Econometric Modeling for Search Marketers · Econometric modeling for marketers Dr. Sid Shah |...

35
© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential. #AdobeSummit @drsidshah @HelenMarie21 © 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential. #AdobeSummit 1 Econometric modeling for marketers Dr. Sid Shah | Director, Business Analytics, Adobe Helen Southgate | Online Marketing Controller Strategy & Planning, BSkyB

Transcript of Econometric Modeling for Search Marketers · Econometric modeling for marketers Dr. Sid Shah |...

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21 © 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit 1

Econometric modeling for marketers Dr. Sid Shah | Director, Business Analytics, Adobe Helen Southgate | Online Marketing Controller Strategy & Planning, BSkyB

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

2

Dr. Sid Shah Director, Business Analytics, Adobe [email protected] | @drsidshah

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

3

Helen Southgate Online Marketing Controller Strategy & Planning, BSkyB

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

4

Econometric models specify the statistical relationship between various variables and a phenomenon of interest

What are econometric models?

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

5

Econometrics let you

measure the unmeasurable!

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Econometric

6

Provide predictive insights 1 Answer complex what-if questions 2

Be applied to a wide variety of problems 3

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

A common attribution problem

7

0

2000

4000

6000

8000

10000

12000

0

20000

40000

60000

80000

100000

120000

140000

160000

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

SE

M O

rde

rs

Tv

Sp

en

d

TV Spend vs SEM Orders

TV Spend SEM Orders

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Too Many Moving Parts?

8

No

rma

lize

d

Imp

ress

ion

s Brand Impressions

No

rma

lize

d

Imp

ress

ion

s Fitness Impressions

No

rma

lize

d

Sp

en

d

TV Spend

Nov Dec Jan Feb

Nov Dec Jan Feb

Nov Dec Jan Feb

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

TV spend vs. SEM orders: The econometric analysis

9

Build

Model

Build deterministic econometric model Explicitly state assumptions

Apply Econometrics

Remove seasonal effects (de-trending) Statistically connect impressions with TV spend

Predict

Predict impressions as a function of TV spend Quantify impact of TV on search

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

10

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

The model

11

Assumptions • Additive linear model (OK for small changes) • Generic impressions (fitness) represents the seasonal trend • Brand keywords are usually at position 1 and hence bid is

relatively constant

𝛽𝐵𝑖𝑑𝐵𝑟𝑎𝑛𝑑 = 𝐼𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛𝑠𝐵𝑟𝑎𝑛𝑑@𝑃𝑜𝑠1 = 𝐶 𝑓 𝑆𝑒𝑎𝑠𝑜𝑛𝑎𝑙𝑖𝑡𝑦 ~𝐼𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛𝑠𝑁𝑜𝑛 𝐵𝑟𝑎𝑛𝑑

𝐼𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛𝑠𝑆𝑒𝑎𝑟𝑐ℎ_𝐵𝑟𝑎𝑛𝑑 − 𝑓 𝑆𝑒𝑎𝑠𝑜𝑛𝑎𝑙𝑖𝑡𝑦 = 𝛼𝑆𝑝𝑒𝑛𝑑𝑇𝑉 + 𝐶

𝐼𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛𝑠𝑆𝑒𝑎𝑟𝑐ℎ_𝐵𝑟𝑎𝑛𝑑 = 𝑓 𝑆𝑒𝑎𝑠𝑜𝑛𝑎𝑙𝑖𝑡𝑦 + 𝛼𝑆𝑝𝑒𝑛𝑑𝑇𝑉 + 𝛽𝐵𝑖𝑑

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

12

Someone told me that

each equation I included in the book

would halve the sales

Stephen Hawking

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Detrending

13

Ind

exe

d C

ha

ng

e t

o W

ee

k 0

Brand Impressions Change

Ind

exe

d C

ha

ng

e t

o W

ee

k 0

Seasonality

-20%

0%

20%

40%

60%

80%

100%

120%

140%

0%

5%

10%

15%

20%

25%

30%

35%

40%

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

Ind

exe

d C

ha

ng

e t

o W

ee

k 0

TV Attributable Change

Tv Attributeable Change Tv Spend*Data presented is for exposition only

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Regression

14

y = 1E-06x + 1.1351 R² = 0.8628

0

0.5

1

1.5

2

2.5

3

3.5

0

20

0,0

00

40

0,0

00

60

0,0

00

80

0,0

00

1,0

00

,00

0

1,2

00

,00

0

1,4

00

,00

0

1,6

00

,00

0

Imp

ress

ion

Vo

lum

e J

um

p

Media Spend

Impression Volume on Brand terms vs. TV Spend

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

TVs effect on search impressions

15

0

8%

21%

31%

38% 44%

49% 53%

57% 60%

63% 65%

0%

10%

20%

30%

40%

50%

60%

70%

TV Spend

TV Attributable SEM Brand Impressions

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Predict

16

0

500

1,000

1,500

2,000

2,500

3,000

0 5,000 10,000 15,000 20,000 25,000 30,000 35,000

Ord

ers

SEM Budget

SEM Performance for different TV Budgets

TV Spend Baseline TV Spend 4x

TV drives awareness which drives impressions and raises the SEM curve

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Multiplicative models for multi channel marketers

Sophisticated multi channel models

Account for multiplicative channel effects (synergies)

Assume non linearity ( Concave/sigmoidal response)

Account for time lags

Account for feedback mechanisms in inputs and responses ( Vector Auto Regressive Models)

17

𝑶𝒖𝒕𝒄𝒐𝒎𝒆 = 𝐵𝑎𝑠𝑒𝑙𝑖𝑛𝑒 ∗ 𝐷𝑟𝑖𝑣𝑒𝑟1𝛼1 ∗ 𝐷𝑟𝑖𝑣𝑒𝑟2

𝛼2 … ∗ 𝐿𝑎𝑔 𝐹𝑎𝑐𝑡𝑜𝑟𝑠𝛾

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

A Continuous

Process

18

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

19

Econometrics let you

measure the unmeasurable!

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Econometric

20

Provide predictive insights 1 2 Answer complex what-if questions

Be applied to a wide variety of problems 3

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21 © 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit

Adobe Summit BSkyB

UEFA Champions League

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

22

So what is a

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Attribution models

Attribution = a method of assigning credit to a sale across marketing channels

Last click First click Even / linear distribution

Time decay Position based

(First, middle, last)

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Long live the last click model

Just make sure you understand what drives it

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Why do we need to look at attribution?

For our last campaign 83% of orders involved multiple clicks across key digital channels

Single Click 17%

Multiple Click 83%

Nu

mb

er

of O

rde

rs

Number of Clicks

1 2 3 4 5 6 7 8 9 10

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Three key objectives

Identifying inefficiencies

Identifying opportunities

Allocating spend to be most effective

What can attribution analysis tell us?

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Why is it still an unexplored area?

The challenges of attribution analysis C

on

fusi

on

Data Available

Volume of data Cross channel devices Support & knowledge gap

?

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

28

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Earlier participation varies per channel and inclusion of impression data

Channels add value beyond last click

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Affiliate Display Natural Search Paid Search

Not Last Exposure

Last Exposure

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Drill down into affiliate, search term, creative, partner etc.

Conversion

Partner 1

Partner 2

But be careful as within each channel there are variances

Single Click 37%

Multiple Clicks but Pure Channel

16%

Last Click but Multiple Channels

27%

Not Last Click but Involved

20%

Single Click 19%

Multiple Clicks but Pure Channel

10%

Last Click but Multiple Channels

24%

Not Last Click but Involved

47%

Partner 1

Partner 2

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Shows value of search higher up the funnel

Search key word analysis

0

0.5

1

1.5

2

2.5

3

Brand Generic

Ratio of Assist to Every Conversion

0

0.5

1

1.5

2

2.5

3

3.5

Individual keywords ratio of assist to every conversion

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

32

So what actionable insight is there

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Actionable Insight Attribution analysis is pointless unless you do something with it

33

Understand the wider value of some of our marketing partners, particularly in the affiliate channel which has led to differing remuneration models and supported a 5% increase in affiliate channel orders 1

Evaluate the cost effectiveness of generic search terms in terms of the customer purchase journey. Increase / decrease spend on

2 customer journey across all channels 3

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Customer is the heart

Beyond online attribution

© 2013 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.

#AdobeSummit @drsidshah @HelenMarie21

Dr. Sid Shah [email protected] Twitter: @drsidshah LinkedIn: Siddharth Shah

Helen Southgate Email: [email protected] Twitter: @HelenMarie21 LinkedIn: http://www.linkedin.com/in/helensouthgate