Casestudy media-mix-modeling[1]
- 1. Case Study: Media Mix Modeling (1/4) (For Retail) A national retail chain in U.S. with more than Client 300 stores spread over 51 different Background geographical clusters Retailer witnessed store sales declining Business significantly over past few years despite Situation same/increased level of media spend allocation ■ To understand media spend and sales for improving its return on media at store level for Objectives geographic clusters ■ To optimize & reallocate the media spends 1
- 2. Case Study: Media Mix Modeling (2/4) Methodology: Modeling ■ Baseline factors data gathering, cleansing, transformations Prototype Modeling and model prep was done Identify appropriate analytical process and determine base ■ Baseline was developed for each geographic cluster using predictors of promotional sales lift. Identify data limitations that may multi-staged regression impact scope of analysis ■ Media mix models were developed at the store cluster level ■ Different regression models were attempted in arriving at a best fit model that captured the effect of media on sales Media Mix Models • Compute baseline sales ■ Additional models were tested, to incorporate other • Develop media mix models using possible effects: historical sales, media spend, and store information Model/Concept Tested Details Generalized Additive • Cumbersome interpretability Modeling • No explanatory variable specific co-efficients as output which was a pre-requisite for next Media Mix Optimization level of mathematical optimization Optimize sales lift due to media Multiplicative Modeling Overall model fit and predictive capability spend, for each market by were low for most clusters reallocating total media spend among media channels Non-Linear Constrained • No observed improvement as compared to Regression Modeling linear models, for significant # of clusters it was not a good fit • Processing of NLCR was too slow 2
- 3. Case Study: Media Mix Modeling (3/4) Methodology: Optimization ■ Objective of Optimization process was to re-allocate individual media spends with a view to maximize total sales lift* ■ Optimization was done for the Q4 2007 using Q4 2006 figures for all the 51 geographic clusters – Total ad spend was kept the same as that of Q4 of 2006 total ad spent – But, each individual media spend was constrained to different upper and lower bounds in two different scenarios ■ In one scenario, each individual media spend was constrained to + 5% change as compared to individual media figures based on Q4 2006 spends ■ In the other scenario the change was constrained to + 10% *Sales lift here means predicted sales lift 3
- 4. Case Study: Media Mix Modeling (4/4) The Final Outcome Media Reallocation Results for Fall 2006 Season Market Call Code # of Total Store Total Ad Promotion Sales Lift Promotion Sales Lift % Change in % Change Stores Sales Spend Predicted Optimized Promotion in Store ($ MM) ($ MM) ($ MM) ($ MM) Sales Sales Aberdeen, SD KABR 1 $5.7 $0.14 $2.80 $2.84 - $2.88 1.5 - 3.0 0.7 - 1.5 Northern Iowa KALO 4 $19.4 $0.53 $10.29 $10.38 - $10.47 0.9 - 1.8 0.5 - 1.0 Wausau, WI KAUW 3 $13.5 $0.39 $5.51 $5.53 - $5.55 0.3 - 0.8 0.1 - 0.3 Scotts Bluff, NE KBFF 1 $5.7 $0.13 $3.12 $3.13 - $3.14 0.2 - 0.6 0.1 - 0.3 Billings, MT KBIL 1 $6.2 $0.42 $2.87 $2.90 - $2.92 0.8 - 1.7 0.4 - 0.8 … …. … … … … Broadsheet… 25% … … 23% … … … 26% … …. … … … … … … … … … … ROP 5% 5% … …. … … … … … … 6% … … 7% … … 6% Radio 4% Rochester, MN KRST 4 $21.2 $0.63 $8.71 $8.99 - $9.27 3.2 - 6.5 1.3 - 2.7 Michigan City, IN KSBN 1 $7.1 $0.31 $3.99 TV $4.14 - 22% $4.30 3.9 - 23% 7.8 2.2 - 4.3 20% Springfield, IL KSPI 2 $14.4 $0.76 $7.80 $8.05 - $8.30 3.2 - 6.5 1.8 - 3.5 9% 8% St. Cloud, MN KSTC 4 $33.9 $0.95 $18.48 Catalog Insert $18.91 - $19.33 2.3 - 11% 4.6 1.3 - 2.5 Sioux City, IA KSUX 1 $8.7 $0.17 $3.93 $4.08 - $4.23 3.8 - 7.6 1.7 - 3.5 Traverse City, MI KTVC 1 $5.9 $0.22 $3.37 Catalog Mailer $3.42 - $3.47 33% 1.4 - 2.9 0.8 - 1.7 36% 31% TOTAL 132 $1,194 $41.9 $604 $617 - $630 2.2% - 4.3% 1% - 2% $18.5 MM $ 20 MM $21 MM Budget Budget Budget Our collaborative efforts allowed the Suggested Current Suggested Retailer to reduce spend by $2M and Allocation Allocation Allocation generated 2% in incremental sales 4