Engage Your Seasonal Customers During Their Off-SeasonPosted in Customer Retention, E-Commerce, Data Analysis, Personalization on 27 February 2014 by:
Many types of online businesses have customers who tend to be active only during particular times of the year. Just a few examples are online retail stores (many customers use particular stores to shop only for particular types of seasonal products, mostly ignoring the store the rest of the year), sports betting sites (many customers are only active during the season in which their favorite sports are played) and travel sites (many customers tend to cluster their purchases during summer and/or winter vacation periods).
In all these cases, an obvious opportunity presents itself: entice these customers to make different kinds of purchases during different times of the year. In addition to simply generating additional revenues from these currently one-season customers, the increased year-round engagement enhances loyalty to your brand and improves the chances that these customers will not go elsewhere when their primary season comes around again next time.
In this post, we’ll take a step-by-step approach to identifying seasonal customers and how to encourage them to make purchases outside their primary product category. This approach is based on analytic methodologies similar to those used in market basket analysis, affinity analysis and collaborative filtering.
Step 1: Segment Your Customers into Personas based on their Product Affinities
The first thing we need to do is segment our customers using cluster analysis, with the goal of discovering interesting “personas,” which, in this case, are groups of customers with strong affinities for a particular product/category. We’ve covered this step in detail in a previous post, Behavior-based Customer Segmentation for More Effective Retail Marketing. You might want to read that post before continuing here.
To illustrate the process, let’s walk through a particular example from the world of retail: Tim’s Sports is an online sporting gear shop, owned by Tim. Following the guidelines in our customer segmentation post, Tim has segmented his customers into five particular personas, according to their sporting gear buying preferences. As we can see in the following results, these five customer segments exhibit dominant purchase behaviors in one particular product category (click any image in this post to enlarge it):
In a sports betting site, these personas might be Football, Basketball and Hockey bettors. For a travel site, these personas might be Caribbean Vacations, Europe Vacations, Asia Vacations and North American Vacations. Whatever the industry, cluster analysis will almost always reveal a number of clear personas.
Step 2: Plot Each Persona’s Annual Revenue Trend
Once we have identified our personas, we want to plot each one’s purchase trends throughout the year, to see if there are clear seasonal buying trends for the various personas.
When Tim did this with his five personas, he got a very clear picture of how seasonal each one is. Runners make their purchases, as a group, relatively evenly across the whole year. Cycling Enthusiasts and Mountain Climbers spent more in the summer than in the other seasons, but their purchases in the other seasons are not insignificant. However, Fishing Enthusiasts and Skiers each placed the majority of their orders in one particular season of the year:
Based on this analysis, Tim decided to try develop effective marketing campaigns with the goal of enticing his Fishing Enthusiast and Skier customers to make additional purchases outside of their regular buying seasons. Needless to say, bombarding these customers with irrelevant offers would probably just annoy them, push them towards the competition and make them apathetic to Tim’s brand and promotions. So it is important that Tim figure out what kinds of offers his customers want to receive.
OK, so let’s see how Tim determined what offers to send to his customers to encourage them to make purchases during other times of the year.
Step 3: Discover the Most Promising Cross-product Correlations
Once we have our interesting customer groups in mind, we need to find those additional product categories with the greatest likelihood of appealing to the customers of each persona at other times of the year.
The method of doing this is to analyze the data, looking for instances where customers purchased from two different product categories and calculating the likelihood of such crosses to occur across all customers. The metric expressing this connection between two categories is called “lift” (see Appendix 1, below, for details).
Let’s use Tim’s example to illustrate: Since Tim is interested in the Skiers and Fishing Enthusiast personas, he calculated the lift between purchases in the skiing department and all of the other product categories. He did the same for his Fishing department:
Tim’s results (detailed in Appendix 2, below) indicated that there are clearly additional product categories with higher statistical likelihood of appealing to customers in each of our two personas. For example, he discovered that Skiers are more than 2.5 times more likely to purchase running products, as compared to any randomly chosen customer in Tim’s customer database. Likewise, Fishing Enthusiasts are 30% more likely to buy cycling gear.
Step 4: Take Action!
Armed with this extremely valuable insight, Tim crafted a number of customer marketing campaigns specifically designed to entice his Skiers to purchase running and cycling products in the spring and summer, and to entice his Fishing Enthusiast customers to purchase cycling gear in the fall and winter.
Tim was careful to run these campaigns as marketing experiments, in order to measure the success of the campaigns compared with control groups. He also tested the conclusions of his cross-sell analysis by experimenting with sending similar campaigns to groups other than those he identified with his analysis. Across the board, the results were impressive – and statistically significant!
The next step is to track customer retention (i.e., churn rates) among the targeted groups the next time their primary season rolls around. The goal of this is to check if customer attrition in those customers who received these cross-selling campaigns is lower than in those who didn’t – which would indicate a greater long-term impact of this approach than simply increasing orders in the customer’s “weaker” periods. Tim is waiting to answer this question over the next 12 months.
An important part of understanding your customers holistically – and maximizing their brand loyalty – is discovering how to engage them, even outside their normal buying periods.
Today’s customers know what they want, and expect you to know as well. If you’re not catering to your customers’ exact needs and passions, you’re almost certainly falling behind your competitors.
While customers are sometimes overwhelmed with marketing, when you charm your customers with relevant offers at the right time, your messaging will get through and customer loyalty will grow. Loyal customers are much more engaged with your brand, spend more and are less likely to churn.
Appendix 1: Lift
Lift, in the context of this article, indicates the likelihood of a customer that purchased from category X, to purchase from category Y in the future, as compared with a randomly selected customer from the entire customer base.
For example, Tim calculated the lift between Skiing gear and Running gear by determining what percentage of all customers who purchased skiing gear, purchased both skiing gear and running. Next, Tim calculated what percentage of all customers purchased running gear. Finally, Tim determined the lift from Skiers to Runners by dividing the first result by the second result. In mathematical terms, it looks like this:
- P1 = number of customers who bought product category 1
- P2 = number of customers who bought product category 2
- C = total number of customers
To learn more, take a look at the Wikipedia article, Lift.
Appendix 2: Tim’s Calculation Tables
This is what Tim’s calculation table looked like:
Tim applied the lift calculations (shown in Appendix 1, above) to the figures in this calculation table, resulting in this table: