RTB Demystified: Machine vs. Human, Live vs. Static Data, 1st vs. 3rd Party
1.
2. Demystifying RTB
Machine Learning vs.
Human Optimization
Trading Desks
DAA
Dynamic Targeting
Live vs. Static Database
Dynamic Algorithms
Search Retargeting
1st Party Data vs. 3rd Party Data
Private Bidders
Click Fraud
NAI
Consoles
DAAC
Keyword Targeting
Click Bots
Brand Safety Organic Data
Remarketing
3. Demystifying RTB
Machine Learning
Human Optimization
vs.
Live Static Database
vs.
1st Party Data
vs.
3rd Party Data
Click Fraud (Death of the Click)
Brand Safety
5. Machine vs. Human Optimization – Case Study
Beauty and the Beast Theatre Production submitted 492 keywords for search retargeting,
which resulted in a CTR of 0.030% (way below average of 0.4%)
Human Analysis indicated that 67% of all points were generated by generic terms shows,
march break & gifts which were diluting the effectiveness of the segment.
*eyeDemand Word Cloud
6. Machine vs. Human Optimization – Case Study
Removing generic keywords and inserting new more specific words - reducing the
total search terms to 207 - resulted in a CTR of 0.57% - an increase of 1800%.
*eyeDemand Word Cloud
7. Human Optimization
Analyze Brand Safety
Web Site Analysis
Content Quality,
demographic data,
language
Base success on conversions,
not clicks
Creative evaluation
Time-relevant search terms
(blocking and targeting)
Evaluate performance of machine
learning
Monitor click fraud
Automated filters can be helpful
but constant human monitoring
required to catch new bots
vs.
Machine Learning
Multi-Variable Analysis
Ad type/Time of day
recency and frequency
creative format
performance
Automated Optimization to
various KPI’s
Algorithms
Monitor click fraud
Automated Filters based on
human learning
Our Recommendation…choose
the cyborg approach, best of
both worlds!
9. Live vs. Static Databases
89 campaigns spanning a one year period (April 2012 – March 2013) for 3 major automotive brands
Conversion Rate
Live “Organic” Data
Static Data
Time
10. Live vs. Static Databases
89 campaigns spanning a one year period (April 2012 – March 2013) for 3 major automotive brands
Cost Per Acquisition
$18.00
$Cha-Ching! $
$16.00
$14.00
$12.00
$10.00
$8.00
$6.00
$4.00
$2.00
$-
Live "Organic"
Data
Static Data
11. Live Data – Retargeting Study
Vertical: Automotive
Campaign Type: Retargeting
Campaign Goal: CPA
•
Behavioural targeted campaign with increasing weight shifting towards “live data” retargeting
Retargeted Users and Cost per Acquisition
Acquisition (conversion) is measured by 3 KPI’s:
requesting a meeting
asking for a quote
virtually creating a custom product
Retargeted Site Stickiness
Percentage of visits to relevant pages (conversion
pages as defined by customer) as a percentage of
homepage visits
12. Live Data vs. Static Data
To Buy or To Build? That’s the big question with data!
Live Data is by far the richest source of information for your
brand and client.
On average, Live Data is 47% more cost effective* when
judging performance within campaigns.
The downside is that reach can be limited when only looking
at Live Data.
Our recommendation is to use a combination of Live Data,
Contextual Data and other relevant 1st party data, thereby
reaching both potential new and current customers.
14. 1st Party vs. 3rd Party Data
Every company defines data differently
1st party data
$$
…they even know who has
the remote in their hand.
15. 1st Party vs. 3rd Party Data
Other companies request personally identifiable information
when signing up for a free email account or free services
…this is 1st party data
$$
Make sure on our push for
privacy – we understand the
landscape.
17. Click Fraud (Death of the Click)
In a 2012 study entitled “Evaluating and Optimizing Online
Advertising: Forget the Click, But There are Good Proxies”
researchers found that:
"Generally, clicking does not correlate well with purchasing.…few
clicks lead to actual product purchases and that most clicks are
generated by a very small subset of browsers.
…clicks are unsuitable as a criterion for designing and optimizing
targeting strategies…Targeting models built on clicks do a poor job
of identifying browsers who later will purchase”
*Evaluating and Optimizing Online Advertising: Forget the Click, But There are Good Proxies
Brian Dalessandro, Rod Hook, Claudia Perlich, m6d research and Foster Provost, NYU/Stern School of Business and m6d research
18. Click Fraud (Death of the Click)
Click fraud comes in many forms; but these issues can affect all
online advertising campaigns!
Bots generated by a fraudulent website
Layered ad units (companies that put 40 ad units in one ad spot
and then place them in reputable web sites)
Incentivized click schemes (some benefit to a user for clicking on
an ad)
Malware on a real user’s computer
Some pop-up blocking software (attempts to click on all ads on a
page while it loads to test for pop-ups)
19. Click Fraud – Prevention
Optimize on CPA vs CPC or CTR
CPC algorithms can actually target ads towards click fraud activity
Employ brand safe rules (next section)
Include Targeting
Majority of fraudulent clicks are found in campaigns without additional
targeting criteria (mass reach /RON)
Demand Transparency
20. Click Fraud (Death of the Click) – Case Study
Vertical: Automotive
Campaign Type: Performance
Campaign Goal: CPA
Campaign required large reach at low cost, while targeting smaller markets like the Prairies.
When comparing vendors on hits to the landing page (ie – clicks), eyeDemand performance questionable.
However when we analyze the results on the relevant conversion pages (dealer located or test drive booked
following car build) eyeDemand becomes the vendor with the highest conversion rate!
Click Through Rate
Total Relevant Conversions
Conversion Rate - Relevant Action
21. Consoles & Proprietary Bidders – Death of the Click
Client:
Click Through Rate
Agency: Gaggi Media
Vertical: Entertainment
Campaign Type: Performance
Campaign Goal: CPA
Both systems employed segment targeting, site
retargeting & RON.
Cost Per Acquisition
Console impressions resulted in 300% better
performance when analyzing CTR and CPC
metrics.
But, the CPA from the Private Bidder was
150% lower!
Cost Per Click
23. Brand Safety
With access to so much inventory, how do you control the setting
in which your message is displayed?
1
Use exchange console filters
CONSOLE
FILTER
30. Summary
Bottom line: It’s tricky! You can’t afford to wait to find that an
algorithm has failed!
Push for transparency
Employ multiple brand safety methods depending on
campaign goals
Work with an expert in the field that can adequately
monitor exchanges & audit content to ensure optimal
brand results
Measuring success only by click must die!