneficial if you need a segmentation strategy that’s predictive and segmented. Segments are different in their predicted behavior, and also tend to differ in logic, as well. This helps you develop offers and creative messages that target differences among customers.

The number of segments is kept to a reasonable, actionable total (usually 10 to 30 segments), and each segment is unique. There’s no overlap among segments, but they’re often combined for marketing efforts.

Example: Three of your 18-to-35 segments might get one offer, while customers age 35 and older might get another.

Segmentation Leads to Contact Strategy

Marketers often begin an attrition effort by looking for areas of waste. Attrition models are good at estimating when customer segments will become unprofitable to re-contact. However, they can’t predict how well new efforts, targeted to profitable customer segments, will perform, or how well offers perform out of season.

Using statistical segmentation to reduce attrition likely will implement a different, and often more complex, contact strategy than before. While it may make fewer overall contacts, the more frequent, more tailored contacts require more work to produce on an individual basis. Fortunately, companies tend to underestimate the amount of money spent on re-contacting unprofitable customers, so overall cost increases aren’t always necessary to implement a new strategy.

Increased profits from more tailored, and in some cases more frequent, communication with customers is the goal of a good strategy based on an attrition model.

Attrition is Not Just a Marginal Customer Problem

Effective use of an attrition model starts way before determining which customers are marginal, and then sending them win-back offers. It begins with identifying which customers can be contacted profitably before they become marginal, and seeking repeat purchases that will cause them to remain profitable.

The best time to attack attrition is when someone becomes a customer. Seek repeat purchases, increased average orders or purchases of additional product categories quickly while their patronage is fresh.

Example: An offer that accompanies a welcome to a new customer might perform 10 times better than a “we-want-you-back” offer. It would be a shame for customers to get a win-back offer without ever having a welcome offer.

All customers have some likelihood of responding to offers. The challenge is timing the offers, finding the most productive and actionable segmentation. Waiting to create win-back offers for customer segments about to drop below break-even profitability is action that’s too little, too late.

Strategies Must be Tested and Tracked

An attrition model can be used to formulate an initial strategy, but once the marketing efforts change, testing is crucial. While it’s unlikely that customers who spent a small amount two years ago suddenly will respond better than recent buyers of high-ticket items, they may behave differently if given targeted offers geared toward their segments.

Why? For some segments, increasing average order is important, and offers should be created accordingly. For others, response might be low, and offers are geared toward getting another sale. Some segments respond to offers that are non-seasonal, while others may wait for a seasonal offer.

Testing and tracking new results will enable you to learn from the new, segmented offers. That will build on your knowledge gained from the attrition model, and allow for better results in time. Alan Weber is CEO of DataPlus Millennium, a Prairie Village, KS-based provider of marketing database consulting and analytical services. He is the former president of the Kansas City Direct Marketing Association, co-author of “Desktop Database Marketing” and a teacher at Kansas University. You can reach him at (913) 432-8311 ext. 25, alan@marketinganalyticsgroup.com.