Business Analyst and Data Science Professional

1. Customer Modeling Challenge 2014 Stage 1 Submission
Submitted by
Name: Jatinder Bedi
Ideatory username: jmbedi
jmbedi@gmail.com, +919717394374

2. Objective
To identify the relevant parameters from the CRM data, which are significant in reference to the propensity of customer
The objective of the study is to predict propensity of customers buying add-ons or upgrades by modeling CRM data.
Tracking effectiveness of the campaign and changes required to improve the conversion
To evaluate that whether CRM data can be good input for BIG Data Analytics leveraging the methodology of the social listening
Try to evaluate the campaign effectiveness by analyzing panel dataset.

3. Conversion optimization strategy
Conversion optimization strategy for increasing use of debit card among the females of 18-24 age group can be listed in steps like:
1. Customer Segmentation into small groups and addressing individual customers based on actual behaviors – instead of hard-coding any pre-conceived notions or assumptions of what makes customers similar to one another, and instead of only looking at aggregated/averaged data which hides important facts about individual customers
2. Tracking customers and how they switch segments (i.e., dynamic segmentation), including customer lifecycle context and cohort analysis – instead of just determining in what segments customers are now without regard for how they arrived there
3. Accurately estimating the future move of customers (e.g., convert, churn, spend more, spend less) using predictive customer behavior modeling techniques – instead of just looking in the rear- view mirror of historical data
4. Using advanced calculations to determine the Life Time Value (LTV) of every customer and basing decisions on it – instead of looking only at the short-term revenue that a customer may bring the company
5. Knowing, based on objective metrics, exactly what marketing steps to do now, for each customer, in order to maximize the long-term value of every customer – instead of trying to figure out what to do based on a dashboard or pile of reports.
The Above steps will lead to the animated view of the customer which will help us to make accurate customer behavior predictions.

4. Hypothesis
Hypothesis which I would like to test are listed below:
Initial Hypothesis:
H0: The young female(18-24 age) customer will not be applying from face-book page for a debit card.
Alternate Hypothesis:
Ha: The young female(18-24 age) customer will be applying from face-book page for a debit card.
Initial hypothesis:
H0: The young female(18-24 age) customer will not be applying from face-book page for a DBS debit card encapsulated with entertainment offers( like movie tickets, utility bill waiver, college fee scholarship etc)
Alternate Hypothesis:
Ha: The young female(18-24 age) customer will be applying from face-book page for a DBS debit card encapsulated with entertainment offers( like movie tickets, utility bill waiver, college fee scholarship etc) .

5. Testing
The Hypothesis will be tested via analytics of multivariate experiments which includes:
1.A/B and MVT testing
2.Full factorial and fractional factorial experimental designs
3.Then analyzing the MVT experimental data
The above methods are selected by me due to following reasons:
1.Granularity of results: MVT finds impact at factor level
2.Interaction effect between factors can be done by MVT
3.A/B tests best for dramatic design
To track the campaign effectiveness:
1.Identify performance metrics
2.Optimize Ad publishing strategy
3.Optimize campaign strategy
4.Calculate KPI effect

6. Testing-continued
To achieve the objective : data will be required from various sources:
•CRM data
•Time-Series data
•Survey Data(including sentiments)
•Panel dataset from search engine advertisements on a cross-section of keywords.
Keyword level (Paid Search):
– Date and ad id
– number of impressions
– number of clicks
– click-through rate
– cost per click (CPC)
– rank of the keyword
– number of conversions
– conversion rate
– revenues from a conversion
– ad quality score
– landing page quality score
Product Level : Quantity , Category, Price , Popularity

7. Analytics Approach/Techniques
•Correlation: The degree to which two variables have a tendency to vary together
•Cross Tabs: Organizes data into a table, showing the frequency at which “combinations” occur in the data.
•Social Listening aspect: Text mining (Sentiment analysis) of the customer posts on FB/twitter about DBS bank in the targeted group.
•Dividing into groups via: Randomization and propensity scores
•Propensity scores: Conditional probability of being treated given individual background characteristics. It ranges between ( 0 to 1.0) estimated using logistic regression model.
•Matching using Propensity scores.
•Propensity scores control for covariates in regression analysis
•Modeling via LOGIT Regression: is a type of regression analysis used for predicting the outcome of binary variables . The LOGIT function bounds the limits of the outcome variable to 0 and 1, thus overcoming the limitations of OLS . LOGIT regressions can employed in marketing applications such as prediction of a customer's propensity to buy or not.
•Recommendations or insights from regression results will help us to make the future campaigns more effective

8. Field Experiment Design
From the CRM data of the female between age of 18 to 24, randomly select the treatment group and randomly select the control group too. We will show the new face-book DBS advertisement to the treatment group We will show the normal DBS bank advertisement to the control group

9. Plausible Result
Expected results from analysis:
•Identification of relevant parameters from the CRM data, which are significant in reference to the propensity of young female customers
•With the help of propensity score will be able to segment the young female customers.
•With the help of LOGIT results coefficients , we should be able to predict the change in log odds of young female customer applying for debit card upon one unit increase in the independent variable.
•From the analysis of the CRM data, we can either reject the H0 or we can leave it as it is.
•From the panel dataset analysis we get the factors with which CTR(Click Through Rate) is proportional, also their magnitude & direction. This will help us to come up with the changes required in the independent variables to increase the CTR.
•With the help of social listening : we can get customer insights, new product features and new offers. which customer wants from debit card.
•With the help of the results we can predict the offers we can have debit cards to increase conversion
Changes via result:
•Will include the changes required to the face-book advertisement as per the insights discovered from the data.
•Will do the changes as per the sensitivity and specificity analysis results.

10. Word cloud of customer posts

11. Mocked up FB advertisement to improve user experience post results
Targeting the young female to apply for a student account debit card and enroll for a chance to win first year university tuition fees.

12. Social Media Challenges
•Measuring the effectiveness of a social media campaign on product purchase , the campaign has its effect on persuading and influencing people to purchase
•But people may adopt and purchase a product because they see that their friends have also purchased the item, regardless of the campaign’s effects
•Simultaneity here results an overestimation of the effect of the social media campaign
•The modeler may have ready measures of length of ad campaign, click through rates, spending on digital marketing, and other measures of digital buzz
•Such external conditions can create OVB(Omitted Variable bias), affecting results drawn from the analysis
•Each entity has its own individual-level ,characteristics that may influence the dependent variable
•These intrinsic sources of variability across units of analysis can bias the regression coefficients
•In econometric analyses, we term these sources of variability as drivers of “unobserved heterogeneity”
•Such unobserved heterogeneity can be understood as a form of omitted variable bias (OVB)

13. Appendix- Debit Card usage Dynamics

14. Debit Card Usage Patterns

15. Sources Referred
•Google.com
•“Consumers’ Use of Debit Cards: Patterns, Preferences, & Price Response” by Ron Borzekowski, Elizabeth K. Kiser & Shaista Ahmed