Application of Statistical

Modelling in Marketing and

Advertising

Mindshare Business Planning

7-Nov-12

“Half the money I spend on advertising is wasted.

The trouble is I don't know which half.”

- John Wanamaker

1895

Cookie Based Attribution Modeling

It is difficult to measure ROI…

Last click is not everything

Integrated thinking rather than “swim lane” thinking

Photography credit

6

Search

Display

Digital analytics need to work with technology

Internet cookies are small pieces of information in text format that are downloaded to your computer when you visit a web sites. With visitor ID, cookie data create a diary of visitor’s online activities (eg. clicks on an ad).

First party cookie: cookies come from web site you visited

Third party cookie: cookies from ad server:

Photography credit

All digital data come from cookies. Cookie = Respondent

7

Chinese Fortune Cookie

• Typical client questions to be addressed by attribution

modeling:

• Which elements have driven the conversion – was it the last

click, the first click or a particular combination of clicks?

• How can I accelerate this journey by improving the cross-

channel efficiency?

Attribution Modeling

Customers interact with online ads multiple times

To create a new presentation based on this template, simply

delete the slides you don’t want and save the presentation

with a new name.

The next few slides provide tips to help you use this template

more easily and avoid editing complications.

Source: Microsoft Atlas 9

Transaction

Attribute sales to every touch points; Not just the last

10

Transaction

Brand

Search Display Aggregator

(price comparison)

Affiliate

LC + 1

Last click

LC + 2

LC + 3

LC + 4

Non-brand

Search

Bubble size shows number of new

customer start their journey;

Line width shows cumulative traffic

over a journey.

Direct click path to drive multiple business

outcomes

11

Purchase online

Cross Purchase

Purchase next time

Active Engagement

Loyalty (active

visit + purchase)

Life time

value

Response to

offline

campaigns

Engagement

Discover synergy between online media

12

Customers travel across search and display are more likely to convert

0%

2%

4%

6%

8%

10%

Search only Search + Display

Journey Conversion Rate

Customer Group

+33%

Case Study: Consumers start journey further away and

finish with brand terms or display

11/7/2012 13

Brand PPC

Generic PPC

Display

Sales

Customers start with brand PPC

or generic PPC then change to

display show the highest

conversion rate – 19%. % inside bubble shows conversion rate of customers travelling on

particular route.

Dig

ital A

naly

tics –

An

Intr

oduction

Application Result 1: Individual media plan could gain

up to extra 20% sales than we thought

-30%

-20%

-10%

0%

10%

20%

30% Journey Sales vs. Last Click Ratio

11/7/2012 14

Journey Sales

> Last Click

Sales

Journey Sales <

Last Click Sales

No difference

Media Campaign Name

Dig

ital A

naly

tics –

An

Intr

oduction

How about mass-media ?

How about offline sales?

Marketing Mix Modeling

A brand’s volume sales constantly changes & it changes for a reason

Marketing Mix Modelling is a mathematical approach that explains how each factor drivers sales & share

Year 2010 Year 2007

What is Marketing Mix Modelling?

Sales (Brand

metrics)

ATL

Yt = b0 + S bk f(X)kt + et

BTL

ƒ Distribution

Others

TV, Internet,

Print, OOH

Price promotion,

Sampling, gift

Number of stores,

sales representative

Category

Dynamics

Economic

Indicators

Innovation

Competition

Seasonality

Dependent Variable Independent variables

=

0

10

20

30

40

50

6-O

ct-

02

3-N

ov-0

2

1-D

ec-0

2

29

-De

c-0

2

26

-Ja

n-0

3

23

-Fe

b-0

3

23

-Ma

r-0

3

20

-Ap

r-0

3

18

-Ma

y-0

3

15

-Ju

n-0

3

13

-Ju

l-0

3

10

-Au

g-0

3

7-S

ep

-03

5-O

ct-

03

2-N

ov-0

3

30

-No

v-0

3

28

-De

c-0

3

25

-Ja

n-0

4

22

-Fe

b-0

4

21

-Ma

r-0

4

18

-Ap

r-0

4

16

-Ma

y-0

4

13

-Ju

n-0

4

11

-Ju

l-0

4

8-A

ug

-04

5-S

ep

-04

3-O

ct-

04

31

-Oct-

04

28

-No

v-0

4

26

-De

c-0

4

23

-Ja

n-0

5

20

-Fe

b-0

5

20

-Ma

r-0

5

17

-Ap

r-0

5

15

-Ma

y-0

5

12

-Ju

n-0

5

10

-Ju

l-0

5

7-A

ug

-05

4-S

ep

-05

2-O

ct-

05

Vo

lum

e U

nit

s

Base TV Online Halo TV Print Radio

Mathematical Model

Nonlinear model

11/7/2012 18

TV

In

cre

me

nta

l V

olu

me

Re

sp

on

se

Threshold

Diminishing Return

Dynamic model

11/7/2012 19

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

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

Le

ve

l o

f A

dve

rtis

ing

Im

pa

ct

(Fu

ll I

mp

ac

t=1

)

Weeks After Advertising Airing

More than a simple linear regression

1. Non-linear model

2. Dynamic model:

• incorporating lags, time (t, t-1…) and flexible adstock to include the media carry-over

effect, post-promotion dip effect etc

• Incorporating wear-in, wear-out and different stage of market development

3. Multivariate model:

• Incorporating multivariate relationship between different outcomes and inputs, e.g.,

TV’s the impact on search and social media, sales’ impact on social media etc

4. Hierarchical model:

• Incorporating hierarchical Bayesian model to address heterogeneity for panel data

11/7/2012 20

An example: Google case

21

Understanding how offline and online work together to drive sales

MindShare First Direct Google Case Study

Google case study: http://services.google.com/advertisers/uk/first-direct-case-study-final.pdf

Search, Display,

Social Media

Search, Display, Social Media

TV, Press, Radio, OOH

Sales

Superior methodology results in superior business results

1. Much more reliable strategic recommendation and higher ROI based on accurate measurement of media effectiveness and ROI

2. Media expert: More actionable recommendation, e.g.,

1. How have cross media (especially TV, digital, OOH, PR) impacted our business? What is the synergy between them? What is the optimal mix?

2. What is the ROI of different TV channels (PSTV, PTV and LTV), and what is the optimal mix?

3. What are the recommended weekly/monthly GRP levels for different purpose? Threshold (minimal)? Sweet spot (between maximum marginal response and maximum ROI)? Saturation?

4. What is optimal mix of GRP level, duration and flight pattern?

5. What is the effectiveness by campaign/creative?

6. What is the halo effect? What is the optimal mix of different sub-brands?

7. How many messages/copies should be on simultaneously?

8. What is the ROI and optimal mix of different types of message (functional vs emotional)?

9. What is the right mix of copy lengths (15’ vs 30’)?

10. What is the impact of competitors? Optimal SOV?

11/7/2012 22

Addressing Full Marketing Mix

BTL

ATL

Distribution

Competition

Seasonality / Climate

How responsive is my brand to price; what will happen if I raise or lower price by 5%?

Which promotions make the biggest difference to my business?

What is my ideal media mix or weight?

Does getting an additional 5% distribution really matter when I have 90% already?

How much of an impact does the time of year have on my sales volume?

Price

Which competitor activities have really affected my brand?

The effectiveness with which we use media can increase by 12% through a

simple reallocation of current budgets

17%

17%

2% 31%

33%

Product 1 Product 2Product 3 Product 4Product 5

Current plan (22.3m RMB in Beijing)

Optimized plan (22.3m RMB in Beijing)

20%

22%

3%

22%

33%

Product 1 Product 2 Product 3

Product 4 Product 5

11.4% increase in

unit sales from

advertising and

12.1% increase in

revenue from media

advertising

Product portfolio optimization

Media mix optimization

4/14/2011 25

TV 15 8%

TV 30 32%

MG 22%

NP 11%

OOH 26%

OL 1%

TV 15 6%

TV 30 33%

MG 15%

NP 14%

OOH 30%

OL 2%

Under the same annual budget the revenue can be improved 5.7%

among the incremental sales generated by media

i) Current spending ii) Suggested spending

+5.7% in incremental

sales by media

Spending optimization

TV 15s -30%

TV 30s +4%

Magazine -30%

+30%

+15%

Online +30%

The media adjustment range is from -30% to +30%

Understanding how investments work differently by

Tier leads to improved planning

26

TV Discounts

Retail Gifting

Re

ven

ue

re

turn

pe

r R

MB

R

eve

nu

e r

etu

rn p

er

RM

B

Re

ven

ue

re

turn

pe

r R

MB

R

eve

nu

e r

etu

rn p

er

RM

B

Beyond ROI

Beyond MMM

1. Computational advertising

• Provide the right product and message to the right consumer in the right time through

the right channel

• Huge data mining, machine learning and optimization problem

• Ad exchange, Real time bidding, Demand side platform

2. Consumer and media research

• Know what consumer really wants and how to contact them

• Conjoint analysis. discrete choice modeling, factor analysis, clustering, latent class

modeling, structural equation modeling, graphical modeling

11/7/2012 28

An example of RTB

30

Email: mliao1@yahoo.com

Weibo: 廖明bluedevil