+ All Categories
Home > Documents > The Marketing Mix Crisis, Attribution Models & The...

The Marketing Mix Crisis, Attribution Models & The...

Date post: 09-Mar-2018
Category:
Upload: hadan
View: 217 times
Download: 1 times
Share this document with a friend
27
Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 1 May 1, 2014 The Marketing Mix Crisis, Attribution Models & The Experimental Design Renaissance
Transcript
Page 1: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 1 May 1, 2014

The Marketing Mix Crisis, Attribution Models & The Experimental Design Renaissance

Page 2: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 2

How Do I Make Sense of All the Ways to “Attribute”?

Audience Verification

Behavioral Analysis

Brand Lift

Cross Media Modeling

Sales Lift/ROI Analysis

Marketing Mix Models

Attribution Modeling

Cross Media Reach Analysis

Page 3: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 3

How Do I Make Sense of All the Ways to “Attribute”?

Use the ABCS of Marketing Effectiveness Funnel as an easy way to organize the measurement ecosystem in your brain!

Audience Verification*

Behavioral Analysis Brand Lift/ CMM

Sales Lift/ROI Analysis Marketing Mix Models

Cross Media Reach*

Attribution Modeling

Page 4: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 4

How Do I Make Sense of All the Ways to “Attribute”?

*Audience Verification and Cross Media Reach neither models nor experiments

Experiment Design Econometric Modeling ! Methodology "

Sales Lift/ROI Analysis Marketing Mix Models

Attribution Modeling

Cross Media Reach* Audience Verification*

Behavioral Analysis Brand Lift/ CMM

The two methods for attributing:

Page 5: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 5

The Marketing Mix Crisis

Page 6: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 6

The Marketing Mix Crisis

Or NOT….

Page 7: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 7

Growing Concerns Old and New

• What about long term effects?

• Who has time to wait a year?

• How do I measure digital media ROI in an MMM?

• Display

• Video

• Mobile

• Tablet

• Owned (website, email)

• Social

• Search

Page 8: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 8

The Marketing Mix Crisis

Or NOT….

Page 9: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 9

Correlation is Not Causation

“A statistical relationship, however strong and however suggestive, can never establish causal connection: our ideas of causation must come from outside statistics, ultimately from some theory or other.”

M. G. Kendall and A. Stuart – The Advanced Theory of Statistics

“To ascribe causality, one must appeal to a priori or theoretical considerations.”

D. N. Gujarati- page 21 Basic Econometrics

Page 10: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 10

Top Criticisms of Marketing Mix Modeling

• Correlation is Not Causation

• Proper MMM overcomes this by drawing from Economic/Econometric Theory

• And scientific cross validation with experimental design

• Not always apples to apples comparisons of channels

• Price, promo, in-store favored vs. all media (Long Term Effects)

• Measures only one return objective of advertising (Sales)

• Supplement with other techniques if you seek to optimize

Page 11: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 11

Marketing Mix is Not in Crisis but it will Change

• MMM remains an excellent tool for marketers, the tires have been kicked for years

• MMM is increasingly in need of supplemental analysis especially for understanding of the interactions with the Digital Media Mix, as well as understanding what’s happening earlier in the consumer journey

• MMM is being disrupted by “Big Data” that is single source consumer level which will lead to a hybrid of econometric, existing frameworks and experimental design techniques

Page 12: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 12

So What About Attribution Models?

Top of the food chain? Bottom Feeding?

Page 13: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 13

“Attribution Models”

• What exactly are attribution models?

• What is the difference between attribution models and mix models?

A) “Marketing mix models are top down approaches and attribution models are bottom up”

B) Attribution models are mix models that haven’t been properly validated

C) There is no difference

Page 14: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 14

What is Attribution Modeling?

•  “Digital attribution — the measurement of the value of each digital marketing contact that contributed to a desired outcome.” IAB

•  “The measurement of the value of each digital marketing contact or exposure that contributed to an outcome.” Forrester Consulting !

How$is$that$anything$different$than$marke3ng$mix$or$other$techniques?$

•  “Marketing Mix Modeling is a top down approach, while Attribution Modeling is a bottom up approach.” Conference on Attribution Modeling

NOT$TRUE…$Consumer$Level$Marke3ng$Mix$Models$have$existed$for$over$a$decade!$

Page 15: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 15

Attribution Modeling…It’s better than last click

Page 16: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 16

Attribution Modeling… But Its Very Assumption Driven

• What would you like to assume? Take your pick!

1. Last Interaction/Last Click Attribution model. 2. Last Non-Direct Click Attribution Model. 3. Last AdWords Click Attribution Model. 4. First Interaction/First Click Attribution Model. 5. Linear Attribution Model. 6. Time Decay Attribution Model. 7. Position Based Attribution Model. 8. Customized/Personalized Attribution Model.

Page 17: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 17

Attribution Modeling: Assumption Driven

“Pre-determined weighted systems lead to arbitrary insights… Because simple attribution models use

predetermined rules to assign credit to every touch points, even advanced systems offer somewhat arbitrary conclusions….

Data-driven attribution ensures impartiality.”

Page 18: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 18

The Bigger Issue with Attribution Modeling

Standard Approach to Measuring the “individual” using cookie level data

Page 19: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 19

Consumer Reality

Home$Devices$

Browsers$Cookies$

Work$

The$Individual$

Page 20: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 20

The History of Data Needs

0 5 10 15 20 25 30

National Monthly Data (1950)

National Weekly Data (1960)

Market Level Weekly Data (1970)

Store Level Weekly Data (1980)

Household Weekly Data (1990)

Household Daily Data (1995)

Consumer Level Data (2011)*

Cookie Level Microsecond Data (1997)

Granularity (Log Space and Indexed to National Monthly Data)

A R

ecen

t His

tory

of D

ata

Col

lect

ion

Granularity (log space)

**Dates based on my memory of events #

Data Nerds got what they asked for: For the first time in the history of data it is too granular.

The Code of Hammurabi (1772 BC)

Page 21: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 21

Misattribution on Audience Verification

1.9

3.3 4.1

1.9 2.5 2.3 2.4

3.7 3.1

2.3 3.0

10.5

12.3

3.2

4.6 3.3 3.5

7.9

4.8 5.7

3.3

11.0

17.3

4.0

7.7 8.0

4.2

9.9

5.0 6.1

Campaign 1 Campaign 2 Campaign 3 Campaign 4 Campaign 5 Campaign 6 Campaign 7 Campaign 8 Campaign 9 Campaign 10

Online Campaign Impression Frequency by ID Technology Browser Cookie Device Level/Metered/LSO InsightExpress Universal ID

Page 22: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 22

Misattribution from Cookie Level Attribution

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Test Control Test Control Test Control

B-Cookie Device Level Universal ID

% E

xpos

ure

Advertising Effectiveness Analysis Contamination

(Illustrative)

The Majority of Attribution Modeling is done on Cookie Level Data not Consumer Level Data…

* Illustrative Example

Page 23: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 23

Thoughts on Attribution

It is difficult to distinguish a meaningful methodological difference between most “Attribution Models” and Marketing Mix Modeling.

What’s referred to in the marketplace as Attribution Modeling tends to be more dashboard driven, and faster than MMM, but be careful what you ask for:

• Real-time ROI may sound like a good idea, but the problem is that ROI takes time (Often years: See seminal IRI Study “How Advertising Works”)

• Many Attribution Modeling approaches in the marketplace can provide deceivingly plausible insights that may be “arbitrary insights”

• Things to watch out for:

• Fine print assumptions?

• How is it validated?

•  Is it cookie level or consumer level?

• How is the software constraining?

• How is sales data being integrated? Consumer level or aggregate?

• The best attribution modeling techniques are actually converging on and rediscovering long standing Marketing Mix Econometric Methods and Experimental Design methods

Page 24: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 24

The Experimental Design Renaissance

Page 25: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 25

The Experimental Design Renaissance

History: Before there were Mix Models and Attribution Models the business of attribution was done using Test vs. Control Experimental Design (AKA the Scientific Method)

• Historic Weakness(in marketing research):

• Cost

• Scale (can’t measure when always on everywhere: TV)

• But the weaknesses are going away:

• Automation, cost of data acquisition

• Big Data – Shift from Market Level to Consumer Level data

• The big advantage of Experimental Design is that it lets the data speak for itself:

• Far fewer assumptions, and its Causal

• Not black box => very transparent

• Easy to understand

• Experimental Design measurement is now re-emerging for virtually every marketing tactic!

Why Assume When You Can Know!

Page 26: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 26

Closing Thoughts

• The marketplace accepted best ways to do proper attribution remain Econometric and Experimental Design driven methodologies

• Experimental Design is the scientific gold standard for data driven proper attribution. It requires the fewest possible assumptions. “When Life is on the Line use Experimental Design”

• Proper Econometric methods like Marketing Mix Modeling are great and generally been validated, often using experimental designs, which is one of the reasons they are trusted by the market

•  “Attribution Models” are certainly better than last click attribution, but be careful; what’s under the hood matters and methodology remains the wild west.

Page 27: The Marketing Mix Crisis, Attribution Models & The ...nymetro.chapter.informs.org/prac_cor_pubs/04-2014 Jerome Shimizu... · The Marketing Mix Crisis, Attribution Models & The Experimental

Confidential and Proprietary – Not for Public Distribution – Do Not Copy © 2014 InsightExpress 27 May 1, 2014

Thank You Jerome Shimizu SVP, Chief Research Officer (203) 406-3218 [email protected]


Recommended