1 of 34

Economics of Digital Ad Identity

Garrett Johnson, Assistant Professor

Questrom School of Business, Boston University

1

garjoh_canuck

2 of 34

About me

Assistant Professor of Marketing, Questrom School of Business (Boston U)

Economics of ad identity expert

Dissertation research
2020 Marketing Science publication

Award-winning researcher examining digital privacy (GDPR) & digital ad effectiveness (ghost ads)

Paul Green Award winner

2

garjoh_canuck

3 of 34

Agenda

What is the value of cookies?
What does advertising look like without cookies?
How can privacy & modern advertising coexist?

Prof. Johnson - Digital Marketing Analytics

3

4 of 34

What is the value of cookies?

(In the status quo)

4

5 of 34

What work do cookies do?

Cookies allow for cross-site, digital ad identity
Digital ad identity powers 3 key value generators:

Targeting

Behavioural targeting including retargeting
Database match
Ad frequency & reach management

Measurement

Ad effectiveness measurement
Cross-site attribution

Automation at scale

Optimization & deployment loop

5

6 of 34

Literature review: Value of a cookie estimates

Study	Data	Method	Outcome	Estimate
Goldfarb & Tucker (2011)	9,596 ad campaigns	Natural experiment (e-Privacy Directive)	User purchase intent (surveyed)	65%
Beales & Eisenach (2014)	2 ad exchanges + "significantly diversified [company] operating multiple Internet-based enterprises"	Regression adjustment	Exchange/ publisher price	>66%^†
Johnson, Shriver, & Du (2020)	Ad exchange (10K+ advertisers, publishers)	Regression adjustment	Exchange price+ Publisher, SSP, DSP, Advertiser	52%
Marotta, Abhishek, & Acquisti (2019)	large, multi-site publisher	Augmented inverse probability weighting	Publisher revenue	4%
Google (2019) (Ravichandran & Korula)	Google top 500 publishers	Experiment	Publisher revenue	52%
UK CMA Report (2020)	Google study's UK users	Experiment +subsampling + imputation	Publisher revenue	70% (Upper bound)

Notes: Value estimates measure loss in e.g. price without a cookie. Industry studies in grey. †Marginal effect estimates for new cookie (Figure A-1).

Studies:

Goldfarb, A. & Tucker, C. (2011). Privacy regulation and online advertising. Management Science.

Beales, J. H. & Eisenach, J. A. (2014). An empirical analysis of the value of information sharing in the market for online content. Technical report, Navigant Economics.

Johnson, G., Shriver, S., & Du, S. (2020) Consumer privacy choice in online advertising: Who opts out and at what cost to industry? Marketing Science.

Marotta, V., Abhishek, V., & Acquisti, A. (2019). Online tracking and publishers’ revenues: An empirical analysis. Working paper.

Ravichandran, D., & Korula, N. (Google 2019) "Effect of disabling third-party cookies on publisher revenue" (Original blog post here)

UK Competition and Markets Authority "Online platforms and digital advertising market study" Appendix F

7 of 34

Ad effectiveness fell 65% after 2002 e-Privacy Directive in EU

Measured using purchase intent through ad surveys

7

Study	Data	Method	Outcome	Estimate
Goldfarb & Tucker (2011)	9,596 ad campaigns	Natural experiment (e-Privacy Directive)	User purchase intent (surveyed)	65%

Strengths	Limitations
Published, peer-reviewed View of "post-cookie" world	No revenue figures 20 years old

8 of 34

Ads without cookies have >66% lower prices

All else equal: User OS, devices, browser, language, media type, ad position, ad size

Older cookies fetch higher prices

Study	Data	Method	Outcome	Estimate
Beales & Eisenach (2014)	2 ad exchanges + "significantly diversified [company] operating multiple Internet-based enterprises"	Regression adjustment	Exchange/ publisher price	>66%^†

Strengths	Limitations
Revenue/prices 2 ad exchanges+	Industry study Attention to detail lacking

9 of 34

Opt-out users have 52% lower ad exchange prices

All else equal: Ad placement, day, hour, user browser & region, past categories browsed

Drop is roughly proportional for: advertiser spend, DSP cut, exchange cut, SSP cut, & publisher revenue

Study	Data	Method	Outcome	Estimate
Johnson, Shriver, & Du (2020)	Ad exchange (10K+ advertisers, publishers)	Regression adjustment	Exchange price+ Publisher, SSP, DSP, Advertiser	52%

Strengths	Limitations
Published, peer-reviewed View of ad tech revenue funnel	Measures opt-out impressions (vs cookieless) 1 ad exchange

10 of 34

Ads without cookies have 11% lower publisher revenues

All else equal: Date, hour, DSP, website, advertiser, user audience
Another model (not included) says 4% lower publisher revenue

Study	Data	Method	Outcome	Estimate
Marotta, Abhishek, & Acquisti (2019)	large, multi-site publisher	Augmented inverse probability weighting	Publisher revenue	4%

Strengths	Limitations
Premium publisher	"Preliminary Draft", not peer-reviewed Raw difference is 37% Ads w/ cookies: $1.18 CPM Ads w/o cookies: $0.74 CPM

11 of 34

Publisher revenue 52% lower without cookies

Measured using an experiment
News publishers worse off: 62% less revenue without cookies

Study	Data	Method	Outcome	Estimate
Google (2019) (Ravichandran & Korula)	Google top 500 publishers	Experiment	Publisher revenue	52%

Strengths	Limitations
Randomized experiment Top publishers	Industry study ("Don't be evil") Some detail, not a lot

12 of 34

Reanalyze Google experiment data
Publisher revenue at most 70% lower without cookies

Removed users without cookies (e.g. Safari & Firefox), which are immaterial for the value of a cookie
Accounted for missing unsold impressions (by adding 0s)

Study	Data	Method	Outcome	Estimate
UK CMA Report (2020)	Google study's UK users	Experiment +subsampling + imputation	Publisher revenue	70% (Upper bound)

Strengths	Limitations
Regulator study Thorough	Cookies only shut off for Google, not the rest (e.g. other DSPs)

13 of 34

Additional evidence

CMA report also cites 3 UK publishers who document [50-60%], [50-60%], & [70-80%] lower revenue from Safari/Firefox users (relative to Chrome or rest) due to lack of 3rd-party cookies
2020 Facebook experiment: "more than a 50% drop in publisher revenue" without personalized ad ranking

Setting: Facebook Audience Network, restricted to mobile app install ads

Miller & Skiera (2017) further show older cookies are more valuable, as they accumulate more information
Bidswitch "Programmatic Insights 2019" report: Ads to matched (identifiable) users have "2x higher eCPMs" than unmatched (non-identifiable) users. Also, Chrome has "1.8x higher eCPMs" than Safari

13

14 of 34

Ad identify increases ad revenue by

2X-3X

(In the status quo)

14

15 of 34

What does advertising look like without cookies?

15

16 of 34

Market will adjust & money will flow—or will it?

Brand advertisers: Cling to premium publishers

Want cross-site reach & frequency

Direct advertisers: Some will reduce spend

Want consumer intent, ad measurement, attribution, & optimization

So, ad revenue pie will shrink, and will be divided more unequally

Case studies of "flight to safety": NYtimes.com increased EU revenue post-GDPR, NPO (Dutch) increased revenue
How does a worse product affect an industry? See post-secondary education revenue post-COVID

16

17 of 34

17

Context

Sign-in

Walled gardens

The birds

18 of 34

Context

Not all context is valuable

General vs. niche content: General & news content does worse (Goldfarb & Tucker 2002; Ravichandran & Korula 2019)
Differences by niche content: The "sperm whale" problem
Valuable niche content may lack sufficient scale to retain value

Benefits publishers that offer scale & brand-safe premium
Context must compete with sign-in & walled gardens

18

19 of 34

Sign-in: Hashed email identifier solutions

Provides identity solution to advertisers

Better: Cross device ID
Worse: Lower scale

Winners / losers based on 1st party data relationships

Publishers: Requires trusted relationship with users
Advertisers: DTC have customer emails, CPG do not
Creates barrier to entry for advertisers & publishers

Tension over coordinating identity solution: More publisher control over audience, less value creation in total
Irony: Cookie (pseudonymous) replaced by e-mail (PII), albeit with consent

19

20 of 34

Walled gardens: the rich get richer

Both will be hurt by limiting 3rd-party conversion tracking

"In the land of the blind, the one-eyed man is king"

Facebook: ~40% of display spending

Cross-device identity
Scale: Facebook + Instagram
Ease of use: Substantial value from long tail advertisers

Google: ~40% of digital ad spending

Cross-device identity among logged-in users
Scale: Search + YouTube + Google Display Network
Multi-product offering: Data-sharing, in-house attribution, one-stop shop for advertisers

20

21 of 34

21

A matter of perspective...

22 of 34

How can privacy & modern advertising coexist?

22

23 of 34

FLOC

TURTLEDOVE

SPARROW

DOVEKEY

PARROT

PTEROSAUR

TERN

PETREL

23

24 of 34

Audience of 1 is not enough

24

25 of 34

Advertisers want to reach audiences

25

26 of 34

Advertisers want to measure clicks

26

27 of 34

Advertisers want to measure conversions

27

28 of 34

Advertisers want to retarget recent visitors

28

29 of 34

Advertisers want to measure reach & frequency

29

2

0

1

2

4

3

0

3

1

2

4

3

0

2

0

3

1

3

1

4

0

3

1

2

0

3

1

4

0

30 of 34

CONTROL

TREATMENT

Advertisers want to run experiments

30

2

0

1

2

4

3

0

3

1

2

4

3

0

2

0

3

1

3

1

4

0

3

1

2

0

3

1

4

0

31 of 34

CONTROL

TREATMENT

Conclusion: Identity solves every use case (!)

31

2

0

1

2

4

3

0

3

1

2

4

3

0

2

0

3

1

3

1

4

0

3

1

2

0

3

1

4

0

32 of 34

TENSION

PRIVACY

Measurement

& Targeting

(De)centralization

Competition

32

33 of 34