By measuring the lifetime value of a customer, companies can make decisions about which what efforts should be made to recruit and/or retain these customers. The article presents a customer value model and explores how to construct it.
Editor’s note: Albert Fitzgerald is president of Visions Research, Del Mar, Calif.
Today’s companies are expending great efforts to retain current customers and win new ones. After all, it is cheaper to keep a current customer than it is to acquire a new customer. But what if some customers are actually costing your company money instead of contributing to the bottom line? Not all customers are profitable. That is why it is beneficial to estimate the lifetime value of a customer.
If you can predict the lifetime value of current and prospective customers, you can make educated decisions on what efforts should be made to retain or recruit these customers, if any. – Dr. Peter Fader, Wharton Business School
Transactional data can be used as a starting point to estimate the lifetime value of each customer in your database. By transactional data, we mean the information that most firms have about their customers – how much they buy, which products they have purchased, how often they purchase, the date of their last purchase, etc. While this is useful information when evaluating current customers, it provides minimal information to identify potential customers and what their value might be. Transactional data tells us what happened but not why it happened. Supplemental research can provide rich insights into why customers buy and provide the insights needed to estimate the value of current and prospective customers.
Past behavior tends to predict future behavior. A company’s current database of transactional data can be used to estimate the current value of a customer. This is done by calculating the customer’s past purchase behaviors/patterns and using them to project future purchase behavior. Included in this is what the customer purchased (contributed to the bottom line) minus what the customer returned/rejected (subtracted from the bottom line). These results are extrapolated and projected over time. The end result is an estimate of the value of a customer. We can calculate this for each customer.
While this gives some estimate of our best customers, it does not tell us if there is a customer who should be buying more from us. This approach only tells us what the customer did and projects that into the future. It does not explain why the customer is buying from us. They may be buying from us because they like our features, we have an attractive price, our service and support meet their needs or they are loyal to our brand. Transactional data alone cannot give us this insight.
To estimate the value of current and potential customers, we can develop a model that estimates the lifetime value of current customers and prospective customers. To accomplish this:
Ideally, we would like to understand why a customer is buying from us rather than our competitor. An ideal method to start is to conduct survey research to classify customers into groups using a segmentation methodology. Asking customers about their needs and attitudes helps us understand their motives and understand why they buy what they buy. Don’t forget to interview non-customers as well. When you interview customers and non-customers, you can see which segments most of your customers fall into.
Let’s say that we conduct a survey of that includes both customers and non-customers. We segment our customers and prospects based on their purchasing attitudes and user needs. As an example, let’s pretend that we uncover three main segments: Price Seekers, Feature Seekers, and Brand Loyalists.
The average lifetime value of customers for each segment is then calculated. This is done by using the data from current customers in each group. Each segment is evaluated for its profitability per customer and by the size of the segment. Using these customer lifetime value results, we can target the profitable segments. Some segments may have profitable customers but they are not large enough to warrant expending resources to target them. We now have a more effective and efficient way of identifying which customers to target and retain – those that fall into the most profitable segments with a sufficient size.
From this example outcome, we determine that we want to focus most of our attention on the Brand Loyalist customer; our second focus is on the Feature Seekers. By looking at the purchases and profitability of our Brand Loyalist customers we can tell they have the highest customer value, followed by the Feature Seekers. However, our customers who fall into the Price Seeker segment are fairly unprofitable.
After the segmentation model has been developed and average lifetime values have been estimated for customer segments, the segments can be profiled. For example, what are the characteristics of our most profitable segment (the Brand Loyalists) that distinguish them from our least profitable segment (the Price Seekers)? Looking at transactional data alone is usually not sufficient. While our historical transactional data may give insights into our current customers, we do not have any transactional data for new prospects. Therefore, we want to add additional demographic data.
There are many sources from which we can buy demographic data on our customers and prospects. One such source is Dun and Bradstreet, but there are many others. Since we know the address of our customers, we can purchase additional data about these customers such as what industry they are in, the number of employees in the firm, their credit scores, the type of office building they occupy, etc. Literally hundreds of pieces of data are available to be purchased. This is true whether our customers are consumers or businesses.
After purchasing the additional data, we append them to the database of transactional and segmentation data. We now have a rich source of descriptive data to use to differentiate between high-value and low-value segments. Mathematically, we can determine which of these descriptive data points is most useful in predicting segment membership. Typically a mathematical model is developed using either discriminant analysis or another data mining technique. Suffice it to say that a highly accurate predictive model can be developed. While we may have surveyed only a few hundred customers and prospects, we are now able to develop a mathematical model that will predict segment membership even for customers who did not take part in our survey. As long as we can gather the descriptive data needed for our model, we can accurately predict the lifetime value of a prospect before they make their first purchase!
Profiles from the new database can be used to target non-customers in the desired segments.