HSBC is a global banking and financial services organization, with 7,500 offices in over 80 countries. The bank’s businesses range from personal and private banking to corporate and investment banking.
HSBC was running Cross Sell campaigns to improve product holding ratio. HSBC’s asset product (Personal Loan) was targeted at a liability base. However the campaigns were hampered by chronically poor response rates.
Xerago identified that to boost the response rate, the base of the Personal Loan customers would have to be analyzed to identify the types of customers within the base who were most likely to respond to a cross sell campaign.
Additionally, Xerago also identified that
- Low response rates hampered the building of a stable statistical model like logistic model, that could predict the likelihood of campaign response
- Building a representative customer base for the purpose of data mining was difficult owing to the skew in the data
To address the above concerns Xerago proposed
- To build a new response model to help score customers for new campaigns
- To use the existing customer data of liability customers to match and offer loan combinations on the scored base
The data study was limited to HSBC’s liability customer base. Responders were defined as those who had taken a personal loan. Customer data from six prior campaigns was selected for the model building process. Xerago was also able to identify those accounts that overlapped with other campaigns and de- duped the data to eliminate duplicate accounts. The following was done as part of this assignment
For each month that the campaign was run, the corresponding customer behavior for the past 6 months was observed closely. Xerago utilized specific tables within the client’s database to extract the relevant data. Using specific variables for each of the table, Xerago was able to summarize data at a customer level and identified a total base of 101696 customers with 778 respondents.
Xerago worked with HSBC to further prune the data. Customers with no long term affiliations with the banks (savings accounts) were excluded from the data. The final customer base stood at 101161 records with 775 respondents; a response rate of 0.77%
Bootstrap and Sampling:
The very low response rates required bootstrapping the customer base at different proportions. SAS’s Proc SurveySelect option allowed the sampling at 10%, 20%, 30% and 40% with the number of respondents’ maintained constant (775). 30% of the sample was found to be an effective representation of the population and gave rise to the altered distribution.
- Non Responders 30116
- Responders 775
- Total Base 30891
- Response Rate 2.51%
The above numbers were used to create the model.
A multitude of variables reflecting the Average Monthly Balances (AMB), Money Inflow and other variables like ATM usage, withdrawals etc. were considered as the model variables. With repeated iterations and detailed studies a list of the final model input variables was derived.
- Average AMB
- Holding Period
- Total number of ATM cash withdrawal
Using this, Xerago was able to construct a response model that was able to identify specific customers within the customer base with high propensity towards responding to a cross sell campaign. And in doing so Xerago was able to significantly increase the response rates of the cross sell campaign.
The model helped achieve a lift of 65% as validated by a sample campaign which was run on the identified base.
During the course of the analysis, Xerago additionally identified that different customers were receptive towards different interest rates. Xerago advised HSBC on how to efficiently utilize existing loan details to offer apt combinations across the scored base.