Analytical model templates: Think one-to-many!
By : Dr. Rahul Vyas
AGM- Analytics Center of Excellence
If we go deeper and analyse many business problems, we can observe them sharing common traits across the industries like the problem of customer churn or margin optimisation. It is beneficial for any analytics practice to pursue opportunities in solving such problems. The effort spent on developing, validating, and optimising analytical models for live deployment is huge, hence it makes sense if we can think one-to-many and have packaged model templates with applicability across the industries.
The horizontal problem of too many choices
In this article, we will discuss one such problem and see the design considerations that can be kept in mind while packaging the analytics capabilities to solve the problem. Enterprises especially in industries like Telecom, BFSI and Retail compete aggressively to acquire and retain customers. This leads to customers always having multiple options to choose from at any point of time like multiple malls or multiple websites to shop or multiple bank accounts or use of multiple SIMs.
This has led to a tectonic shift from mere customer acquisition to customer retention and loyalty enhancement. Let us see how enterprises have been plagued with this common horizontal problem and how analytics practice can approach to building ‘packaged intelligence’ to help business users address this in different industries.
Multi-SIM in Telecom
Usage of multiple SIMs is quite common now a days especially in the emerging markets, where prepaid plans are the norm and subscribers are most price-sensitive. Lately, with the introduction of 4G services and Telcos competing to provide attractive data plans at the lowest prices, multiple SIM ownership has increased sharply. Multi-SIM owners typically have mobile devices which have dual-SIM standby feature so as to utilise the benefit of switching between the telecom carriers. Usage of multiple SIMs may have an adverse effect on the revenues and the churn rate of a Telco. Therefore in a highly competitive market scenario, it is crucial for a Telco to refrain its customers from switching to other telecom service providers.
Multiple Accounts in Banking
Banks and financial institutions these days look at providing a wholesome customer experience so as to sustain competitive advantage by increasing customer loyalty, targeting a greater return on promotional campaigns, reducing costs, and improving operational efficiencies. To improve the experience in line with customer expectations, banks must obtain a better understanding of their customers’ behaviors and motivations. Banking customers are increasingly demanding and insist on being identified with unique needs which are addressed personally. Predictive analytics can provide valuable and actionable insights over banks’ data of users holding multiple accounts so that they can encourage customers to use their account and maximise revenue.
Multiple retail avenues to shop around
Again choices are plenty for customers when it comes to retail sites and malls to shop around. Retailers are looking for ways to win over consumers, whether it’s in-store, online, at a kiosk or on the phone. With the right marketing tools and insights, retailers can make customers feel special. By analysing past click-through behavior, preferences, and history in real-time, retailers can provide a better customer experience by right price adjustments, recommendations and promotions.
One methodology, multiple use cases
The practice applies a distinctive evaluation technique to identify multiple buying alternatives in any industry, like having multiple current accounts in bank or buying from multiple retail malls or buying multiple SIMs. The customer might use the alternatives to get pricing or service quality benefits. Such customer behavior can be identified at an early stage so as to provide customized offers to keep them engaged and increase business with you. It will also help to identify loyal and not-so-loyal customers.
A market research is conducted on a sample set of users to verify and categorise primary and secondary users. A list of questions is prepared to understand the customer inclination towards the primary and secondary choices. This information along with specific usage based KPIs are used to build the predictive model. Generally KPIs reflecting demographic, usage behavior and footfall patterns are used to analyse the customer behavior on choosing and using products from available alternatives. The RFM (regency, frequency and monetary value) model can be used to analyse the usage behavior.
For example in telecom scenario, a survey is conducted on sample base to determine whether the user is using a particular service provider’s SIM as a primary or a secondary SIM. Then that information is attached with specific usage based KPIs like Off-net Minutes of Use (MOU), On-net MOU, average revenue per user (ARPU), handset type etc. to build a model to predict the multi-SIM behavior of larger base.
The market research questionnaire and the KPIs are all considered to be a part of the primary data. This sample dataset is further segregated into test and train dataset. Train dataset is used to train the classification models. The trained model is then tested on the test dataset and accuracy of all the models is compared to find out the best model for prediction.
One model template may fit all
It is highly imperative for analytics practice to understand these business problems with common traits across different industries in depth so as to develop a methodology for identifying and building the best possible model that will help enterprises across different industries to address these problems. Once Analytics practice adopts this approach, they can scale up on delivery of analytics output in an efficient manner. However enterprises will also need appropriate technology platform to implement this collaborative analytics approach, where these model templates can be deployed, tested and optimised for their specific use case.