UN//CLASSIFIED
(TOPIC) MOBILE AD FRAUD WIKILEAKS
EXPOSING THE THREATS
Rev 1.1, 31 July 2017
Presenter: Dale Carr
Email: dale.carr@Leadbolt.com
Position: CEO, LeadboltUN//CLASSIFIED
The Mobile Frontier
Worldwide, $143 billion will be spent on mobile ads this year โ€“ TWICE that of 2015
Dangers From the Front Line
Sources:
Magna Global
Hewlett Packard Enterprises, โ€œThe Business of Hackingโ€, May
2016
Wow, that
was easy!
DSP
Who Lives In This New Frontier ?
Who Is at Risk ?
Everyone
โ€ข Developers &
Publishers
โ€ข Networks, DSPs,
SSPs, Exchanges
โ€ข Attribution
companies &
Solution Providers
โ€ข And of course โ€ฆ.
Agencies,
Marketers &
Advertisers
Threats
โ€ข $$$
โ€ข Resource drain
โ€ข Misdirected focus
โ€ข Business model
alignment
โ€ข Misaligned user
experience
โ€ข Reputation
Mobile Ad Fraud Hotspots
Who Is to Blame ?
The rest
โ€ข The App Stores
โ€ข The Operating System
โ€ข Publishers/Developers
โ€ข Ad
Networks/SSPs/Exchanges
โ€ข Attribution
companies/Solution
Providers
โ€ข Advertisers/DSPs
The bad guys
โ€ข Intentional criminals
Types of Fraud
Areas to consider:
โ€ข Impression // CPM
โ€ข Click // CPC
โ€ข Install // CPI
โ€ข Lead // CPL
โ€ข Injection // Adware
โ€ข Domain Spoofing
โ€ข CMS // Fake Publisher
โ€ข Blending // Audience Extension
Attack Vectors
Non-Human Traffic:
โ€ข Simple Bots
โ€ข Complex Bots
โ€ข Botnets
Human Traffic:
โ€ข Invisible Ads
โ€ข Domain Spoofing
โ€ข Click Spam
โ€ข Click Injection
โ€ข Click Farms
Non-Human Traffic - Bots
Type
โ€ข Simple Bots โ€“ simple scripts that run
from hosting servers with consistent
patterns.
โ€ข Complex Bots โ€“ sophisticated tactics
mimicking normal behavior.
โ€ข Botnets โ€“ array of devices that have been
compromised by bad actors. Send
commands that perform tasks like
โ€˜loadingโ€™ or โ€˜clickingโ€™ on ads or
โ€˜installingโ€™ and โ€˜openingโ€™ other apps
Detection
โ€ข Patterns can be identified then blocked
eg. IDs; agents; known data center IP
addresses.
โ€ข More difficult as less consistent pattern.
Rotating IP; user agents; ids; timings; ctr.
โ€ข Hard to detect and block. When
discovered by law enforcement effectively
shutdown. Patterns can be uncovered by
experts.
Type
โ€ข Invisible ads โ€“ hidden ads with zero
being seen aka Ad stacking
Detection
โ€ข Very low CTR/high CTR with other
characteristics. Detectable using off
the shelf ad verification tools like
Integral Ad Science.
Human Traffic โ€“ Invisible Ads
Type
โ€ข Domain Spoofing โ€“ publishers are
declare their own domain and label
They misrepresent by identifying as
domain. Other cases the publisher
spoofed within the request.
Detection
โ€ข Digging deeper and doing proper
verification and validation of
publishers.
Domain Spoofing
Type
โ€ข Click Spam โ€“ clicks, clicks and
generated in hope of โ€œwinningโ€ or
install.
Detection
โ€ข Low/fixed conversion rate
โ€ข High amount of clicks
โ€ข Patterns in click frequency
Click Spam
Example of Click Spam
Installs
Even distribution (flat lines) over a number of hours
is an indication of spamming activity
Early Installs show a normal pattern
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
Hours
Install Time Analysis
Type
โ€ข Click Injection โ€“ a process on the
to โ€œbroadcast intentsโ€ for app
a click before the app is opened in
the install due to last click
Detection
โ€ข Android only
โ€ข Conversion rates are 100%
โ€ข Low CTIT/MTTI
โ€ข High concentration of installs within
moments of click
Click Injection
CLICK TRACK DOWNLOAD OPEN
NEW CLICK !
Last Click
Wins Attribution based
on โ€œopenโ€ event
Click Injection
Example of Click Injection
Installs
Extremely short install times (within seconds) indicate an injection pattern
Install Time Analysis
Seconds
Type
โ€ข Click Farms - a large group of low-
are hired to click on paid
Detection
โ€ข Very difficult as visitors are real
โ€ข Repeated patterns
Click Farms
Example of How Itโ€™s Done
Prevent Mobile Ad Fraud In The First Place
Detection:
Look for Deviations from Patterns / Baselines
โ€ข Establish a baseline; Compare activity against it.
โ€ข Flag/report any abnormalities, discrepancies asap
Monitor and measure everything
Notice Installs from Suspicious Sources
Assign an Internal Stakeholder
โ€ข Dedicated internal person to look at performance
Prevention:
Buy Direct When Possible
โ€ข Direct relationships improve quality and transparency
Buy Premium
โ€ข Cheap traffic comes at a high price
Partner Up
โ€ข Ad fraud security services
โ€ข Measurement + attribution partners
Research
โ€ข Reputations matter
Protect Yourself
โ€ข Contract that outlines what you will/will not accept
Prevent Mobile Ad Fraud
THANK//YOU!
QUESTIONS
Dale Carr
E: dale.carr@leadbolt.com
THE//END

Mobile Ad Fraud Wikileaks: Exposing the Threats

  • 1.
    UN//CLASSIFIED (TOPIC) MOBILE ADFRAUD WIKILEAKS EXPOSING THE THREATS Rev 1.1, 31 July 2017 Presenter: Dale Carr Email: dale.carr@Leadbolt.com Position: CEO, LeadboltUN//CLASSIFIED
  • 2.
    The Mobile Frontier Worldwide,$143 billion will be spent on mobile ads this year โ€“ TWICE that of 2015
  • 3.
    Dangers From theFront Line Sources: Magna Global Hewlett Packard Enterprises, โ€œThe Business of Hackingโ€, May 2016 Wow, that was easy!
  • 4.
    DSP Who Lives InThis New Frontier ?
  • 5.
    Who Is atRisk ? Everyone โ€ข Developers & Publishers โ€ข Networks, DSPs, SSPs, Exchanges โ€ข Attribution companies & Solution Providers โ€ข And of course โ€ฆ. Agencies, Marketers & Advertisers Threats โ€ข $$$ โ€ข Resource drain โ€ข Misdirected focus โ€ข Business model alignment โ€ข Misaligned user experience โ€ข Reputation
  • 6.
  • 7.
    Who Is toBlame ? The rest โ€ข The App Stores โ€ข The Operating System โ€ข Publishers/Developers โ€ข Ad Networks/SSPs/Exchanges โ€ข Attribution companies/Solution Providers โ€ข Advertisers/DSPs The bad guys โ€ข Intentional criminals
  • 8.
    Types of Fraud Areasto consider: โ€ข Impression // CPM โ€ข Click // CPC โ€ข Install // CPI โ€ข Lead // CPL โ€ข Injection // Adware โ€ข Domain Spoofing โ€ข CMS // Fake Publisher โ€ข Blending // Audience Extension
  • 9.
    Attack Vectors Non-Human Traffic: โ€ขSimple Bots โ€ข Complex Bots โ€ข Botnets Human Traffic: โ€ข Invisible Ads โ€ข Domain Spoofing โ€ข Click Spam โ€ข Click Injection โ€ข Click Farms
  • 10.
    Non-Human Traffic -Bots Type โ€ข Simple Bots โ€“ simple scripts that run from hosting servers with consistent patterns. โ€ข Complex Bots โ€“ sophisticated tactics mimicking normal behavior. โ€ข Botnets โ€“ array of devices that have been compromised by bad actors. Send commands that perform tasks like โ€˜loadingโ€™ or โ€˜clickingโ€™ on ads or โ€˜installingโ€™ and โ€˜openingโ€™ other apps Detection โ€ข Patterns can be identified then blocked eg. IDs; agents; known data center IP addresses. โ€ข More difficult as less consistent pattern. Rotating IP; user agents; ids; timings; ctr. โ€ข Hard to detect and block. When discovered by law enforcement effectively shutdown. Patterns can be uncovered by experts.
  • 11.
    Type โ€ข Invisible adsโ€“ hidden ads with zero being seen aka Ad stacking Detection โ€ข Very low CTR/high CTR with other characteristics. Detectable using off the shelf ad verification tools like Integral Ad Science. Human Traffic โ€“ Invisible Ads
  • 12.
    Type โ€ข Domain Spoofingโ€“ publishers are declare their own domain and label They misrepresent by identifying as domain. Other cases the publisher spoofed within the request. Detection โ€ข Digging deeper and doing proper verification and validation of publishers. Domain Spoofing
  • 13.
    Type โ€ข Click Spamโ€“ clicks, clicks and generated in hope of โ€œwinningโ€ or install. Detection โ€ข Low/fixed conversion rate โ€ข High amount of clicks โ€ข Patterns in click frequency Click Spam
  • 14.
    Example of ClickSpam Installs Even distribution (flat lines) over a number of hours is an indication of spamming activity Early Installs show a normal pattern 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 Hours Install Time Analysis
  • 15.
    Type โ€ข Click Injectionโ€“ a process on the to โ€œbroadcast intentsโ€ for app a click before the app is opened in the install due to last click Detection โ€ข Android only โ€ข Conversion rates are 100% โ€ข Low CTIT/MTTI โ€ข High concentration of installs within moments of click Click Injection
  • 16.
    CLICK TRACK DOWNLOADOPEN NEW CLICK ! Last Click Wins Attribution based on โ€œopenโ€ event Click Injection
  • 17.
    Example of ClickInjection Installs Extremely short install times (within seconds) indicate an injection pattern Install Time Analysis Seconds
  • 18.
    Type โ€ข Click Farms- a large group of low- are hired to click on paid Detection โ€ข Very difficult as visitors are real โ€ข Repeated patterns Click Farms
  • 24.
    Example of HowItโ€™s Done
  • 25.
    Prevent Mobile AdFraud In The First Place Detection: Look for Deviations from Patterns / Baselines โ€ข Establish a baseline; Compare activity against it. โ€ข Flag/report any abnormalities, discrepancies asap Monitor and measure everything Notice Installs from Suspicious Sources Assign an Internal Stakeholder โ€ข Dedicated internal person to look at performance
  • 26.
    Prevention: Buy Direct WhenPossible โ€ข Direct relationships improve quality and transparency Buy Premium โ€ข Cheap traffic comes at a high price Partner Up โ€ข Ad fraud security services โ€ข Measurement + attribution partners Research โ€ข Reputations matter Protect Yourself โ€ข Contract that outlines what you will/will not accept Prevent Mobile Ad Fraud
  • 27.