Cross-sell & Up sell

Objective

Objective of the project was to identify prospective Consumers that can be targeted for cross selling various products. For this a scorecard was needed to be built calculating the propensity of the target Consumers to buy various products.

Background and Challenges

Our client, a leading financial services group, was experiencing an extraordinary growth mainly fueled by their strong market presence and efficient Consumer services. The client wanted to harness its huge Consumer base using cross-sell strategies that are smartly designed. As this problem involved techniques in predictive modeling, the challenge was to identify :

• who should be targeted
• with what product(s)
• and at what price

The reason this business problem became special was because the data captured by the client consisted mainly of Consumer coordinates. Other valuable information that we use in predictive modeling to segment our Consumers was found missing. Apart from this the available data fields for different products were not consistent. Henceforth, to solve this business problem we had to rely on creating surrogate data for all products that should not only facilitate a comparative analysis of the Consumers of different products but also building robust models that bring excellent result. Besides developing the ‘smart cross-sell strategy’, we also aimed to provide a comprehensive solution that could help the client fix the gap in the data management practice and identify important information required for future capturing.

Our Approach

As the data provided consisted of mainly Consumer co-ordinates, it became more important to study its quality. We observed that the portfolio consisted of phone numbers that didn’t belong to the respective Consumers as it turned out that the agents were providing their own numbers to shield their Consumers.

For cross-sell purposes we need to devise techniques where two profiles of Consumers in different products can be compared. In our case we solved this problem in a very unique way. We started with small in-house data on which appreciable effort in cross-selling was performed in the recent past. Using that information we developed a model and identified the important factors that determine buying tendencies of Consumers in the Indian market.

Then, in our actual data we produced these fields by creating surrogates using a robust logic. New data was created (e.g. income range for vehicle owners using information on vehicle-make and vehicle-age) on which we used statistical techniques (e.g. correlation between two fields), survey results (e.g. percentage of income of Consumers who invest in insurance), micro-economics factors (e.g. purchasing power of people depending on demographics) and the business logic to create these surrogate fields. The new information thus available became the powerhouse to develop models whose output was a comparative score (normalized between 0 and 100) such that higher the score of a Consumer for a given product higher is the propensity of cross selling of that particular product to the Consumer. For each product individual scorecards were developed. These scores then were aligned with the profitability to decide on the strategies for the cross-selling.

In the above mention approach we tried to identify the potential Consumers by comparing their profiles from that of existing Consumers. The second approach was to group the Consumer on the basis of their profiles to identify if they have any traits in common. For these grouping, clustering techniques were used. The characteristics of the clusters were assigned based on the majority of its members.