Retail Customer Lifecycle Segmentation and Modeling

Customers of a retail brand are diverse in many ways. They have different lifestyles and demographics. They physically live in different geographies. They have varying levels of disposable income and utilize that money in different ways. They have different tastes in regards to products and services that they purchase from other brands. And perhaps most importantly, they have varying levels of engagement and transaction history with the retail brand.

The rich, historical transaction data possessed by the brand for their customers are the raw materials needed to answer many fundamental questions about the business: How often should I expect my customers to purchase? How many customers do I really have who are 'high value'? What products purchased by new customers may indicate that they will likely purchase again? Are the customers I am acquiring today similar or different than the customers I acquired last year? Of my high value customers, who are likely to stop shopping my brand? We could easily fill 3 or more pages with insightful questions such as these.

Lifestyle Segmentation is a framework that we've designed at Wiland for the express purpose of asking and answering these questions. We've considered many of the common questions that retailers ask themselves and have built a process around delivering those answers. Much of what is to follow in this paper outlines these core questions with description of how we answer them.

But why are we seeking to answer these questions? Fundamentally, we are seeking to assist retail organizations by leveraging the insights that are generated from the answers. Most typically this takes the form of marketing promotions to customers. The marketing promotions are intended to influence purchase decisions in one way or another. Marketing to the customers with the right message at the right time with the right promotion is the goal. In addition, customer insights can be leveraged for merchandising decisions, operational considerations, or simply a deeper understanding of the brand's customers. 

Count and Describe

More specifically, Lifestyle Segmentation is an enumeration of existing customers across 3 dimensions: Value, Recency, and Business Segments. We determine Value based upon the predicted amount of future spend with the brand. Recency is the simple measure of how long it has been since the last purchase. Business Segments are the way the brand currently views customers within their marketing programs. Typically, these Business Segments are very much aligned with Recency and are simplified to acquisition (non-customers), new customers, core customers and lapsed customers. Once we have defined these 3 dimensions, the fun begins.  

Identifying the Recency and Business Segment of each customer is a simple exercise. What about identifying Value, however? We address Value by developing a predictive model that ranks all customers with the objective of predicting future sales. The development of this prediction utilizes multiple historical points-in-time in order to smooth out potential seasonal bias in the data. Once we have the final Future Value prediction, we can then understand where customers exist within the segmentation framework. A general illustration of the segmentation is illustrated below.


The most basic customer segmentation is now complete. We've counted customers and understand the size of the Business Segments. Marketing opportunities are starting to present themselves. For example, is the volume of Lapsed Buyers large? How many customers do we typically win back in a given season? How can we improve these efforts? Before we start pressing ahead with opportunities, however, we can add more color to this work via some interesting descriptive reporting.

Each cohort of the Lifecycle Segmentation can be compared on many different levels. Transactional data from the brand can be summarized to illustrate the magnitude of differences across segments. Demographic and Lifestyle overlays can be observed via Wiland's Demographic Insights Report for each segment. Distribution of the segments based upon proximity to the brand's physical locations can also be generated. Finally, transactional data from across Wiland's cooperative clients can be summarized to illustrate how the brand's customers are spending their disposable income.

Key Advantages: Data Depth and Superior Strategy

Wiland offers multichannel retailers a tremendous advantage: a massive database with a diversity and depth of transactional data that surpasses other industry data sources. We track not only what, when, and from whom consumers buy, but also the channel for each purchase. With over 2,500 participating clients, the Wiland database includes transactions from large, well-known retail brands, huge internet retailers, small specialty merchants, and everyone in between. Collectively, our client base offers consumers virtually every product category imaginable.

In addition to merchandise transactions, Wiland has unsurpassed publishing and nonprofit data. Our database includes the magazine and newsletter subscription activities for 100 million consumers, as well as their donation activity across hundreds of charities. These non-merchandise behaviors are often strong predictors of merchandise purchases. Our database also includes demographic and psychographic data on virtually every consumer in the United States. The result is a robust, 360-degree view of customer buyer behavior that provides retailers deep understanding of their customers that goes well beyond primary research activity.

Another key advantage is our dedication to good marketing strategy and advanced analytics. We understand the problems that face marketers in today's environment. We provide a flexible analytical platform and consultation that delivers actionable insights. This difference in strategic approach makes us a superior analytics partner. Clients that work closely with us and share their goals will ultimately improve their brand as a result. We strategize. We improve. We refine. And ultimately, we deliver measurable results.

Now Let's Act

Whether the main objective is new customer acquisition, improving value of current customers, or reactivating lapsed customers, Wiland stands ready to help retailers realize a better future.

In the sections that follow, we will delve further into successful Lifecycle Segmentation strategies and discuss how Wiland helps clients execute them to meet their objectives.

The Fundamental Opportunities

Business Segment Goals  

Each Business Segment presents an opportunity for the retail marketer to improve productivity or campaign results. We typically assume that improved productivity or improved results refers to increased ROI of marketing campaigns. Let's start by outline the fundamental opportunity for each Business Segment:

Acquisition Audience

How do we attract new customers to the brand?

Identifying the right consumers who will purchase from the brand and become solid, high-value customers is critical to the long-term success of the business. It is so critical, that we've produced a separate paper outlining our methodology for success. This paper is called New Customer Acquisition for Multichannel Retailers. Please refer to this work for an extensive perspective on how to acquire the right new customers.

New Customers

Of all the new customers, who will turn into a high value, repeat buyer?

We've successfully convinced consumers to purchase from our brand. Great. Now how will we treat these new customers with marketing messages? Should we treat them all the same? Understanding which of the new customers have the propensity to turn into high value repeat buyers provides the marketer with a tool to accelerate the process and remove marketing waste on low value new customers.

Core Customers

Which of the core customers are at risk of leaving the brand?

The core customer is the lifeblood of the brand. They produce all the revenue and profits. The brand wants to keep these customers happy by delivering relevant products at the right price. Identifying customers who are likely to never purchase again and customers who will provide positive ROI if promoted is imperative for the brand. These predictions are very nearly inversely related, but not always as directly as one would suspect.

Low Value Customers

Which of the low value customers can be transformed into high value customers?

Converting recent, but low value customers is an attractive way to grow the business. Often times a consumer tries out a brand with a small purchase within an introductory product line. Other times, they are simply buying a gift for someone else. The identification of the most likely converts from low to high value provides the marketer with an important tool. The marketer now understands which consumers will provide future value.

Lapsed & Dormant Customers

Which older customer can be won back to make a new purchase?

Most retailers have a very significant volume of buyers who have left the fold for one reason or another. Some customers are buying competitive product while others have moved on from that type of merchandise. Determining which lapsed buyers are most likely to purchase again and which should be avoided improves marketing efficiency and ROI. Utilizing the vast data within the Wiland database is critical in the prediction.

We've put forth these goals for each Business Segment. There are, of course, many slight variations of these goals and some others not yet discussed. We seek to provide the best solutions for the common goals, but also stand ready to adjust and develop solutions that vary from this path. 

Periodicity

Up to this point, we have discussed Recency and Business Segment in general terms. An important aspect of Lifestyle Segmentation is the understanding of a customer's typical cycle of purchases. What is the typical lag pattern between the first and second purchase? For Core Customers, what is the typical interval between purchases? The type of retailer and product sold obviously leads to completely different patterns. A car dealership will have many one-time buyers and extremely long periods between second and third purchases, if any. A grocer will have the expectation that a good customer will visit their store weekly, if not even more frequently.  

The concept of Business Segment should align with likely lag patterns of purchase. Most retailers intuitively devise the Business Segment and frequently settle upon a couple of common views. One view slots New Customers as those with a first purchase in the last 90 days, Lapsed Customers as those without a purchase in the last 2 years and the Core and Low Value Customers living in between. The second common view has New Customers as those with a purchase in the last month and Lapsed Customers without a purchase in the last year.  

We validate or help to establish appropriate Business Segments within the Lifestyle Segmentation work that we complete.

Customer "Migration"

Marketing campaigns are designed to grow awareness of the brand and ultimately increase sales. Each marketing campaign can be viewed independently and be measured on an ROI basis. If marketing campaigns were tailored to Business Segments, however, the success could be measured both on a campaign ROI basis as well as by how much the campaign affected customer purchase patterns. For example, if a series of campaigns were designed to upgrade customers from low value to high value, we could calculate the rate of movement from the Low Value to the Core Customer segment. But what do we have to compare that rate of movement? Is the rate of movement acceptable or not? The method that we use to determine these answers is calculating customer migration through all Business Segments.  

In the Count and Describe section above, we discussed enumerating each segment of the grid and presented the generic illustration the concept. Customers are not confined to a Business Segment. The customers' Value and Recency change over time as they either make or do not make new purchases. There are certainly many influences on the change of customer behavior including marketing efforts, product assortment, promotions, competition, the economic conditions, and so on. Assuming all things being equal, we can observe the migration rates of customers from one Business Segment to another. We accomplish this task by viewing the customer database at a point in the past and compare counts by Business Segment in this historical view to where those customers ended up today. This concept is illustrated in the following table.

 


Seasonality

Measuring the migration rates of customers from one point in the past to today's view of the customer file is insightful. It can also be a bit misleading depending upon how today" is defined. Seasonality in the business drives different migration rates at different times of the year. For example, it is typical for the holiday season to drive a large volume of new customers. Conversion of the New Customers to the high value, Core Customer segment is generally more difficult and thus is accomplished at a lower rate. In off seasons, the raw number of new customers is lower, but the rate of conversion is higher.

It is for these reasons that we will actually observe multiple windows of migration and average the rates across the whole year. Using this approach yields an average migration for the whole year from segment to segment. Once again, note that these observed rates are based upon "business as usual" and influenced by many factors both within and outside of the brand's control.

Improving upon Baseline Migration

With the knowledge of customer migration rates in hand, marketing efforts can be measured both on the short-term, campaign ROI basis as well as a longer term view. What is this longer term view? The answer: improve the rates of migration in a favorable way for the business as follows:  

  • Convert New Customers to Core Customers at a higher rate
  • Lose less Core Customers to the Lapsed Customer Group
  • Upgrade Low Value Customers to Core Customers at a higher rate
  • Return Lapsed Customers to current Customers at a higher rate  

The goal seems pretty simple. How can a marketer make it happen? It takes the marriage of the science of predictive analytics and the art of creative marketing concepts and ideas. Wiland is skilled at delivering the science for our clients. Let's dig deeper.

Act via the Power of Predictive Modeling

New Customers: Just How Valuable Are They?

There is a wealth of information available on your brand's new customers. On the date of first purchase, assuming capture of name and address, the customer's demographics and lifestyle information is known. All of the details of the first purchase are fully understood including order channel (e.g., store or web), payment method, and transaction detail (revenue, coupon redemption, product category and specific SKU level information). Through the power of the Wiland database, the same type of purchase data from over 2,500 other brands is also known. This data is extremely powerful.

One practical use for all of this data is to immediately assess the potential future value of new customers based upon this "Day 1" knowledge. How do we do it? We develop the New Customer Future Value model to predict sales from a given new customer. To do so, we examine new customers at a historical point in time with all of the characteristics of their first purchases and then observe future spending, typically over the time period defining an active customer. Consider the following timeline.


This illustration is for just one cohort of new customers. Remember the previous discussion about seasonality? It is clearly an important consideration when developing the New Customer Future Value model. We actually identify many new customer cohorts throughout the entire calendar year and roll all of it together into one analytical dataset to remove the bias from the final model. 

Once complete, we have a prediction of future value that can be applied to all new customers. We then have a deeper understanding of which new customers will be of value and those that are highly unlikely to transact again. This information can be used in near real time within marketing campaigns to message and promote appropriately. A simplified illustration of model gains and the cumulative effects of more targeted campaigns are illustrated in the table below.



Finally, with improved targeting and messaging for New Customers powered by the newly developed Future Value model, we can measure the long-term conversion rate to the Core Customer segment. We established the brand's typical conversion rate in the Migration Analysis. The end result will now be an improved migration rate resulting in more Core Customers.

Core Customers: Who's at Risk and Who has Long-Term Value?

Core Customers pay the bills. They are the reason why the brand is in existence and operating. To keep the business healthy, the retailer must continue to deliver products that are relevant and timely for the Core Customer segment. Virtually all of the profitability of the brand comes from this Business Segment. It is, therefore, imperative that this segment is well understood in two ways: which of these customers are at risk of not purchasing again (attrition risk) and which of these customers have the highest future value.

Predicting future value for all Core Customers can be executed in the exact same way that future value was predicted for New Customers in the previous section. For this reason, we will focus on the Attrition Risk model.

In the migration analysis, we determined the typical rate the Core Customers stop purchasing and move into the Lapsed & Dormant segment. The goal is to reduce this rate of decay. Once again, we utilize historical views of customers to construct our analytical data for model development. For example, we observe all Core Customers with a purchase specifically in the window of time 12 to 13 months ago. At this point in time, we have all of the historical purchase data on these customers from the brand, demographic and lifestyle information as well as the historical transactional data from the Wiland database. We then observe which of these customers makes a subsequent purchase from the brand. We utilize all of the historical data to predict attrition, or a lack of a subsequent purchase. This analytical design is illustrated below:

 

 


Once again, seasonality is an important factor. In order to effectively manage the seasonality effect, we construct an analytical sample from multiple points in time as described above. The resulting model now paints a clear picture for which customers are likely to never shop the brand again.




Marketing promotions and messaging targeting the Core Customer segment can now be customized based upon Future Value and Attrition risk predictions. For example, customers with high attrition risk and low future value can be removed from direct marketing campaigns as their responsiveness is too low for the marketing cost. Conversely, high value and high attrition risk customers may be given special offers to keep them engaged with the brand.

Low Value Customers: Who has Potential for High Value?

All brands have a set of customers that remain somewhat active with purchases yet they don't contribute large sales and profitability. This Low Value segment tends to be a mixture of customers that includes those relatively new to the brand, those that have limited but constant affinity with the brand, those with high value potential and those who were simply "samplers". It is important to identify those with high value potential and those with constant affinity. The Upgrade Propensity model accomplishes this goal.

We construct our analytical data in much the same fashion as we did in the previous modeling exercises. Historical cohorts are determined over a length of time to remove seasonality and customers are tracked to determine who significantly increased their spending, thus moving from the Low Value to the Core Customer segment. We build the model to predict this movement. The illustration of analytical design and model gains are as follows:


 

 

Armed with this information on the Low Value Segment, marketing efforts can focus on the right customers and save cost on those that have virtually no chance of ever buying again. Effectively, an acceleration of movement from Low Value to Core Customer segment can be achieved.

Another useful, but complicated analysis can also be pursued: Are there characteristics of purchasing early in a customer's interaction with the brand that indicate that a customer will never upgrade? This type of analysis can be completed and is discussed in the paper called New Customer Acquisition for Multichannel Retailers

Lapsed & Dormant Customers: Who can be Won Back?

Convincing a lapsed customer to purchase again is the cheapest way to grow the business and the current Core Customer segment. Most established brands have a considerable volume of Lapsed & Dormant Customers making this segment very attractive for marketing activity. Due to the fact that the segment is large, marketing efforts could be costly if an effective predictive model is not in place. Enter the Reactivation Model.  

There are generally two different methods used to develop the Reactivation Model. The first entails a similar analytical design to the previous models discussed in this section. Namely, historical views of the Lapsed & Dormant Customer segment are created and then reactivation purchases are tracked. Historical data elements are then used to determine which customers are most likely to purchase again. 





The second method for the Reactivation model comes directly from the use of past reactivation campaigns. If direct marketing promotions were used to spur reactivation to a cohort of Lapsed Customers, these campaigns and their responses can be used directly to develop the model. The following model gains chart illustrates the gains from one of these promotion-based Reactivation models.





There is a general rule of thumb that the longer that it has been since a customer's purchase, the more that customer starts to "act" and to "look" like a pure prospect. Along this line of reasoning, the power of the Wiland database transactional data becomes increasingly more important the longer it has been since the last purchase.

Other Consideration & Thoughts

Some Observations on the Key Drivers in These Models  

The last section closed with a brief point about the power of the Wiland database for Lapsed & Dormant customers. Throughout all of the model discussions, however, there was little detail given about "variables" and key characteristics often found in the modeling exercise. The following table speaks to this point, summarizing generally the experience derived from building literally thousands of models over the last 10 years here at Wiland.




It can be seen from this summary that the brand's transaction data is most important when dealing with current customers. However, we have observed that the Wiland transactional data helps to sharpen the line for "promote/do not promote" decisions. The absence of Wiland data in the model solution increases inefficiencies in the targeting and thus reduces campaign effectiveness.

Modeling for Specific Outcomes

This paper has discussed the common methods for modeling and predicting customer behavior. There are certainly other actions and specific business initiatives that can be predicted with the rich data.  

Some brands utilize proprietary credit and seek to have more customers possess this line and use it when buying. The brands, therefore, create and execute specific promotional campaigns focused on credit adoption. Response modeling can be utilized to support these campaigns and eliminate wasteful promotions on those customers highly unlikely to respond.  

Incremental revenue models are also used to measure the truly attributable revenue associated with a campaign. To accomplish this type of model, the brand must select an appropriate random selection of customers from the treatment population and hold them aside as a control group and not promote them. The modeling exercise then seeks to predict incremental revenue instead of total revenue. This type of modeling often leads to curious results where portions of the Core Customer segment are predicted to not be influenced by marketing promotion. Removal of Core Customers with high predicted Future Value from future campaigns can be a risky proposition. We recommend extreme caution and conservative test approaches to mitigate risk.  

Another broad area of modeling revolves around predicted sales within certain product lines or product categories. Understanding which customers have affinity with a given product category can be useful in many different marketing channels. Optimizing landing pages on the web site, creative content for digital advertising and email marketing, or circulation for direct mail with specific product focus are a few of the common applications.

There is one important point to make in regards to the development of predictive models. Wiland has created a model development environment that utilizes a rich volume of data. Our systems allow for the creation of any custom model that can be conceptualized given the availability of the data. Further, we expect to share the findings of the analysis in great detail with our clients in a consultative manner at each step of the process. This improves our client's understanding of their customers. It also provides Wiland with all of the necessary background information needed to deliver a superior product.

No Need to Boil the Ocean

The descriptive statistics available for a brand's customers and the predictive modeling tools discussed in this paper are not trivial in their creation and execution. Marketing professionals need flexibility and creativity in their jobs to fully utilize the work. It doesn't have to be difficult, however. Each predictive model and descriptive report can be generated and utilized on its own. Often a great way to get started is by identifying an area of opportunity and then executing that one area well. A very common example of this approach is Reactivation Model. Clients understand that the Wiland data provides insight into Lapsed Customers and their activity with other brands in the same category. This translates into a good predictive model used for promotions. The client tests these concepts first before branching into other customer segments.

Conclusion

Hopefully the material presented here is an eye-opening new view of the potential of predictive modeling in the context of customer marketing. We call it Lifecycle Segmentation because we want to understand where the customer resides in their cycle and we recognize that the customer's behavior may change. This helps the marketing pro understand their customer in a much deeper way. That understanding is great, but it is the actions available given the predictive modeling that allow the marketer to manipulate the migration of customers in way that is more profitable for the business. Let's acquire better customers, upgrade them when necessary, keep them engaged longer and win them back if they get away.

We're Here to Help

We are committed to helping retailers thrive in today's challenging and evolving marketplace by improving their customer base and supporting their brands. The solutions that we provide generate superior results and contribute to the viability and vitality of every brand we serve.  

Contact us today to learn more about our entire suite of customer solutions and other services for retailers. We are here to help.

Contact Us