Social Analytics Andy Fisher Chief Analytics Officer Merkle Inc. November 15, 2013 © 2013 Merkle Inc. All Reserved. Confidential © 2013 Merkle Inc.Rights All Rights Reserved. Confidential. 1 Agenda • About Merkle • Social Definitions • Media and Analytics History • Social Integration Approaches • Industry Examples • Takeaways © 2013 Merkle Inc. All Rights Reserved. Confidential. 2 Company Overview Distinctive experience Extraordinary expertise • Privately held by current management since 1988 • 2,000+ Employees • 150+ world class clients • 250+ advanced degreed statisticians • Manage over 140 marketing databases • 300+ dedicated digital professionals • Manage 1.6+ petabytes of customer data • 650+ marketing technology professionals • Inform over $10 billion marketing decisions annually • 100+ creative professionals Sustained 25% growth since 1989 $400 $365 Revenue in millions $350 $303 $300 $255 $250 $211 $223 $180 $200 $150 $320 $148 Awarded and recognized • Bronze Stevie Award “Business to Business Marketing Campaign of the Year- Business Services” ‘13 • Ad Age A-List: “Agency to Watch in 2012” • Largest privately-held agency in U.S., Ad Age ’12 • 6th Largest CRM/Direct Marketing Agency, Ad Age ’12 • Recognized by SmartCEO Magazine as a Future 50 Company ‘11 • NCDM Awards ’04, ’05, ’06, ’09, ’10, ‘11 $100 • Recognized by Forrester ’03, ‘06, ’07 &’10 $50 2006 2007 2008 2009 2010 2011 2012 2013E • Multiple MAXI Award Winner ’10 ’11 ‘13 • Multiple DMA Innovation Award Winner ’10 ‘13 © 2013 Merkle Inc. All Rights Reserved. Confidential. 3 Our clients represent many of the best global brands © 2013 Merkle Inc. All Rights Reserved. Confidential. 4 Definitions Type Definition Example Paid Placements that and advertiser pays for on •Ads in FB newsfeeds a social website •Sponsored content on Twitter Owned Experiences where advertiser controls the content •Advertiser website •Advertiser FB Page Earned Content about an advertiser where the advertiser does not control the content •Tweets from users •Facebook likes © 2013 Merkle Inc. All Rights Reserved. Confidential. 5 A brief history of Online Media Audiences aggregated by content 1995-2005 Differentiation created by Media Skills Audiences aggregates by scaling niche content 2005-2009 Differentiation Created by Optimization Audience aggregated by third party data Audience aggregated using known relationships 2009-2012 Differentiation Created by Technology 2013+ Differentiation Created by Data Integration and Analytics © 2013 Merkle Inc. All Rights Reserved. Confidential. 6 Reality Check Sell side is far ahead of the buy side Legal and operational infrastructure is problematic To do this well you need A new combination of skills • Agency • Programmatic expertise • Classical planning expertise • CRM • First Party data management Open Issues Very new approach • Legal • Scale • Business standards (T&C’s, makegoods, etc) • Integration of Creative and Analytics We believe this will become a major driver of value Est. $8bn Totus spend by 2017 © 2013 Merkle Inc. All Rights Reserved. Confidential. 7 History of Marketing Mix Optimization and Attribution MMO begins as custom one-off projects Modern MMO emerges in CPG Audiences Industry 1940s-1970s 1980s Low adoption, lack of data, lack of computing power aggregated by content Syndicated scanner data revolutionizes industry • Mainly academic until 1970s • Computing problematic • First MMO product in 1979 • Regression • MMO is panel based, similar to attribution today • Panel approaches fade (will remain as forecasting tools) MMO scales outside CPG to Audiences include Auto, aggregated Finance and by content Pharma 1990s MMO (top down) and Attribution (bottom up) unify 2000s Computer power increase (still mainframes though) • MMO becomes the standard approach for CPG • Models become more complex • First digital models in 1999 Digital media disrupts MMO Audiences industry. aggregated Recovers by late by content 2000s Mathematics of digital need to be created. • Bayesian, Markov, agentbased and other models emerge • First attribution models in 2005 (based on 1979 panel models) © 2013 Merkle Inc. All Rights Reserved. Confidential. 2010s Mathematics of paid digital fixed, computation cost falls to $0 • Focus on speed and actionability • Implementation becomes limiting factor • Social become the next frontier 8 Common Approach for Integrating Social © 2013 Merkle Inc. All Reserved. Confidential © 2013 Merkle Inc.Rights All Rights Reserved. Confidential. 9 Step 1: Integrate Paid and Owned Compile Data Online Email Activity (opens, clicks) Web behavior (weblog) Email subscription User communities Online registration Coupon downloads Digital media exposure Social Behavior Offline POS Data Coupon Redemption Shopper Panel FSP In store promo Objective Function Model the Data and Iterate Tie Engagement to Revenue (Sample) Brand X 1.5x Brand Y 1.3x Brand Z 2.3x Evaluate Individual level Activity level Segment Based Determine Critical KPIs Constant Evolution Data drives what to measure Segment level © 2013 Merkle Inc. All Rights Reserved. Confidential. 10 Step 2: Build Attribution Models Day 8-30 Day 1-7 Day 0-1 Transaction/ Conversion Actual experience $ Direct or Rules Based (Basic) 0% 0% $ 100% 100% Probabilistic Model Based (Advanced) 3% 14% 3% 0% 5% 5% 5% 15% 5% Mass and Offline Direct mail sent TV view 5% $ 40% Digital Newspaper view Display view Website visit Social visit © 2013 Merkle Inc. All Rights Reserved. Confidential. Paid search click 11 Step 3: X-channel modeling with Paid, Owned and Earned TOP-DOWN MEDIA MIX MODEL (POS Data) National media (TV & radio) Local media (TV & radio) Direct mail Social Digital $140 $200 $180 $113 $83 calibration layer BOTTOM-UP CONSUMER MODELING (Individual Response Data) @ Display/ Video Social Paid Search Email $60 $113 $91 Direct mail $80 $180 Campaign $ Network $ Engine $ Program $ Program $ Placement $ Program $ Branded $ Campaign $ Campaign $ Creative $ Campaign $ Keyword Segment $ Segment $ Offer $ Segment $ Campaign $ SEGMENT BASED ATTRIBUTION Enlightened Consumer Time Savvy Mom Attribution needs to tie engagement to sales and be customer centric by enabling segment-based performance results of marketing © 2013 Merkle Inc. All Rights Reserved. Confidential. 12 Reality Check To do this well you need MMO + Attribution • Media Mix Optimization together with Attribution • One model • Integrates Offline and Online • Most vendors doing some form of this • Historically the two are related Better Math Attribution gaps • Incrementality • Decay Factors • Interplay of media • Math is understood X-Channel experts • Data Scientists • All channel data • Data Viz Experts • Hard to find Open Issues Data Quality – Cooke Data has Issues • Idea 1: Leverage CRM data • Idea 2: Leverage Panels/”Good” samples • Idea 3: Statistical Identification/Fingerprinting • Idea 4: Model Cookie Deletion • Idea 5: First Party Data • No Silver bullet Lack of Validation Social Challenges • Rare in MMO • Very rare in Attribution • No agreed upon methodology • Predicitvity • Models • Sentiment Problems © 2013 Merkle Inc. All Rights Reserved. Confidential. 13 Step 4: Deploy into optimization and personalization process Creative selection Increased Sales Force Prospecting Decisioning Digital Media For display, optimization output is directly fed to the Demand Side Platform For SEM, bidding engine weights are output of analytics correlating keyword to high-value segments Agent Optimization Feed Sales force automation / call planning prioritized by agent’s segment and value mix For site, real-time integration of segments into offer management system Search Increased bid amount on “retirement planning” Site Served Fixed Income X-sell offer © 2013 Merkle Inc. All Rights Reserved. Confidential. 14 Reality Check Many people doing within channel personalization well Far fewer are doing x-Channel personalization To do this well you need Cross Channel Infrastructure • Data source across channels • Enterprise segmentation • Cross channel Technology • Within-channel activation • The technology is good enough Open Issues Organizational Silos • Different Goals • Different notion of customer • Lack of incentive • Different measurement approaches Think this will become a large practice moving forward Big opportunity for people in the room © 2013 Merkle Inc. All Rights Reserved. Confidential. 15 Industry Successes (and Otherwise…) © 2013 Merkle Inc. All Reserved. Confidential © 2013 Merkle Inc.Rights All Rights Reserved. Confidential. 16 Tying social media initiatives to business outcomes CHALLENGE Company X needed a way to optimize movie promotion dollars in an environment where every dollar is spent prior to the first ticket sale. APPROACH: • Model buzz as the response variable to media • Determine how buzz maps to box office sales – Substantial historical database for modeling – Bass-diffusion models simulate buzz build – Incorporate HSX data • Leverage buzz as campaign optimization tool • Real-time forecasts of box office • Learnings – – – – Long dev process/custom Treat as experimental Required deep vertical knowledge Success driven by lack of feasible plan B © 2013 Merkle Inc. All Rights Reserved. Confidential. 17 Incorporating buzz metrics into media mix models CHALLENGE Coca-Cola wanted to leverage buzz to improve their MMO and potentially replace expensive brand tracking surveys. APPROACH: • Incorporate buzz into MMO (input and output) • Results: Buzz showed no correlation with sales • Caveat 1: FB, Blogs, Twitter, YouTube only • Caveat 2: Sentiment analysis challenges • Next steps – refine approach and continue • Learnings – Long dev process/custom – Treat as experimental – Industry Backlash/PR issues © 2013 Merkle Inc. All Rights Reserved. Confidential. 18 Getting past aggregate-level data CHALLENGE In order to leverage social data for CRM, companies must get to segment-level and even userlevel data which isn’t available from most social media platforms. APPROACH A technology retailer followed a four-step process to build out a social data warehouse that will connect to the CRM database. 1. Identify available social data assets and sources 2. Evaluate applicability of assets via three criteria: • • • Identifiable - does the data source help us connect to a customer/prospect record Actionable - can this data source be used to power more targeted marketing – via email, social, or site content or analytics/insights Scalable - how large is the population of identifiable and actionable records now and what is its expected growth 3. Prioritize social data assets based on score and review privacy implications 4. Define use cases to test CRM program impact © 2013 Merkle Inc. All Rights Reserved. Confidential. 19 Forecasting Tune-in Challenge: Making Social Measurement Predictive Example: Sharknado (SiFy) Sharknado runs on SyFy Major social media sensation 5000 tweets per MINUTE! SyFy plans a rerun of Sharknado SyFy green lights Sharknado 2 Twitter competition for tagline Sharknado rerun airs and… 1.89 MM Viewers Not too bad. But not great either. Sharktopus = 2.5 MM Viewers Average = 1.5MM Viewers GoTH Red Wedding = 5.13MM © 2013 Merkle Inc. All Rights Reserved. Confidential. 20 Takeaways • Constantly Evolving Space • Many problems and opportunities • Trend towards first party data • Pockets of measurement success • Long way to go • Social cannot be meaningfully measured in a silo • Holistic approaches are necessary • Lots of hype – and lots of content too! © 2013 Merkle Inc. All Rights Reserved. Confidential. 21