In a recent interview with The Wall Street Journal, JCPenncy CEO Marc Rosen said that they aren’t trying to win new or trendy, younger buyers. The 120-year-old retailer plans to focus its energies on customers that ‘love’ the brand.
Who are these customer’s who love a brand?
For JCPenny these are customer’s who already shop there : budget-conscious American families. But in the age of supply chain issues powered by the Ukraine war & overall rise in cost of living, being cost competitive is a ‘mission impossible’. Hence, customers love a brand when they select the brand within purchasing criteria while checking with others based on price, delivery available, options etc. So they check out the websites, maybe click on the emails that they get for coupon codes and if the price is right make that purchase.
How to identify these customers?
It all starts with the customer – brand relationship that we had discussed in the previous post here and below is the visual representation for the same.
To understand the different behaviors represented above and define the appropriate approach, lets assume purchase behavior of 4 customers over a time T as represented below.
For ‘Non – Contractual’ Customer Relationships
As the definition gives away, when customer relationships that are not bound by any contract or subscription purchase transactions can happen at any time. Based on Image 2, the following inferences can be made about the behavior of the 4 different customers –
- Customers A, B & C made their first purchase at the same time. It could be that they received a coupon or invite or interest from a mutual friend that motivated the first purchase. There were no further purchases from A thereafter which means they do not need the product or have found better options somewhere or are a light shopper
- Both Customers B & C purchase at regular intervals. But Customer C has been inactive for a long time. We can expect C to comeback for some seasonal purchases same time next year or month, while there’s a lot of purchase expected from B at such regular intervals
- Customer B’s purchase behavior makes it the perfect profile for sending subscription related messaging
- Customers B & D last purchases happened at the same time, so they both have an equally high likelihood to make the purchase at time T. But with lower number of purchases from D, we can expect fewer or no purchases
Hence, the questions that business leaders would ask and we need to solve are :
- Based on the customer histories, which customers are ‘alive’ or ‘active’ at time T ?
- What is the next purchase expected from them in the future?
When Customer’s show continuous behavior
For Customers A & D that show ‘buy till you die’ behavior, a Pareto/NBD model can predict next transactions. They have 2 phases of the relationship – ‘alive’ for an unknown period of time and then permanently inactive. When “alive,” the NBD model represents the customer’s purchasing behavior. To implement this model parameters the only customer-level information required are recency & frequency.
The ‘buy till you die’ is not the only framework to capture the aggregated random and slowing down behavior, other models are quite difficult to implement. Additionally no one has yet derived expressions for definitions such as P(alive) and conditional expectations.
When Customer’s show discrete behavior
Customers like B&C who buy at regular intervals are often assumed to be seasonal purchasers who buy regularly at a certain time of the year/month. Assuming that the probability of purchase in one period is intendent of whether or not a purchase was made in the preceding period, a Bernoulli Purchasing process can be assumed.
Using the ‘buy till you die ‘ framework, the Beta-Geometric/BetaBinomial (BG/BB) model with a similar logic to the Pareto/NBD model, has been shown to be very good at modeling discrete-time transactions.
For ‘Contractual’ Customer Relationships
Assuming Customers B & C are making 5 transactions based on an annual contract for free deliveries and reduced rates (yes like Amazon Prime !) the business would ask the following questions –
- <li>Which customers have the highest rate of churning?</li><li>How long would each customer stay with the brand based on past behavior?</li>
The problem of modelling customer retention has received much attention from the areas of marketing, applied statistics, and data mining which have led to the development of a number of models that attempt to either explain or predict churn.
For any segment of customers, retention rates have been observed to increase with time. Such as you have a less propensity to cancel your Netflix subscription if you have already been a customer since the past several years.
Across continuous & discrete scenarios the GBM model has been found to predict customer performance very well out-of-sample and the parameters are useful for generating granular insights for individual customers. The key advantage being that it accommodates week-to-week fluctuations as well as trends in usage rate over time.