Big Data + Big Analytics = Big Wins September 19, 2012 Marco Vriens The Modellers My Background • SVP at the Modellers, Innovation & Analytics • Microsoft and GE • Academics and marketing research firms • Editor of The Handbook of Marketing Research (Sage, 2006) • Author of The Insights Advantage: Knowing How to Win (2012) Discussion Points 1. The Big-Data Challenge 2. New and Big Data 3. Smart Use of Big-Data 4. Potential Value Areas 5. Some Examples 6. How “True” and Valuable Are Big-Data Insights 7. Key Take-Aways The Big-Data Challenge “Everywhere you look, the quantity of information in the world is soaring. Merely keeping up with this flood, and storing the bits that might be useful, is difficult enough. Analyzing it, to spot patterns and extract useful information, is harder still.” The Economist; “The Data Deluge”; 2/10/2010 Gartner says: • “The Big Data Challenge Involves More Than Just Managing Volumes of Data” • “The real issue is making sense out of the data and (…) helping organizations make better decisions.” Definition of Insights • “Insights are thoughts, facts, data, or analysis of facts and data that induce meaning and further understanding of a business challenge and answer essential questions and create an urgency to act or rethink a business challenge in terms of its problems or solutions.” New & Big Data New & Big Data Specific New Data Sources • Source: Google.com/Insights/Search • Data: Search frequencies for search terms (e.g. Red Wine, Flu) • Source: Blogpulse.com • • • • Data: Blog content mapped to # of mentions, positive, negative Worldwide no. of blogs estimated btw. 120-184M, 120,000 added every day 50% of US Internet population reads at least 1 blog Blogs can be a good predictor of market outcomes for new products • Source: Tweets scraped from Twitter and twitter.com/trendistic • Data: Time series of key phrases (Tweets cleaned for Key phrases) or Time series of Mood measures (Tweets mapped to mood states) • Source: twitter.com/trendistic.com • Text scraped from across the Internet • Data: conversations • Source: proprietary firms Data Strategy Data Strategy • Data inventory • Data readiness • Integrated designs Insights Discovery Process Big-Data Should Start with a Business Problem * Potential Value Areas New & Big Data Cases 1. Quantitative Trend Spotting 2. Early Warning Systems 3. Market Forecasting 4. New Product Idea Generation 5. Crisis Identification 6. Marketing Mix Modeling 7. Better Prioritization Question: Do Big-Data studies help the firm make better decisions? Big-Data, Different Decisions? KEY VALUE AREAS Time series on 40+ search terms on Wine varieties identified key trends Microsoft accurately identified in 2003 the size of the Linux threat BETTER DECISION? Doubtful Yes, $ 150 Million Ad campaign impacted Tweets mapped to mood states improves prediction of the DJIA No Intuit uses Twitter comments to identify product improvements Yes Unilever quickly assessed brand risk after Dove’s Anti-Age Ad was pulled Yes Various cases: Improved the predictive power of marketing mix models because Big-Data facilitates the inclusion of indirect effects and latent demand Don’t Know, but likely More opportunities for validation help prioritize insights Yes Two Scenarios Grab Big Data and… SCENARIO 1: Do a stand-alone analysis to get insights Example: Twitter Mood improves predicting the Dow Jones Industrial Average over and above its own past. SCENARIO 2: Mix the Big-Data with with more traditional marketing research data to get more and better insights Use Social Media data and mix it with traditional marketing mix data to show value of social media marketing in context of all other media options Scenario 1 Example Twitter Mood Predicts DJIA • Hypothesis: Emotions can drive behavior in addition to new behavior • Doing surveys at this scale to measure public mood would be very expensive and time consuming • Twitter-feeds-based sentiment tracking tools have been developed • Two tools were used: 1. Opinion Finder (software tool): provides a positive/negative daily time series of public mood 2. Google Profile of Mood States (GPOMS): composes a 6dimensional daily time series: Calm, Alert, Sure, Vital, Kind, Happy • To make original psychometric instrument applicable to Twitter data the researchers expanded the original 72 items in the POMS instrument to a lexicon of 964 associated terms • This allowed mapping to the 6 mood states (for details: www.terramood.informatics.Indiana.edu/data) Scenario 1 Example Twitter Mood Predicts DJIA • Data: • 10 million tweets by 2.7 Million users • Yahoo Finance closing values of DJIA • Opinion Finder & GPOMS were cross-validated against significant events such as Election day, etc. • Analysis: • Granger Causality Tests: Indicated that Calm Granger caused DJIA • Self-Organizing Fuzzy Neural Networks: • Base model: DJIA predicted based on its own history: 73% direction accuracy • Adding Calm dimension: 86.7% direction accuracy Scenario 2 – Example 1 Big Data Shows Value of Social Media • Firm-facilitated social media interaction can help explain variations in mind set metrics such as awareness, consideration, and preference • Firm-facilitated social media interactions explain an additional 25% of the variance in sales over and above the amount that is explained by media expenditures • This work was feasible through the combination of Big-Data + Advanced VectorAuto Regression. Source: de Vries et al. (2012), VAR Model for Effects of Social Media Interactions on Firm Outcomes, presented at the Marketing Dynamics Conference, 2012 Indirect Effects Improve Marketing Mix Models Better Estimates + Better Understanding Total effect = C + (A*B) Indirect Effects Improve Marketing Mix Models Better Estimates + Better Understanding Total effect = C + (A*B) Indirect Effect of Adwords Leads Quotes Orders Profit Online Online Online 27 % Offline Offline Offline 73 % Adwords Online Visits Scenario 2 – Example 2 Big-Data Insights Into Car Advertising • One published case study for a car manufacturer showed how online product information and quote request significantly added to the insights of a marketing mix model • Questions: • Does advertising lead to increased search? • Does advertising lead to accelerated sales? • Does advertising create new demand? • Data: • Advertising expenditures • Online search (117 online sources were used) – “hand-raising” data • Unit sales data Based on Dotson, et al (2012), Identifying the Dynamic Role of Advertising Through Online Search and Online Sales, Working Paper. Scenario 2 – Example 2 Big-Data Insights Into Car Advertising • Offline sales and online search dependent variables and indicators of latent demand • Modeling approach: Dynamic Linear Modeling TV Offline Sales Radio Latent Demand for Car Brand Online Search Print Adding online search significantly adds to the forecasting ability Scenario 2 – Example 2 Big-Data Insights Into Car Advertising • Modeling approach + Big-Data allows for better forecast quality • Results show that TV advertising drives demand creation, not just purchase acceleration • Approach allows “nowcasting”: • This is a great benefit – because in the car market online search data can be observed in real time, but reporting of actual sales often lags • This feature helps managers intervene much earlier if sales are expected to go down and gives them a vital advantage over their competitors Based on Dotson, et al (2012), Identifying the Dynamic Role of Advertising Through Online Search and Online Sales, Working Paper. Scenario 2 – Example 3 Data Fusion & Interactive Decision Support Tools • Survey data – indicating what engagement types customers used for engaging with firm; correlate with overall satisfaction • CRM data on what engagements customers actually used • Monthly reach data by country (35+) by engagement (15+) – resulting in huge cumbersome Excel spreadsheets • YTD Spend on engagements • Several survey studies + macro economic data that allowed us to estimate by country the size population Integrating Diverse Data Streams Survey Data Correlation of Engagement with Overall Satisfaction Internal Engagement Data Reach Data (Monthly) External data Audience Population Numbers YTD Spent Data Reach Penetration Cost per Touch Engagement ROI Benchmarking + Recommendations Scenario 2 – Example 3 Interactive Decision Support Tools Big Analytics DEGREE OF ADVANCED OR BIG ANALYTICS ‘Basic’ Intermediate Highly Linear regression Logit & MNL Regression Support Vector Machines Asymmetric loss functions Factor Analysis Structural equation modeling Hierarchy of effects models K-Means Latent Class Models Time Series Regression Vector Auto Regressive (VARX) Kalmam Filters/State Space models Big Analytics Can Yield Big Returns Case Studies Return on Modeling John Deere Double digit profit increase ABB Electric $100 Million Rhenania BauMax Profits increase 4x + Significant advantage of competition 8% profit increase Inofec Double digit profit increase Marriot $200 million initially, LT 1 Billion Qantas/Jetstar 4% market share growth, $45 million increase in revenue Using Big Analytics Comes With a Challenge Why Managers Don’t Trust Predictive Models Using Big Analytics Comes With a Challenge Why Managers Don’t Trust Predictive Models Analytics Trend Troubles Scientists In 2010, two research teams separately analyzed data from the same UK patient database to see if widely prescribed osteoporosis drugs increased the risk of esophageal cancer. They came to surprisingly different conclusions… How True? How Valid Are Big-Data Insights? • Target’s prediction of pregnant teen • Twitter-based prediction of flu Managers do not trust predictive models because they know that different datasets can Some Validation Risks Correlation is not causation give different results – different analysts or different modelers can get different insights – and they realize correlation is not causation Misinterpretation Stability and Validity Lucas Critique Key Take-Aways • Big-Data should be part of broader data strategy. • We should start with a business problem and assess whether it’s a logical target for Big-Data. • Using a mix of “old,” “new,” and “big” data is a more powerful approach than using Big-Data alone. • To leverage the potential in Big-Data, we need sophisticated Advanced Analytics – e.g., VAR/VEC models, DLMs, Fuzzy Neural Networks, Data Squashing etc. Key Take-Aways • Big-Data is not the solution to everything. Right now the killer apps seem to be marketing mix modeling and customer sentiment and satisfaction. • Validation of Big-Data insights is key, and… • Last but not least, Big-Data will not replace survey research. To get a true foundational understanding of why people behave the way they do we will often need to ask them specific questions that don’t just “arise” naturally in Big-Data. Thank you. Marco Vriens, Ph.D. SVP Methodology (801) 290-3838 Marco.Vriens@themodellers.com References Vriens, M. (2012), The Insights Advantage: Knowing how to win, i-Universe. Grover, R., and M. Vriens (2006), Handbook of Marketing Research: Uses, Misuses, and Future Advances. Thousand Oaks: Sage Publishing. de Vries, L. et al. (2012), VAR Model for Effects of Social Media Interactions on Firm Outcomes, presented at the Marketing Dynamics Conference, 2012 Dotson, et al (2012), Identifying the Dynamic Role of Advertising Through Online Search and Online Sales, Working Paper Bollen, Mao, and Zeng (2011), Twitter Mood Predicts the Stock Market, J of Computational Science, 2, 1-8. My blogs… • 10 Steps For Stretching Marketing Research For More And Better Insights (http://www.greenbookblog.org/2012/05/29/10-steps-for-stretching-marketing-research- for-more-and-better-insights/) • • 3 Reasons Why Big Data is Relevant (www.allanalytics.com) • Avoiding Big Data Disaster (www.allanalytics.com) • www.theinsightsadvantage.com Nucleus Research Note: The Big returns of Big Data