180 likes | 303 Views
Maximize Profit(A)=. i = 1, …, M customers j = 1, …, N offers A = (a ij ) is a matrix where a ij = 1, if offer j is to be targeted to prospect i and 0 otherwise. A ij *( profit,cost,score ) ij. Market Optimization Phase VI. May 2012. Optimization & Market Expansion.
E N D
Maximize Profit(A)= i = 1, …, M customers j = 1, …, N offers A = (aij) is a matrix where aij = 1, if offer j is to be targeted to prospect i and 0 otherwise Aij*(profit,cost,score)ij Market OptimizationPhase VI May 2012
Optimization & Market Expansion • Optimize current screen criteria to local market conditions • Include all zip codes in current markets • Integrated Data Warehouse “Buckets” across all sources - ITA, PA, Trigger, Vertical, TBD • Optimizing target selections via individual screens • Relevant Messages • Model Scores • Performance Scores by Branch, Channel, Branch, Proprietary Targeting Cells
Extreme Situation: Too tight of Screens and 91.% of Homeowners were screened out of Target base
INCOME TU has only individual level of Income, this is aggregated income per household Even though we aggregated income per household, TU income is lower than Census Info. Prospects with less than TU Defined $30K individual Income, eliminated over 15% of the market opportunity
HOME EQUITY New Intellidyn Home Value – Maximum of TU home value, Mortgage 1 balance or Mortgage 2 balance About 5% homeowners have a loan which has higher mortgage balance than home value 01 - 02
Scored individual records from multiple master files Selects based on deciles from multiple buckets All phone numbers Verified/corrected for Telemarketing Pre-determined mix of ITA/PA/Vertical Same Select based on model score from one aggregate bucket Solicitations by telemarketing and/or email based on model score and optimization No pre-determined mix of sources. Optimization chooses records most likely to respond and convert Optimization Vs. Current Production Current Production Intelli-Optimization
Maximize Profit(A)= i = 1, …, M customers j = 1, …, N offers A = (aij) is a matrix where aij = 1, if offer j is to be targeted to prospect i and 0 otherwise Aij*(profit,cost,score)ij Mathematical Function for Optimization The multidimensional optimization technique requires the exact mathematical solution to the maximization or minimization of multi variable functions (i.e. business goal of a campaign such as profit, budget or sales). The final objective of the analysis is to exactly determine which specific offers (aij) to target to each individual prospect simultaneously optimizing business goals while satisfying business rules/constraints and local market conditions.
Required Input for Market Optimization -Business Goals (Maximum Profit, Minimum Budget, Maximum Number of Sales) -Dimensions (Offers by channels, etc. ): Campaign typically consists of several offers for targeting to a large set of prospects*. Each channel is associated with the following attributes necessary to calculate the business goal to be optimized: Response Model score for calculating probability of a prospect to respond to the offer Conversion Model score for calculating probability of a prospect to be funded. Profitability model score for calculating the profitability of a prospect for the offer Eligibility condition for determining eligible prospects for the offer Economics such as the delivery costs and selling costs of the offer -Constraints: Campaigns have several associated constraints such as budget limitations, minimum profit requirements and maximum number of offers that can be targeted to a prospect. * Prospects for this example were a 10% sample of three buckets, ITA, PA, Vertical
How Market Optimization Works • Calculates Net Present Value (NPV), expected profit, number of sales, etc. across multiple campaign parameters • Applies Constraints to determine eligibility of each prospect to each campaign (i.e.need a phone number for telemarketing) • Applies Business Rules (I.e. 1 offer per Household, Budget constraints, etc.) to determine the optimal offer for each prospect
Performance*, Score & Cost Metrics *All performance metrics are based on MIS report as of 4/4/2012
Score Function Business Goal: Find most appropriate objective function Probability to respond =( Response model score + Historical Performance by channel )/2 Probability to Convert =( Conversion model score + Historical Performance by channel )/2 Cost = Source cost + Cost by channel Budget vs. Branch Budget Exercise: Optimize Budget allocation Probability to respond = Response model score Probability to Convert =( Conversion model score + Historical Performance by branch )/2 Cost = Source cost + average Cost for all channel
Overall Scenario Results Scenario 2 has lowest CPL but no other higher results Scenario 3 has most responders but Resp. Rate and Fund Rates are low Scenario 4 has highest fund rate but also a higher CPL Red Highlighted numbers represent max/min for each column
Market Optimization Next Steps • Use to Determine or test market Screens • Development of channel specific models will increase performance of Optimization Software