Day1 - NMA-NIELSEN _ Mix Intro.pptx

SDVKDKFNSDOVHSODVHOSVHOSVHOSCHOAVHCODUASVCALSCNALDNCVDVNSKDVNKSDVBAVBGJZZCBKZXCH8ASDHCISIKDBVKVBIASUCBKN BXKVBKSJDVBAKDJVBAKDVBASKDVBAKDVBAKVBAKJDJVBKSDJAVBKSDVBKSDVBKSDBVKSDBVKSDVBKSDVBKSDVBSKDJVBKSDVBKSJDVBKSDVBKSJDJBVKSDVBSDKVBSKDVBSKDJVBJKSDVBQdNJACKJBZXKJHB

1. NMA and NIELSEN Mix Introduction

2. History of NMA Pioneered MMM automation Pioneered Bayesian estimation for marketing mix modeling Solved aggregation bias with Store-Group Model Pioneered real-time MMM tracking tools Stabilized real-time MMM with Intelligent Priors SM 1992 1994 1995 2001 2010 2012 Stabilized MMM response curves by tying to effective frequency Founded by Ross-boy Link, Marketing Analytics became part of Nielsen in 2011, and has continued to push the frontiers on innovation. R&D TIMELINE: MARKETING ANALYTICS Industry leading Digital Media Consortium to establish best practices for a) modeling digital assets, b) modeling granularity 2014/2015

3. Shanghai` Singapore Geneva New York Mexico City Regional Headquarters Local Offices Mumbai Nielsen’s MMM Practice As the largest MMM practice in the world, Nielsen helps clients optimize their media mix allocation via a holistic, customized approach that ensures sustained profitable growth.

4. MARKETING MIX MODELING

5. Marketing Mix modeling (MMM) is an application that helps answer key marketing questions Is my marketing paying off? What is the ROI from each of my marketing vehicles? How much Incremental impact is my marketing spend driving to my brand? How can I OPTIMALLY ALLOCATE my spend, given my limited budget?

6. Some specific questions answered by marketing mix models include What marketing tactics drive my sales? What tactics explain changes/growth in sales year-over-year? How effective is each tactic? What is the ROI on my marketing investment? How does marketing response/ROI vary by brand? Is there cross halo impact across brands? What is the halo effect from Brand A advertising on Brand B? What are the saturation points for each of my tactics? Are some tactics approaching Diminishing Returns? How can I optimize allocation of my current marketing dollars ? By vehicle? By Brand? By Market? Cross Country?

7. Role Of Marketing Mix modeling : KEY DECISIONS ENABLED BY MMM Marketing professionals strive to make fact-based, insight-driven decisions about Marketing investment levels, and resource allocation. Marketing Planner helps guide those business decisions. How much should we spend on marketing? How much should we spend in a given geography or product? How should we allocate spending across channels ? What is the best way to execute within a channel? Key Decisions MMM is NOT a decision making tool. It is a decision support tool to help you guide make more fact based decisions

8. A multi- variate OLS regression model Spend is not an input into the model* Volume impact is the modeled outcome. Cost/ROI impact is derived, against activity spend and margin *Spend can be an input if data for media weight does not exist. Using spend as an input , however, is not ideal. Mix Model- How does it work? Competition Product: Category/TM/Brand/PPG; Geography: Store, DMA, Regions/Other Dimensions Dependent: Volume (POS Data) Model Dimensions Modeled Output TV GRPs Digital Imp/GRPs Print GRPs Radio GRPs Price Seasonality/Weather/ Holidays/ Derived Output Incorporate Defined Spend and Margin Multivariate Regression Model (Mix Model) Discount Trade Promotion ROI Volume Impact from each driver

9. Data Inputs : extensive list of relevant media, trade, pricing & base Factors Pricing In-store Trade Promotions Paid Media (campaign level, if Available) POCM/Accessories/ Estra Point Regular Price Discounts Key Price Points 3 2 1 4 Competitor Price Features & OR Display Coolers FSI/Coupons 2 4 3 5 1 In-Store Displays 2 1 Television Magazine Radio 3 5 4 Digital (Banner, Video, Mobile, Search, Social) In-Store/ Shopper Mktg. Out-of-Home 6 Distribution & Product External 2 1 Avg. # of items Distribution Product mix changes/New Product 3 5 4 Competitive Distribution Product Inventory/Out of Stock 2 1 Competitive Marketing (Data Permitting) Temperature/Precipitation vs. Normal Seasonality/Key Holidays 3 4 Economic Conditions 5 Category /Industry Trend 6 Any other Regulatory Changes 3 years weekly (or daily) history aligned with POS By geography where execution varied, e.g., store, zipcode, DMA National if it is the only data available In each modeling round, ABI is modeling 140 million data points. So far, we have modeled ~500 million observations

10. We build models at the most granular level data that is available Uses 2-3 years of weekly or monthly data sales/volume data (preferably weekly data) Predictors (base, marketing, competition/external factors) are collected or mapped to geographies at which we are modeling *Geog. can be DMA, Market, Region, Cities, or any level of granularity that data is available for O O O WK 1 2 3 4 t O O O WK 1 2 3 4 t O O O WK 1 2 3 4 t O O O WK 1 2 3 4 t O O O .COM Store Level Other Geog.* Level Store 1 Store 2 Store n WK 1 2 3 4 t O O O .COM WK 1 2 3 4 t O O O .COM O O O Geo. 1 Geo. 2 Geo. n OR, 1. Level at which the models are built 2. Rolled up to higher geographic levels of interest Reporting Level

11. Weekly Volume Sales Examp l e Only Marketing Mix Analysis takes total sales volume…

12. Weekly Volume Sales Example Only …And Quantifies Changes In Sales Due To Individual Executions So in one line, “Marketing Mix Modelling quantifies the sales driven by each ingredient in the marketing mix” Stat Units

13. What is happening? Decompose sales patterns to understand volumetric impacts of marketing and operational activities Marketing effectiveness/efficiency and base trends are calculated to explain volume change We provide tangible recommendations to optimize your marketing mix Why is it happening? Future improvements Volume Contribution This Very Basic Concept Allows Us To Do Many Things!

14. {Expected everyday volume} No Trade or Media {Incremental Volume} Trade {Incremental Volume} Media INCREMENTAL VOLUME BASE VOLUME Base Media-Traditional Trade Distribution Seasonality Base Price Competitive Base Price Competitive Trade Macro-economic Trend Discount Feature-Newspaper Ads In-Store Display Traditional Media (TV, Print , Radio, Out of Home) Digital Media (Display, Online Video, Facebook) Media-Digital Models quantify Incremental volume (separate from base) Models separate and quantify incremental volume from base

15. Models decompose volume into driver contribution Marketing and Base Factors Across Time Base Base vs. Marketing driven Incremental Sales Macroeconomic factor putting a downward pressure on base Aligns sales with all drivers, both marketing and non-marketing Statistically attributes “lift ” to each driver Lift from Trade Lift from TV Models Separate Base from Driver Sales (Incremental) Base = Sales that would have occurred even in the absence of marketing Incremental = Sales that occur only because of marketing Regression Model…. and….

16. Incremental Volume Support TV support data from Agency Output from Model Volume Response = 657,673 3,720 TV Volume Response = (per 100 TVGRP) = 17,679 X100 Volume Response Calculation

17. Incremental Vol. X Price (Net or Gross) Spend of Activity Predicted by model, scaled to shipment volume = If Payout is $1.00, ROI is breakeven ( For every $1 spent, $1 in Revenue is gained) If Payout is < $1.00, ROI is negative If Payout is > $1.00, ROI is positive ROI Efficiency Incremental volume along with financial info are used for ROI calculation

18. Mix modeling zeroes in on the determinants of Effectiveness and CPM CPM Margin Effectiveness ROI Decomposition Effectiveness Driven By: Copy Strength Saturation and Flighting Execution Core volume ROI can be impacted by multiple factors CPM = Cost per mille  (CPM), also called cost per thousand (CPT) (in Latin, French and Italian, mille means one thousand), is a commonly used measurement in advertising. It is the cost an advertiser pays for one thousand views or clicks of an advertisement.

19. Mix MODELING Flowchart- sequential & iterative process Collect data for POS, Media, Trade, Sponsorships etc. QC and Review Data Data transformations ( adstock , saturation etc.) Check Model Results using Diagnostics Run Model Determine Model Specification Iterative Process Create Response Curves Load into Optimizer & run resource allocation scenarios Final Presentation with ROI and resource allocation results 1 2 3 4 5 6 7 8 9

20. How do we do it?

21. Our Aggregate MMM is done via simple, OLS regression Regression is a technique that involves fitting a line to a scatter of points. For example: We see, visually, that there could be a relationship between Sales and Television execution, namely as Television execution increases, Sales also increase. Multiple Linear Regression How Does It Work? Television Execution Sales

22. We could estimate Sales by assuming the mean at all points (blue line below), but this does not take into effect changes due to shifts in Television execution. We see that Sales trend up as Television execution increases. But how do we test this to show statistically that this relationship does exist. A line like the red one would be a more realistic model of these changes. But there can be many lines, which one do we pick? How do we quantify the effect that increasing Television execution has on sales? Introduction The Concept Of Regression Television Execution Sales

23. Simple Linear Regression Y = β + β 1 x + ε β 0: The y-intercept of our estimated line β 1: The slope of our estimated line ε : The error β ε β 1 Television Execution Sales The regression line chosen is the one that minimizes the sum of ε 2

24. Simple Linear Regression Back to the Example Y = β 0 + β 1x + ε Sales = 100 + 10*TV + ε Interpretation : A one unit increase in TV execution leads to a 10 unit change in sales Simple Linear Regression Example

25. In Practice, There Are Many Marketing Executions That Need To Be Accounted For, Isolated, And Quantified By The Regression

26. In a multiple linear regression model the dependent variable, or response, is related to k independent, or regressor, variables as: Y = β + β 1 x 1 + β 2 x 2 + … + β k x k + ε This follows the same logic as the simple linear regression model except that we have multiple variables predicting the value of Y. In multiple regression, β 1 shows the impact of a one unit change of x 1 on sales holding x 2 through x n constant. That is, the impact of each variable is independent of the other variables in the model. Multiple Linear Regression Multiple Linear Regression

27. Multiple Linear Regression The Marketing Mix Model Equation (Additive Equation) Sales t = Constant(s) + b 1 * Price t + b 2 * Distribution t + b 3 * TV Advertising t + b 4 * Print Advertising t + b 5 * Trade Promotions t + b 6 * FSI Couponing t + b 7 * Competitive Activity t + b 9 * Seasonality t + …… + Prediction Error where, b i is the coefficient of the respective variables, and “t” denotes week “t”.

28. Multiplicative Model function is Used TO Capture the synergistic impacts across drivers Multiplicative model function to capture the synergistic impact Price & Trade modeled at the lowest level of granularity available Media is modeled with Adtosck & Saturation transformation Log (Sales) = f(Distribution, Price, Trade, Advertising, Base/Other Factors..) Sales = f(Distribution) X f(Price) X f(Trade) X f (Advertising) X f(Other Factors..) Variables are transformed to determine the best model fit in a highly iterative process Each model goes thru 25-35 iterations (runs) before it’s finalized.