Case 1
Client: One of the largest dating / matrimony portals in the world
Opportunity: Optimize customer targeting by propensity to have paid profiles
Solution:
- Analyze key variable and compute the importance of variables
- Develop a seriousness score by creating a compound variable based on value of profile components, fields filled, nationality and customers outside their community / city zone
- Determine a conversion rate by determining regression with seriousness score by nationalities
- Train the model on 50% data, millions of record and test on remaining
- Implement the data for active targeting of customers with high propensity to pay
Result:
- Improved response rate by 11X based on the algorithm for propensity to pay
- Improved online marketing ROI by 6X
Case 2:
Client: One of the largest dating / matrimony portals in the world
Opportunity: Improve search and “view similar profiles” algorithm for better hit rate, stickiness leading to revenue
Solution:
- Identify behavioral variables such as contact initiated, contact accepted, login frequency and length of the profile instead of self-declared profile details
- Create a network of individuals by using past viewing history or action taken and degrees of separation thereof
- Calculate similarity in tastes by creating a factor which depends on overlap of network of two individuals
- Devise a score of likelihood to contact by factoring in days since contact in the network. Smooth the formula using an inverse exponential factor
- Implement the search algorithm and “view similar profile” algorithm on the website with millions of records
Result
- Improved contact rate by 7X based on the new algorithm of search and “view similar profiles”
- Improved online marketing ROI by 4X