Notes Version: The Paradigm From Sales To Profits Using Optimization Analytics
0 likes206 views
CVS Caremark utilized lift modeling and constraint-based optimization to improve their POS coupon program. They developed predictive models to identify customers most likely to purchase incrementally in response to offers. Testing found their initial models did not improve results, but refining the models using more data and a combined analytic and strategic approach significantly increased incremental sales and profit per customer. Lessons included focusing on the right success metrics, challenges of lift modeling, and the need for both analytic techniques and business knowledge.
Notes Version: The Paradigm From Sales To Profits Using Optimization Analytics
- 1. 9/30/2011 What to Expect from This Session Explore the fundamentals and benefits of Lift Modeling and Constraint-Based Optimization Analytics Learn how CVS Caremark has utilized their POS channel to deliver relevant offers and improve ROI - Behind-the-scenes analytic engine that drives differentiated offers to unique types of customers - Program evolution - Lessons learned Introductions 1
- 2. 9/30/2011 Who We Are Consulting firm specializing in information-based marketing and customer intelligence Founded in 1999, with headquarters in Burlington, MA Business problem solvers, skilled in analytics Broad, practical experience across multiple industries Outcomes driven - strive to deliver work that produces significant bottom line impact What We Do Customer Intelligence and Reporting Support Business Decision Multi-media Making / Processes Predictive Marketing Analytics and Effectiveness Scoring Increase Business Growth and Profitability Campaign Customer Planning and Acquisition and Measurement Development Optimize Marketing Loyalty Program Spending Strategy CVS What does CVS stand for? a. Consumer Value Stores b. Caroline, Vanessa, Sophie c. Convenience, Value, Service d. Computer Vision Syndrome 2
- 3. 9/30/2011 CVS Nation’s largest retail pharmacy chain Recently opened 7,300th store CVS fills more than 1 out of every 7 prescriptions in America ExtraCare is the largest and most successful retail loyalty program in the country ExtraCare Loyalty Program (brief video) Do you have your ExtraCare Card? 68 MM active ExtraCare cardholders 70% of CVS transactions are on the card ExtraCare customers visit the store every month, which is twice as often as the typical non-ExtraCare shopper Every second, 32 ExtraCare cards are scanned in the US The very next day, these transactions can be transformed into customer insights 3
- 4. 9/30/2011 Lift Modeling and Optimization Incremental Sales Potential “Best” customers are not always the same ones who drive campaign profitability Best Customers - Top Sales Promotion Sensitive, High Incremental Sales Limit or Exclude from New Population - Discounting Strategy Test Offer Levels Response Model Lift Model Best Offers Lift Modeling Focus directly on the consumer behavior that drives ROI Predicts incremental response rather than total response Model score provides a mechanism to target customer with the highest likelihood to deliver incremental sales The challenge: incremental sales is an aggregate measure, while modeling and targeting must be done at the individual level 4
- 5. 9/30/2011 Offer Optimization Engine Objective: Determine the optimal offer assignment for each customer given certain goals and constraints Automated approach to offer assignment; allows quick evaluation of numerous scenarios Optimization targets: maximum lift sales, program revenue, units, ROI, margin impact; minimum incentive cost Optimization Constraints: budget, incentive cost, ROI hurdle, total sales Flexibility is a key benefit Optimization in Action How do you do it? Start by asking: are you leveraging everything you know about your customer? Shopping Behavior Promotion Category Breadth History / Channel Preference Basket Profiles Digital Behavior Tenure Payment Type Employees / Demographics Non-Employees 5
- 6. 9/30/2011 POS Marketing at CVS POS Coupons - Overview Sophisticated targeting engine drives delivery of POS Coupons Printed at Bottom of Cash Register Receipt Primary Objective: Drive return visits Customer data (what they buy, who they are) drives offer assignment Offers: Product-Specific or “Open-Ended” Open-Ended Offers - Discount on any non-Rx item in the store - Automated trigger program - Redeemable on next visit - Broader offers drive most sales lift and ROI - Largest CVS-funded incentive program CVS Approach – Trigger Offers Customized and Dynamic Targeted Offers Illustrative Customer 1: Customer 2: Customer 3: Spends $12.50 Spends $22 per Spend $15 per per trip, visits trip and, visits 4x trip and, visits 5x 3x Qtr Month Qtr ExtraCare/coupon John Smith Offer Configuration and Cadence determined quarterly by mining millions of data Points: Save $4 on any $20 purchase. (UP TO $4.00 VALUE) Category Past Profit Rx Shopped Demos Offers Level Promos Freq EXPIRES 11/4/2011 ExtraCare® Card #*******8047 *EXTRACARE CARD MUST BE PRESENTED TO GET THESE Customer 2: Customer 1 Customer 3: SAVINGS. EXCLUDES TOBACCO, ALCOHOL, AND PRESCRIPTIONS. NO Matched up to CASH BACK. TAX CHARGED ON PRE-COUPON PRICE WHERE REQUIRED. Matched up to Matched up to LIMIT ONE PER CUSTOMER. $10 off $50 3363805547 $3 off $15 offer $4 off $20 offer offer 6
- 7. 9/30/2011 Evolution of POS Program Ongoing innovation and refinement has ensured lasting program success Step 5 Step 4 Optimize Optimization Step 3 Innovate System Step 2 Refine Lift Modeling Response, Step 1 Target and Modeling, Differentiation Execute Build Fact R, F, M Base Test and Learn We were here This could not be achieved without a commitment to test and learn The Catalyst: Declining ROI Response model targeting was beginning to be unprofitable – need for investigating drivers and potential solutions Campaign ROI 20% 15% 10% 5% 0% Campaign Incremental Discount / Net Net Sales Sales (Lift) Coupons Incremental Incremental -5% Sales Margin -10% -15% Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Key Question: Who drives incremental sales? 2009 2010 Modeling Approach Explore a wide range of variables to capture key customer behaviors - Promotion sensitivity and history, brand preferences, loyalty, demos Create multiple predictive models for each step in the coupon redemption process Print Redeem Spend visit, purchase visit, purchase basket size Estimate a wide range of success metrics Optimize offer assignment, unique to each customer, given predicted behavior in response to each offer scenario 7
- 8. 9/30/2011 Program Structure The final solution involves multiple models that capture the profit dynamics Econometric Business Data Inputs Optimization Modeling Metrics Shopping Sales Behavior / No Offer Loyalty Price & Promotional Sales Lift $3 off $10 Sensitivity Predictive Coupon Usage Models Cost $4 off $20 (14) Category Margin $5 off $25 Patterns Lift Qualifier (1) Redemption (8) Customer Sales (5) Demographics ROI $10 off $50 Recommended Target Changes Significant changes to targeting and offer assignment were recommended Old Target Drop Keep Add New Target Higher incremental sales Don’t Target Stronger ROI The Test A test was implemented to compare the new approach to the old Eligible Customers New Strategy - Lift Existing Strategy - Modeling and Response Model Based Optimization Based Control Group Control Group 8
- 9. 9/30/2011 Test Results It didn’t work! Incremental Sales and Profit Per Customer Incremental Sales Incremental Profit Existing Model New Model $3 and $10 coupon worked well $4 coupon drove negative results Identified disconnect in success metrics If At First You Don’t Succeed… Granted another opportunity to “split” the eligible population in a head-to-head test Refined and implemented models using more recent data Tested alternative approaches - Model Based Targeting and Offer Assignment - “Strategy-Based” approach – art plus science Art + Science Wins! Incremental Sales and Profit Per Customer Incremental Sales Incremental Profit Existing Model New - Model Based New - Strategy Based ROI 4% 15% 16% 9
- 10. 9/30/2011 Lessons Learned Lift Modeling is not straight forward – be open to testing - Most challenging: predicting incremental sales dollars and ROI - There are several alternative modeling approaches Focus on the right metrics, and be consistent with corporate KPIs Optimization is likely to require human intervention – solid business understanding and good intuition In Conclusion Leverage as much data and customer insight as you can to maximize The next generation of marketing analytics is moving beyond traditional response modeling - Understand your success metrics and use analytic tools that focus directly on those metrics - Be aware of the challenges The best solution requires a balance of technique and business knowledge Getting Started Set up randomized testing with appropriate treated / control quantities Capture historical campaign data - Customer characteristics at time of execution - Response results Ensure operational system infrastructure is in place Identify resource to develop solution 10
- 11. 9/30/2011 Thank You! Marcy Riordan Michael Parduhn ANALYTIC VISION. BUSINESS IMPACT. For more information contact: mriordan@iknowtion.com (781) 494-9989 11