Panagiotis Papadopoulos
FORTH-ICS, Greece
Nicolas Kourtellis, Evangelos P. Markatos
The Cost of Digital Advertisement:
Comparing User and Advertiser Views
Advertising
"Advertising is an audio or visual form of marketing communication
that employs an openly sponsored, non-personal message to promote
or sell a product, service or idea"
William J. Stanton. Fundamentals of Marketing. McGraw-Hill (1984).
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Ad-slot
• Publisher provide a slot in its medium:
• radio/TV: time slot
• newspaper: ad-slot in paper
• website: ad-slot frame
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Same pattern:
• advertiser pays a monetary cost
• buys ad-space in a content provider’s medium
• places their advertisement on the consumer’s side
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Traditional Vs. Digital Advertising
Same 3 participants:
1. Advertiser pays to deliver ad
2. Publisher gets revenue and provides space
3. User gets ads inside the received content
Same roles too??
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Digital advertising
• Digital advertising moves towards more personalized ad
delivery
• Advertisers programmatically buy available ad-slots
• based on how well the advertised product matches the profile of
the user.
• Users get content from ad-supported publishers for free!
Do they?Panagiotis Papadopoulos ~ panpap@ics.forth.gr
In Digital Advertising
Same 3 participants:
1. Advertiser pays to deliver ad
2. Publisher gets revenue and provides space
3. User also pays to get ads inside the received content
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
In this paper…
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
In this paper...
1. examine the hidden costs of mobile advertising
Ø for both advertiser and user
2. compare them for the same user profiles
3. how fairly they are shared among the two sides?
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Goal: Transparency in the costs of digital advertising
for both advertiser and user
Large scale analysis
Dataset:
1. 12 months duration
2. 1270 mobile users volunteered
Ø Redirect traffic through our proxy
3. 250M HTTP-only requests
4. 44997 programmatic bought ad-slots
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
OpenDAMP
Open Digital Advertising Measurement Platform*
1. Traffic Classification
2. Network Consumption
3. Approximation of Power Consumption
4. Privacy loss
5. Estimation of the per ad-slot price paid by advertisers
*open-source: https://github.com/panpap/openDAMP
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
The view of the user
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
The view of the user (1/3)
Portion of HTTP requests per content
category for average user.
• Categories:
• Advertising,
• Analytics,
• Social (Facebook, Twitter plugins),
• 3rd Party Content (CDNs, widgets),
• Other
• 23% of HTTP requests associated
with content user is not actually
interested in.
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Actual
Content
The view of the user (2/3)
How much of the downloaded
volume is related to ads?
• Across the year:
8.2% of total transferred bytes
Ad and Analytics related
• Ad volume increased compared
with 6 years ago:
Øfrom 5.6%* to 7.3%
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
*Narseo Vallina-Rodriguez, Jay Shah, Alessandro Finamore, Yan Grunenberger, Konstantina Papagiannaki,
Hamed Haddadi, and Jon Crowcroft. Breaking for Commercials: Characterizing Mobile Advertising. IMC’12
The view of the user (3/3)
Ø power consumption of network component:
• 7.98% related to ad-traffic
Ø mobile device’s battery:
• 10 hours of ad-free browsing
• 9.2 hours of ad-supported browsing.
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Based on the model: Chanmin Yoon, Dongwon Kim, Wonwoo Jung, Chulkoo Kang, and Hojung Cha.
AppScope: Application Energy Metering Framework for Android Smartphone Using Kernel Activity
Monitoring, USENIX ATC’12
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
How to measure Privacy loss?
• Loss of anonymity on the web
• Users have a specific budget of aliases (cookie IDs)
• Cookie Synchronization (CSync): a mechanism to increase user identifiability
After CSync:
user123==userABC
userABC
user123
Leaked cookie IDs
User’s exposure to CSync
• 97% of users with regular
activity on the web affected
Ø( >10 HTTP requests per day)
• median user exposed to 1 CSync
every 140 HTTP requests
Ø(every 3-4 website visits)
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
How much do tracking entities
know about a user?
• Median user:
ØAs many as 22 tracking entities
learned up to 20% of her cookie IDs
Ø3 tracking entities learned up to 40%
of her cookie IDs
The view of the advertiser
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
The view of the advertiser (1/3)
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Programmatic RTB auctions
Ad Exchange (ADX)
Real-time Auction
Website on the
user’s browser
Demand Side Platforms
(DSPs)
Bid Request (+user info)
Bid Request (+user info)
Bid Request (+user info)
1.35 CPM (+impression)
0.95 CPM (+impression)
Available ad-slot
The view of the advertiser (1/3)
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
1.35 CPM (+impression)
0.95 CPM (+impression)
Available ad-slot
Ad Exchange (ADX)
Real-time Auction
Website on the
user’s browser
Demand Side Platforms
(DSPs)
WINNER
A
D
Impression delivery
(i) you won,
(ii) here’s the charge price,
(iii) the impression is
rendered successfully
Winning
notification
Programmatic RTB auctions
The view of the advertiser (2/3)
1. RTB auctions
Ø44997 programmatic bought ad-slots
2. Advertiser’s charge price per ad impression
3. Leverage RTB notification URL (nURL)
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Read more: Panagiotis Papadopoulos, Nicolas Kourtellis, Pablo Rodriguez Rodriguez, and Nikolaos Laoutaris..
“If you are not paying for it, you are the product: How much do advertisers pay to reach you?”. IMC ’17,
The view of the advertiser (3/3)
Cost per user for advertisers to
display ads across the year.
• average cost per impression:
0.9 CPM (Cost per thousand).
• total cost paid by advertisers for
the median user is 22 CPM.
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Comparison of both views
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Comparison of both views
Per ad:
• median advertiser paid 0.00071
Euros
• median user paid 0.0022 Euros
(in downloaded bytes)
Ø Considering cost per byte in Europe
users got charged 3x more than
advertisers!
• median user got cookie IDs leaked
to 3.4 different entities
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Rebalancing costs
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Rebalancing costs
• Unlimited dataplan
• Ad caching when on WiFi (e.g., CAMEO middleware [MobiSys‘13])
• User compensation (e.g., Basic Attention Token of BRAVE Browser)
üusers are rewarded for their attention
• Choose paid, ad-free content
üNo trackers… (hopefully)
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
Panagiotis Papadopoulos ~ panpap@ics.forth.gr
1. Ad- and Analytics- related traffic associated with:
• 19% of the total requests
• 8.2% of downloaded volume
• consume more than 9% of the smartphone’s power
2. 97% of regular users are exposed to Cookie Synchronization.
3. Median user:
üanonymity loss through synced cookie IDs
üpays 3x more to download ads, than
what advertisers pay to deliver them!
In summary

The Cost of Digital Advertisement: Comparing User and Advertiser Views

  • 1.
    Panagiotis Papadopoulos FORTH-ICS, Greece NicolasKourtellis, Evangelos P. Markatos The Cost of Digital Advertisement: Comparing User and Advertiser Views
  • 2.
    Advertising "Advertising is anaudio or visual form of marketing communication that employs an openly sponsored, non-personal message to promote or sell a product, service or idea" William J. Stanton. Fundamentals of Marketing. McGraw-Hill (1984). Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 3.
    Ad-slot • Publisher providea slot in its medium: • radio/TV: time slot • newspaper: ad-slot in paper • website: ad-slot frame Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 4.
    Same pattern: • advertiserpays a monetary cost • buys ad-space in a content provider’s medium • places their advertisement on the consumer’s side Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 5.
    Traditional Vs. DigitalAdvertising Same 3 participants: 1. Advertiser pays to deliver ad 2. Publisher gets revenue and provides space 3. User gets ads inside the received content Same roles too?? Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 6.
    Digital advertising • Digitaladvertising moves towards more personalized ad delivery • Advertisers programmatically buy available ad-slots • based on how well the advertised product matches the profile of the user. • Users get content from ad-supported publishers for free! Do they?Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 7.
    In Digital Advertising Same3 participants: 1. Advertiser pays to deliver ad 2. Publisher gets revenue and provides space 3. User also pays to get ads inside the received content Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 8.
    In this paper… PanagiotisPapadopoulos ~ panpap@ics.forth.gr
  • 9.
    In this paper... 1.examine the hidden costs of mobile advertising Ø for both advertiser and user 2. compare them for the same user profiles 3. how fairly they are shared among the two sides? Panagiotis Papadopoulos ~ panpap@ics.forth.gr Goal: Transparency in the costs of digital advertising for both advertiser and user
  • 10.
    Large scale analysis Dataset: 1.12 months duration 2. 1270 mobile users volunteered Ø Redirect traffic through our proxy 3. 250M HTTP-only requests 4. 44997 programmatic bought ad-slots Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 11.
    OpenDAMP Open Digital AdvertisingMeasurement Platform* 1. Traffic Classification 2. Network Consumption 3. Approximation of Power Consumption 4. Privacy loss 5. Estimation of the per ad-slot price paid by advertisers *open-source: https://github.com/panpap/openDAMP Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 12.
    The view ofthe user Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 13.
    The view ofthe user (1/3) Portion of HTTP requests per content category for average user. • Categories: • Advertising, • Analytics, • Social (Facebook, Twitter plugins), • 3rd Party Content (CDNs, widgets), • Other • 23% of HTTP requests associated with content user is not actually interested in. Panagiotis Papadopoulos ~ panpap@ics.forth.gr Actual Content
  • 14.
    The view ofthe user (2/3) How much of the downloaded volume is related to ads? • Across the year: 8.2% of total transferred bytes Ad and Analytics related • Ad volume increased compared with 6 years ago: Øfrom 5.6%* to 7.3% Panagiotis Papadopoulos ~ panpap@ics.forth.gr *Narseo Vallina-Rodriguez, Jay Shah, Alessandro Finamore, Yan Grunenberger, Konstantina Papagiannaki, Hamed Haddadi, and Jon Crowcroft. Breaking for Commercials: Characterizing Mobile Advertising. IMC’12
  • 15.
    The view ofthe user (3/3) Ø power consumption of network component: • 7.98% related to ad-traffic Ø mobile device’s battery: • 10 hours of ad-free browsing • 9.2 hours of ad-supported browsing. Panagiotis Papadopoulos ~ panpap@ics.forth.gr Based on the model: Chanmin Yoon, Dongwon Kim, Wonwoo Jung, Chulkoo Kang, and Hojung Cha. AppScope: Application Energy Metering Framework for Android Smartphone Using Kernel Activity Monitoring, USENIX ATC’12
  • 16.
    Panagiotis Papadopoulos ~panpap@ics.forth.gr How to measure Privacy loss? • Loss of anonymity on the web • Users have a specific budget of aliases (cookie IDs) • Cookie Synchronization (CSync): a mechanism to increase user identifiability After CSync: user123==userABC userABC user123
  • 17.
    Leaked cookie IDs User’sexposure to CSync • 97% of users with regular activity on the web affected Ø( >10 HTTP requests per day) • median user exposed to 1 CSync every 140 HTTP requests Ø(every 3-4 website visits) Panagiotis Papadopoulos ~ panpap@ics.forth.gr How much do tracking entities know about a user? • Median user: ØAs many as 22 tracking entities learned up to 20% of her cookie IDs Ø3 tracking entities learned up to 40% of her cookie IDs
  • 18.
    The view ofthe advertiser Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 19.
    The view ofthe advertiser (1/3) Panagiotis Papadopoulos ~ panpap@ics.forth.gr Programmatic RTB auctions Ad Exchange (ADX) Real-time Auction Website on the user’s browser Demand Side Platforms (DSPs) Bid Request (+user info) Bid Request (+user info) Bid Request (+user info) 1.35 CPM (+impression) 0.95 CPM (+impression) Available ad-slot
  • 20.
    The view ofthe advertiser (1/3) Panagiotis Papadopoulos ~ panpap@ics.forth.gr 1.35 CPM (+impression) 0.95 CPM (+impression) Available ad-slot Ad Exchange (ADX) Real-time Auction Website on the user’s browser Demand Side Platforms (DSPs) WINNER A D Impression delivery (i) you won, (ii) here’s the charge price, (iii) the impression is rendered successfully Winning notification Programmatic RTB auctions
  • 21.
    The view ofthe advertiser (2/3) 1. RTB auctions Ø44997 programmatic bought ad-slots 2. Advertiser’s charge price per ad impression 3. Leverage RTB notification URL (nURL) Panagiotis Papadopoulos ~ panpap@ics.forth.gr Read more: Panagiotis Papadopoulos, Nicolas Kourtellis, Pablo Rodriguez Rodriguez, and Nikolaos Laoutaris.. “If you are not paying for it, you are the product: How much do advertisers pay to reach you?”. IMC ’17,
  • 22.
    The view ofthe advertiser (3/3) Cost per user for advertisers to display ads across the year. • average cost per impression: 0.9 CPM (Cost per thousand). • total cost paid by advertisers for the median user is 22 CPM. Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 23.
    Comparison of bothviews Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 24.
    Comparison of bothviews Per ad: • median advertiser paid 0.00071 Euros • median user paid 0.0022 Euros (in downloaded bytes) Ø Considering cost per byte in Europe users got charged 3x more than advertisers! • median user got cookie IDs leaked to 3.4 different entities Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 25.
  • 26.
    Rebalancing costs • Unlimiteddataplan • Ad caching when on WiFi (e.g., CAMEO middleware [MobiSys‘13]) • User compensation (e.g., Basic Attention Token of BRAVE Browser) üusers are rewarded for their attention • Choose paid, ad-free content üNo trackers… (hopefully) Panagiotis Papadopoulos ~ panpap@ics.forth.gr
  • 27.
    Panagiotis Papadopoulos ~panpap@ics.forth.gr 1. Ad- and Analytics- related traffic associated with: • 19% of the total requests • 8.2% of downloaded volume • consume more than 9% of the smartphone’s power 2. 97% of regular users are exposed to Cookie Synchronization. 3. Median user: üanonymity loss through synced cookie IDs üpays 3x more to download ads, than what advertisers pay to deliver them! In summary