Transcription
Direct Marketing Analytics with RuseR! 2008Dortmund, GermanyAugust, 2008Jim Porzak,Senior Director of AnalyticsResponsys, Inc.San Francisco, CaliforniaRevised Sep08
Outline Introduction– What is Direct Marketing (DM)?– How does “analytics” play a role?– What's Special About DM data & analytics?DM data requirements - Class StructureBasic DM MetricsTestingSegmentationModelingDirections & Questions(Appendix with resources & links)09/03/08useR! 2008 -Porzak - DMA2
Introduction09/03/08useR! 2008 -Porzak - DMA3
What is DM? Also know as “direct response marketing.”Characteristics:––––– Directed at targeted individuals or demographicResponse is asked for and expectedTracking of responses back to sourceEvaluated by counts and value [ , , , .]Testing of alternate elements is implicit in DMElements (in order of importance):1. List2. Offer3. Creative09/03/08useR! 2008 -Porzak - DMA4
Channels used in DMClassical Individual– ––– Direct MailDemographic–InternetIndividual– AdvertisementTV or arch PaidFreeRemember, all of above ask for a response that is traceable back to source!09/03/08useR! 2008 -Porzak - DMA5
Use Analytics to Answer these Questions Directed to whom?–Predicting responses –Segmenting population – To use best offer, creative, & channelEvaluated with accepted metrics–– Which of list, or part of list?When to send?Open definitions are important here.Use confidence intervalsTesting to improve next time around.–09/03/08Show significance of resultsuseR! 2008 -Porzak - DMA6
So What's So Special? Statistically speaking?––Not much.But remember the nature of DM problems: The audience!–– Huge N (typically 104 to 107)Small proportions (often 3% to 0.05% for direct or email)The corporate worldDMers themselvesThe Data Structure––09/03/08Levels of granularity“Campaign” hierarchies drives testinguseR! 2008 -Porzak - DMA7
“It's the structure, stupid!”09/03/08useR! 2008 -Porzak - DMA8
The DM Process (Individual) Postal MailOutbound–Mail a “piece” –Tagged? Recipient responds Return mailCalling 800#Visits––Send a “message” Inbound–EmailOutboundInbound–ISP –WebPhysical location useR! 2008 -Porzak - DMABounceOpt-outRecipient 09/03/08Tagged?OpenClick(Request or Buy)Opt-out9
Data Elements Details– The “List” (perhaps with additional data)– Send Events– Response EventsSummaries– Response counts & rates (total & unique)– Simple “cell-level” metricsCampaign Meta-data– Costs and Values– Time window– Batch or Triggered–09/03/08useR! 2008 -Porzak - DMA10
Class & Method Challenges Detail & Summary classes. straightforwardCampaign wise meta-data. straightforwardCampaign elements & relations. harder!– summary, print & plot should be able to understanda group of campaignsand elements withina campaign.Leverage arules?From package vignette: arules.pdf09/03/08useR! 2008 -Porzak - DMA11
DMA Modules09/03/08useR! 2008 -Porzak - DMA12
Direct Marketing Metrics Direct Mail–– Response counts & rateCost per response (sale, lead, .)Email–all above, plus email specific metrics Opt-out, bounce, open, click counts & ratesAdd unique opens, clicks, responsesGeneral–––09/03/08Campaign ROIList growth (opt-ins / -outs per time period)List fatigueuseR! 2008 -Porzak - DMA13
DM Testing Simple 2-way: A/B, Control/TestMultiple test against control: A/BCD.True MVTGoal is appropriate analysis done based oncampaign meta-data.09/03/08useR! 2008 -Porzak - DMA14
Example A/BC Test09/03/08useR! 2008 -Porzak - DMA15
Segmentation for Targeting Behavior based–– PurchasesUsageAttitudinal–09/03/08Preference / Interest SurveyuseR! 2008 -Porzak - DMA16
Purchase Behavior Example09/03/08useR! 2008 -Porzak - DMA17
Purchase Behavior CategoriesFor executive presentations, we re-draw the segment cells in this way:09/03/08useR! 2008 -Porzak - DMA18
Modeling for List Optimization Model full list to select those recipients withhighest expected response to offerMethods include logistic regression andmachine learning tools like random forest.Supply “model validation” curve (ROC) somarketer can pick “depth of file” to use basedon economics of the offer09/03/08useR! 2008 -Porzak - DMA19
Response Prediction Example09/03/08useR! 2008 -Porzak - DMA20
Future Directions Finalize class structure– Need to work through more use cases Feel free to send examples!– Sketch method dependenciesRoadmap– Independent batch campaigns– 2-way & n-way against control– Triggered campaigns– True MVT– Segmentation– Response ModelingOn R-Forge: https://r-forge.r-project.org/projects/dma/– Collaborators welcome!09/03/08useR! 2008 -Porzak - DMA21
Thanks!09/03/08useR! 2008 -Porzak - DMA22
Appendix09/03/08useR! 2008 -Porzak - DMA23
Links & References Books– Metrics Davis, Measuring Marketing – 103 Key Metrics Every Marketer Needs,Wiley, 2007 Farris, Bendle, Pfeifer & Reibstein, Marketing Metrics – 50 Metrics EveryExecutive Should Master, Wharton, 3rd printing, 2006.– Marketing Libey & Pickering, RFM & Beyond, MeritDirect Press, 2005. Alan Tapp, Principles of Direct and Database Marketing, 3rd Edition, Pearson,2005. A. M. Hughes, Strategic Database Marketing, 3rd Edition, McGraw-Hill, 2006.Links– Related Talks on www.porzak.com/JimArchive/– dma on R-Forge https://r-forge.r-project.org/projects/dma/– Responsys.com Resource Center– Direct Marketing Association International Resources– Email Experience Council Home– EmailLabs: Glossary, Benchmark Data– use R Group of San Francisco Bay Area http://ia.meetup.com/67/09/03/08useR! 2008 -Porzak - DMA24
Direct Marketing Analytics with R useR! 2008 Dortmund, Germany August, 2008 Jim Porzak, Senior Director of Analytics . arules.pdf. 09/03/08 useR! 2008 -Porzak - DMA 12 DMA Modules. . – Direct
