Customer Churn: Is There a Method to the Madness?
When it comes to customer loyalty, churn and recurrence, traditional businesses usually have a basic cause-and-effect approach. This is derived from established consumer behaviour principles. Hence, they have a relatively predictable model for estimating customer lifetime value (CLV). However, the story is different for new age businesses, such as e-commerce (e.g. Flipkart, Amazon), mobile media streaming (e.g. HotStar, NetFlix), search and discovery platforms (e.g. Practo, Zomato), and service aggregators (e.g. Ola, Uber). Standard churn prediction models and CLV analysis are often found to be considerably inadequate here, and even delusive.
In fact, it is not just businesses of the internet age. It is now understood that even industries like aviation and consumer electronics are now in the midst of a chaos in terms of unpredictability in customer loyalty, churn and recurrence. Recently, research-based advances have been made towards modelling the chaos, or bringing a method to the madness. This has been underscored by the introduction of a new customer characteristic – ‘Clumpiness’ – to the equation. We shall discuss the concept in the next section.
From RFM to RFMC – Evolving Customer Analytics
Before foraying into ‘Clumpiness’, let us revisit the conventional analytics framework for customer churn. Until recently, the RFM model (Recency, Frequency, and Monetary) was believed to model customer stickiness successfully across sectors and industries. Conceptually, this model ranks / categorizes customers by three key behavior characteristics, both independently and collectively:
- When was the last time (how recently) the customer purchased a product / service from a business? – Recency
- How often (how frequently) does the customer use products / services of the business? – Frequency
- What is the transaction worth (in money terms) of the customer’s purchases with the business? – Monetary
One of the most important things to note in this model is that, customer churn is perceived to affect customer loyalty directly, i.e. increase in churn is equivalent to drop in loyalty. However, with consumer behavior becoming more chaotic and event-based for modern businesses, this lead to misleading results.
Let us consider the following examples:
- An e-commerce website ClipKart runs a sale promotion campaign offering substantial discounts, which causes customers to make purchases. However, as soon as the promotion is withdrawn, there is a high churn? But, does this churn mean that loyalty is lost?
- During Christmas and New Year holidays, a large number of customers book flights on Splice Airways – both domestic and international – for vacations. However, the bookings drop significantly when holiday season is over. Does this really count as churn for Splice Airways?
- During India-Pakistan cricket test match, a lot of new subscriptions are registered on DotStar – a mobile media content provider. Post the match, majority of the subscriptions are discontinued. Is this actually a customer loyalty issue?
Two important aspects emerge from the above examples. Firstly, unlike conventional churns, the churns in the above situations were driven by a trigger in the environment – i.e. a discount campaign, a holiday, an exciting match, and so on. This makes the churn appear less chaotic when viewed in context of the trigger.
Secondly, the churns in the above cases do not translate into customer loyalty immediately / directly. That is, even if a subscriber discontinues DotStar subscription once the current cricket match is over, s/he is likely to re-subscribe during the next match (depending on his experience of using DotStar). Similarly, the customers who booked flights on Splice Airways might book again during the next holiday season, and the next discount campaign by ClipKart might bring back its previous customers.
Now, if the above trigger-based ‘hot’ and ‘cold’ cycles in customer behavior can be factored into the standard RFM model in some way, the aberrations in CLV analysis can be corrected. This way, better marketing plans can be strategized around managing such event-based triggers.
This brings us to the concept of ‘Clumpiness’. A working definition for ‘Clumpiness’ can be the degree to which a customer does not conform to equal spacing in usage behavior of a company’s products and services. That is, the more event-based the customer-behavior is, the higher the clumpiness.
MECBOT: Modelling Clumpiness to Make Sense of Chaotic Consumer Behaviour
At FORMCEPT, our flagship product MECBOT equips new age businesses with tools to model ‘Clumpiness’ and marry it to traditional customer analytics.
The above image (Reference: Predicting Customer Value Using Clumpiness: From RFM to RFMC, Yao Zhang, Eric T. Bradlow, Dylan S. Small, Retrieved on June 30, 2017) shows transaction histories four customers A,B,C,D over a period time. A’s transactions are equally spaced during this period, whereas B,C, and D exhibit ‘Clumpiness’, although each of different extent and at different points of time. On conducting in-depth analysis, it may be found that each of B,C, and D had their purchases triggered by a different event. From these transaction histories, MECBOT can not only score and categorize, but also provide insights on which customers are more likely to become loyal with the right marketing strategy.
The image below shows how MECBOT calculates RFMC score of customers in a step-by-step manner:
Measuring Clumpiness – Unlike standard RFM, ‘Clumpiness’ is not about a data-point, but rather about data patterns over a period of time. ‘Clumpiness’ is a behaviour pattern that can be detected through time-series analysis (MECBOT uses AutoRegressive Integrated Moving Average – arimA). It is then measured using scores that indicate the extent of ‘Clumpiness’. This score is then rationalized and integrated with RFM analysis to depict a holistic picture of customer behaviour. The scores are separately and collectively fed into other existing databases and dataflows of the business.
Identifying Triggers – MECBOT’s capabilities do not end there. We take businesses a step further by correlating triggers that initiate ‘Clumpiness’ among customers. We identify a pool of triggers from both structured and unstructured data using opinion mining, social media analytics and traditional analytics. These triggers are then narrowed down to the critical ones using various statistical approaches.
Managing Triggers – To help businesses act on these triggers quickly, MECBOT generates action feedback for strengthening loyalty, preventing churn, and maximizing CLV by optimizing ‘Clumpiness’. These are integrated with workflow of each business to generate relevant actionable insights seamlessly. MECBOT can listen to external data sources (such as online platforms, social media interactions, emails, reviews, etc.) to capture external triggers and also read internal triggers that are fed into it. This makes it easy for businesses to identify and manage triggers that lead to customer ‘Clumpiness’.
If you are grappling with the problem of ‘Clumpiness’ in your business, MECBOT can come to your rescue. To know more about what we do and how we can help you, visit www.formcept.com or write to us at email@example.com