Marketing Attribution Problematic michel.bruley@teradata.com Extract from various presentations: RPM Direct, Kevin Hillstrom, Teradata Aster, … September 2012 www.decideo.fr/bruley Marketing Channels and Attribution Mail TV Print www.decideo.fr/bruley TM Quote TM Sale Marketing Channels and Attribution Mail TV TM Quote TM Sale Web Quote Web Sale Print Search Aggregator Banner email www.decideo.fr/bruley Marketing Channels and Attribution Mail TV TM Quote TM Sale Web Quote Web Sale Print Search Aggregator Banner email www.decideo.fr/bruley Marketing Channels and Attribution Mail TV TM Quote TM Sale Web Quote Web Sale Print Search Aggregator Banner email www.decideo.fr/bruley Marketing Channels and Attribution Putting Response in Correct “Bucket” Putting Sales in Correct “Bucket” Calculating Media Cost per Sale Spending Next Tactical Marketing Dollar Building Media Specific Targeting Models Making Strategic Business Decisions www.decideo.fr/bruley Channel Attribution Methodology • Individual Customer Identification Numbers • Marketing Source Codes • Name and Address Match • Factor Analysis • Dynamic Time Series Regression • Proportionately “Factor” Leftovers into Marketing Channels www.decideo.fr/bruley What is a latent conversion www.decideo.fr/bruley What’s wrong with Google Analytics? ROI is attributed to the last (latent) referrer (Most recent keyword, ad, email, blog etc.) What’s wrong with most recent? – Shoppers initiate search using broad categories – Later narrow down to product names/IDs – Perhaps then narrow to store brand Broad categories don’t get the credit they (may) deserve The only exception is direct type-in Google Analytics takes the blame, though many web analytics tools work the same in a default implementation www.decideo.fr/bruley Attribution examples Quiz: An existing customer receives a catalog on September 1. On September 2, the customer uses Google to search for merchandise, visits your website, and purchases an item. What % of the order do you allocate to catalogs, to search, and to organic brand loyalty? Quiz: An existing customer receives a catalog on September 1, receives e-mail marketing campaigns on September 7 and September 9, and purchases on your website on September 10, buying merchandise featured in your catalog and merchandise available only online. What % of the order do you allocate to catalogs, email, and to organic brand loyalty? Quiz: An existing customer receives a catalog on August 1, and receives 17 e-mail marketing messages between 8/1 and 10/1. On 10/4, the customer uses Google to search for merchandise, visits, and buys an item. What % of the order do you allocate to catalogs, to e-mail, to search, and to organic brand loyalty? www.decideo.fr/bruley Test + Results = Attribution Rules The beautiful thing about catalog marketing and e-mail marketing is that you can test, you can see what happens to other channels when you do not mail a catalog, when you do not send an e-mail campaign! Sample 80,000 twelve-month buyers with a valid e-mail address. Group 1 = 20,000: Catalogs = Yes, E-Mail = Yes Group 2 = 20,000: Catalogs = Yes, E-Mail = No Group 3 = 20,000: Catalogs = No, E-Mail = Yes Group 4 = 20,000: Catalogs = No, E-Mail = No Execute for a month, quarter, season, or year! In a controlled experiment, the results of your test tell you what impact catalog marketing and e-mail marketing have on other channels (search, mobile, social, display ads, affiliates), so you can set up reasonable attribution rules! www.decideo.fr/bruley Some Results We have four test panels in this test. We sent one catalog and nine email campaigns during a one-month timeframe www.decideo.fr/bruley What Is The Organic Percentage? The organic percentage is possibly the most important metric a direct marketer / catalog brand can track. It is the percentage of demand that will be generated if no marketing exists. What about our example? Take the $5.80 generated in the no catalogs / no e-mail test panel, and divide it by the $11.37 generated in the catalogs + e-mail test panel. The result is 51%. www.decideo.fr/bruley 13 A 51% Organic Percentage We must execute catalog and e-mail mail/holdout test panels, in order to properly estimate what our organic percentage is. • When the organic percentage is < 20%, your matchback/allocation process is generally accurate. • When the organic percentage is > 40%, matchbacks and allocation programs become increasingly inaccurate. Does The Organic Percentage Vary? • Some customers are “highly organic”, while other customers require “large amounts of marketing”. There are HUGE profit opportunities in knowing this difference! • Customers who mail orders to a company or use the telephone to order require advertising. • Customers who combine catalogs and online channels are a “hybrid”, requiring much less advertising. • Customers who order online or in stores are highly organic, you can reduce advertising! www.decideo.fr/bruley Key Takeaways Have the courage to execute both catalog mail/holdout tests and e-mail mail/holdout tests. Test four panels, test for a quarter or season or year if you can. The results are going to be breathtaking!. Catalogs: In many cases, orders that would have happened anyway are attributed to catalogs, causing us to spend way too much money mailing catalogs. E-Mail: E-Mail is frequently cannibalized by catalogs. E-Mail frequently causes Search/Mobile orders to happen. Search: Search is often the outcome of catalog marketing or e-mail marketing. Social/Mobile: In the early stages of a channel, sales are frequently cannibalized from existing channels, or the existing channels cause the sale to happen in the new channel. Over time, new channels become “organic”, and do not require oldschool channels in order to create sales on their own. Tests can validate this. www.decideo.fr/bruley Multi-Channel Customer Analysis • Business Question(s): • Prior to new product additions? Is there any identifiable pattern of behavior prior to account closure? • If so, what does this pattern look like? www.decideo.fr/bruley Value of Aster Data for Digital Marketing Aster Data Analysis • • • Click-stream • # of visitors, visitor location, browser type • Last click analysis Online behavior • Common interaction behaviors • Optimal paths through website Where to place this button, link, etc. • Search Optimization Multi-channel (online & offline) campaign analysis - Mix of paid per click and organic • Which search terms drive traffic, behaviors Advertising Optimization • Conversion Complete customer interaction history Marketing return on investment • • • • A/B testing • • + Teradata Adds… • How to optimize advertising placement • Where are shopping carts abandoned & why Attribution + cost of conversion Marketing Attribution • What % credit to give each referring channel or campaign www.decideo.fr/bruley + Aprimo Adds… • Campaign management - Take action to influence behaviors • Marketing Resource Management - Take action to optimize marketing spend Attribution Using Aster www.decideo.fr/bruley Aster nPath Identifies the “Last Mile” All interaction patterns evaluated in a single pass userID 10001 Prepares multi-structured data • 20001 Stitches rows together by customer in a time-ordered view Aster MapReduce Platform event time userID event time 10001 10001 20001 20001 10001 20001 Scans all records to produce a complete set of paths Step 1: Pivot data via nPath SQL-MapReduce • No need to define patterns in advance custID channel1 … channeln time1 … timen • Fully parallelized for top performance using MapReduce where SQL falls down 10001 Online Retail … Research products 12:00 PM 1/1/2010 … 3:00 PM 2/15/2010 20001 Store Purchase … BankX Credit Card 1:45 PM 1/1/2010 … 12:20 PM 2/22/2010 Summarize output for business exploration • Rank order the most popular paths and yet represent the long tail too www.decideo.fr/bruley Step 2: Run nPath SQLMapReduce Java Logic Total # of Customers channel1 … channeln 35 Online Retail … Research products 26 Store Purchase … BankX Credit Card Aster nPath example: Account Closure www.decideo.fr/bruley