List Sourcing Optimization. Using Analytics to Own the Advantage

Transcription

1 Sourcing Optimization Using Analytics to Own the Advantage

2 For too long list owners and brokers have held the advantage in list acquisition Now it s your time to own the advantage! 2

3 Knowledge is Power In this session you will learn how to use analytics to gain critical knowledge necessary to: 1. Compare lists unbiasedly 2. Dictate to list owners / brokers list value 3. Eliminate cost in your list acquisition 3

4 Typical Situation High volume mailer Large number of list sources Merge process eliminates duplicates Uncertain just how many sources are really needed Opportunity Reduce paying for duplicate names Especially those that cost more! 4

5 Approach Your Existing Sourcing Optimal Sourcing There is a point at which additional lists provide no additional value. ROI N lists 5 N lists # of s

6 Source Mix Outcome Move from your current source mix to one that achieves the same or better performance at less cost. Multi-Buyer 14% Current Compiled 28% Response 12% Optimized Multi-Buyer 4% Compiled 20% Response 33% PDb 25% PDb 64% 6

7 Source Optimization Evolution Sourcing Optimization Big 3 Unbiased Response Attribution Analysis Level Optimization Individual Level Optimization 7

8 Unbiased Response Attribution Unbiased response attribution enables accurate list level performance analysis TRADTIONAL (CREDITED-ON) RESPONSE ATTRIBUTION Response is randomly assigned to a single list, typically the list that got paid. Remaining lists do not get the credit hence resulting in incomplete attribution UN-BIASED (APPEARED-ON) RESPONSE ATTRIBUTION Response is assigned to each of the lists on which the individual exists

9 Level Optimization Efficiency Index The final metric used to rank order lists The Efficiency Index is a normalized measure of a list s ROI controlling for differences in average mailing depth across lists. This allows an apples to apples comparison that is not impacted by average decile mailed. A B Average Decile Mailed 9

10 Level Optimization (cont.) Efficiency Index The final metric used to rank order lists Norm Resp Index / CPP Index Norm RR Index = Normalized RR for the list / RR Overall CPP Index = CPP for the list / CPP Overall o o Normalized RR = Appeared-on RR * Decile Adjustment Factor CPP = Total cost including list, package, creative / mail volume 10

11 Cum. Response Rate Index Level Optimization (cont.) Normalized RR Appeared-on RR * Mailing Decile Adjustment Factor A Avg Decile Mailed Index of Response Model Used in Selection B Avg Decile Mailed 5.0 Index of All s Avg Decile Mailable 4.5 Index of Response Model Decile Appeared-on RR A =.85% B =.80% All s Avg. Model Index Mailable = 102 Decile Adjustment Factor A: 102/109 =.94 B: 102/101 = A Normalized RR =.85% *.94 =.80% B Normalized RR =.80% * 1.01 =.88%

12 Level Optimization (cont.) Efficiency Index Norm Resp Index / CPP Index Overall Average Appeared on RR =.81% CPP = $.45 A B 2.5 Appeared-on RR =.85% Decile Adj Factor =.94 (102/109) Norm RR =.80% (.85% *.94) CPP = $.42 Norm RR Index = 99 (.80%/.81%) CPP Index = 93 ($.42/$.45) Efficiency Index = 106 (99/93) 5.0 Appeared-on RR =.80% Decile Adj Factor = 1.01 (102/101) Norm RR =.88% (.80% * 1.01) CPP = $.48 Norm RR Index = 109 (.88%/.81%) CPP Index = 107 ($.48/$.45) Efficiency Index = 102 (109/107) A Efficiency Index 106 B Efficiency Index

13 Level Optimization (cont.) Utilize Excel Premium Solver to run list level optimization most efficiently when using multiple constraints. 13

14 Individual Level Optimization Individual Efficiency Index The final metric used to assign an individual to a single list Individual Norm Resp Index / CPP Index The Individual Efficiency Index is different for the same individual on each list they appear on and involves a number of adjustments to get to an apples-to-apples comparison. Uses appeared-on list performance and expected individual and list performance based on the response model. Required Adjustments Needed to Compare: o average decile mailed (vs. average mailable) o Individual decile mailed (vs. average mailable) Joe A 25% 2.5 Joe B 50%

15 Individual Level Optimization (cont.) Individual Adjusted RR Normalized RR * Individual Decile Adjustment Factor Individual Performance Adjustment Individual Joe index A or B 110 A Individual Index at Avg Decile Mailed B Individual Index at Avg Decile Mailed Response Model Percentile Individual Model Percentile = 90 Resp Index = 110 Normalized RR A: =.80% B: =.88% Individual Decile Adjustment Factor A: 110/94 = 1.17 B: 110/89 = Joe s A Individual Adjusted RR =.80% * 1.17 =.94% Joe s B Individual Adjusted RR =.88% * 1.24 = 1.09%

16 Individual Level Optimization (cont.) Individual Efficiency Index Individual Norm Resp Index / CPP Overall Average Appeared on RR =.81% CPP = $.45 Joe A Joe B 25% 2.5 Appeared-on RR =.85% Decile Adj Factor =.94 (102/109) Norm RR =.80% Indiv Adj Factor = 1.17 (110/94) Indiv Adj RR =.94% (.80% *1.17) Indiv Norm RR Index = 1.16 (.94%/.81%) CPP = $.42 CPP Index =.93 ($.42/$.45) Individual Efficiency Index = 1.25 (1.16/.93) 50% 5.0 Appeared-on RR =.80% Decile Adj Factor = 1.01 (102/101) Norm RR =.88% Indiv Adj Factor = 1.24 (110/89) Indiv Adj RR = 1.09% (.88% *1.24) Indiv Norm RR Index = 1.35 (1.09%/.81%) CPP = $.48 CPP Index = 1.07 ($.48/$.45) Individual Optimization Score = 1.15 (1.35/1.07) A Individual Efficiency Index 1.25 B Individual Efficiency Index

17 Individual Level Optimization (cont.) Why is it better to acquire Joe from list B? B s normalized RR (.88%) is much better than A (.80%) but B s list efficiency index (102) is worse than A s (106) B s individual adjusted RR (1.09%) is much better than A s (.94%) and B s indiv efficiency Index (1.26) is just better than A s (125) Joe s expected ROI is slightly higher if purchased from B 17

18 Individual Level Optimization (cont.) Example Assignment Credited-on Actual vs. Credited-on Recommended A B C D Joe Wendy Susan Brian Alex 18 Actual Recommended

19 Individual Level Optimization (cont.) Rollup All individuals are deduplicated by assigning them to their best list then they are rank ordered by their individual efficiency index. Individual selections are then made in order of individual efficiency index until the desired objective is reached (cost, mail volume, responses etc.) Selected individuals are then summed by list considering total individuals on each list and list costs are calculated. 19

20 Individual Level Optimization (cont.) Cost Convergence The full optimization process is run repeatedly with dynamic cost estimation until cost convergence is reached. This is when Effective CPP % Change is less than After each optimization run the list CPP is recalculated based on the recommendation and compared to the CPP from the previous run. If the difference is too great then the optimization runs again and a new CPP is calculated and compared to the previous. This keeps repeating. Convergence typically occurs after about 17 iterations at which point the recommendation is final. Effective CPP = Total Cost / Net Mail Qty Effective CPP % Change = (Effective CPP current - Effective CPP previous) / Effective CPP previous 20

21 Individual Level Optimization (cont.) Individual Optimization Results acquisition quantity in most segments could be reduced substantially order quantity for segments A and B could be reduced by 60% without impacting mailing quantity and campaign performance Actual Sourcing Optimization/Recommendation Reduction% Segment # s Order Qty Mail Qty # s Order Qty Mail Qty # s Order Qty A 79 17,382,322 1,500, ,205,660 1,500,000 62% 64% B ,862,406 12,342, ,165,553 14,152,041 63% 58% C 62 10,484,783 1,500, ,318,329 1,500,000 39% 49% D 4 181,435 44, ,435 89,002 0% 0% E 40 6,753,394 1,000, ,611,377 1,000,000 33% 32% F 20 9,847,706 2,613, ,092, ,957 40% 28% ,512,046 19,000, ,575,234 19,000,000 57% 55% 21

22 Putting It All Together and Individual optimization work together to lower costs without impact to performance. 22

23 Business Case Study Large Non-Profit Mailer 23

24 Background A nonprofit membership organization >50% of new members from direct mail program 200MM pieces of mail/year Monthly mailings Using hundreds of different list sources 24

25 Background (cont.) Response rates decreasing Thus, cost per new member (CPNM) is increasing ** Only way to keep CPNM constant is to reduce costs** 25

26 Cost Per New Member Drivers 26

27 Solution Developed a Data Sourcing and Optimization Strategy Began with a detailed analysis of the historical performance Sourcing Optimization was applied to ensure the purchase of the optimal combination of lists that would yield the highest response and lowest costs Attributed value to data for selectivity, not for the name/address alone 27

28 Input Names (in MM) Source Mix House Compiled PDB Response Multibuyer Even though Compiled names doubled, cost was reduced significantly by licensing a secondary file (primary file is used for PDB) Response list cost have gone down even though input quantity has stayed constant = buying smarter 28

29 Impact Performance Improvement 1.4% over Forecast Savings: $2,651,082 a 49% Expense Reduction 46% Reduction in Cost Per Order Improvement in nets of 8% 29

30 30 THANK YOU!

31 Merkle Analytics Gary Robinson Vice President, Data Solutions

32 Table Questions 1. How do you manage your list costs today? 2. How does the LSO solution compare? 3. Have you ever performed a list level appeared-on response analysis across lists? 4. Did the LSO solution make sense to you? 5. What would prevent you from using an LSO solution in your organization? 32