1. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com
They’re NotWrong,
They’re Just Not Right
Classic Attribution Approaches
Are No Longer Enough
Pavan Korada, SVP Global Data Science
2. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com | 2
Pavan Korada
SVP, Data Science &
Analytics
o Analytics & Data Science practitioner for 15
years
o Drives marketing ROI for Fortune 100 brands
via the strategic application of modeling,
inference and optimization
o 5.5 years at Zeta
3. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
3
The Deterministic
Attribution Difference
o Personal-Level Identifiers
o Non-cookie-based identifiers that
accurately quantify true audience ROI
across channels.
o Incremental measurement
o Heart of deterministic media mix
modeling approach
o Controlled Experimentation
o Quantifies the true impact od
campaigns in a biased-free manner
4. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
Last Click Disaster
Under Prioritization
Fail Explanatio
n
Top Measurement Fails!
I attribute conversations to the last thing that happened and I ignore/under
invest in what came before
Attribution is not a top priority for the organization
It’s Broken, but We Don’t Fix It We know our attribution doesn’t work but…
Politics, Politics Attribution results are related to departmental media budgeting and is not fully
objective
Can’t Pixel I can’t get the right tracking/optimization infrastructure in place on my site
Can’t Share Conversation Data I can’t provide access to the right types of conversation data for attribution
Cookies Not People I attribute to the device not the person
Siloed Channel Measurement
My measurement doesn’t account for online & offline marketing & online &
offline conversations in a unified manner
Black Box I don’t really understand my attribution how my attribution works
Aggregate Level Only I can’t access raw event level data related to
attribution
Insights for Insights Sake My attribution insights are not actionable in media
5. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com | 5
When people see an ad, what percentage of the total effect on sales can be
attributed to the selection effect, and how much is due to the advertising effect?
Study 1 (Retail)
Study 2 (Bank)
Study 3 (E-commerce)
Study 4 (Retail)
Study 5 (E-commerce)
Study 6 (Retail)
Study 7 (E-commerce)
Study 8 (Tech)
Study 9 (E-commerce)
Study 10 (Retail)
Study 11 (E-commerce)
Study 12 (E-commerce)
Study 13 (Retail)
0
%
60% 80% 100%40%20%
Advertising effect Selection effect
Bron:Gordon et al. (2018) LocalFocus
6. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
Attribution Methods
Accurate and transparent measurement is a competitive advantage for brands
o Front-end metrics (CTR) are not
strongly correlated with conversion
o Overly conservative methods (last-
click) disincentivize scaling media
and inhibit growth
o Inaccurate methods (view-through)
waste media dollars
o Black box models can be gamed by
media providers
o Online-only methods omit critical
offline media & conversion data
7. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
MMM and MTA: They’re Not Wrong, They’re Just Not Right
7
o Macro-level analysis performed 1-2 times a year.
o Analyzes aggregated historical data from offline
sources to deliver planning, spending and
performance metrics. Considers internal and
external factors, like marketing and pricing data,
market elasticity, seasonality and weather and
news events.
Multi-Touch Attribution
o Analyzes individual users’ digital journeys and the
path to conversion across multiple touch points.
o Measures the impact of each individual marketing
tactic using data science to turn account-specific,
near-real-time data into insights about how
marketing initiatives are performing at each level.
These and other attribution methods only measure growth
responses…
Marketers Must Achieve Incrementality, But
How?
Multi-Touch Attribution
8. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
Incrementality Is Achieved Through Test and Hold-Out Groups
8
9. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
Data Cloud Signals and Deterministic Measurement Approach
to Generate Massive ROI
Segment Performance Summary
(All ROI metrics measured for campaign flight during 11/02-12/31)
10. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
It’s easy to do this on single-channel and
email, but the need of the hour is to do
this across Programmatic and Walled
Gardens.
You must have an Identity Graph to do
this!
11. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
It Starts with Our Vast Zeta Data Cloud
11
750M+ Permissioned
Users with Identity-
Based Data
600M+ Global
Connected IDs
150M Upvotes &
Comments per
Month
2,500+ Data
Elements per User
3.4BN Unique
Users Monthly
800M+ User Signals
Processed Daily
500+ Intender
Audience Segments
1 Trillion+ Monthly
Content Consumption
Signals
100B+ Bidstream
Signals Monthly
3B location visits
Monthly
Data Sources
Proprietary
+ Disqus Network
+ O&O Properties and
Sources
+ Partnerships with Digital
Publishers
Strategic Partners
+ TransUnion
+ Acxiom
+ Liveramp
+ Affinity Solutions
+ PlaceIQ
High-Volume Data-In
Zeta Data Cloud Assets
+ Identity-based Profiles
with Scoring
+ Behavioral Signals
+ Environmental Signals
+ Transactional Signals
+ Transportable Data
Assets
+ Onboarding and
Distribution Pipelines
High-Value Audiences Out
13. 13
A Robust Test and Learn Methodology is Used to
Drive Omnichannel Performance
o Zeta’s omnichannel execution is
designed to flight channels:
o In the order of their ability to drive
conversions and
o In a manner that allows for optimal
channel-specific impact
measurement
0.009%
0.010%
0.007%
0.011%
0.006%
Total
ESD
ES
ED
E
Incremental Policy Rate
Channel Email Social Display
Control
Single ✓
Multi ✓ ✓
Multi ✓ ✓
Multi ✓ ✓
Omni ✓ ✓ ✓
Simplified Test Design Matrix (example)
14. 2019 Zeta Global – Proprietary & Confidential www.zetaglobal.com |
o Conversion around brand KPIs
o Split into TRT and CTR
o Split TRT into equal budget groups
o Run channel-specific campaigns with
similar creatives and messaging for a
specific period of time
o Calculate lift in conversion
and brand KPIs
o Normalize incrementality
measurements for KPIs of interest to
calculate optimal channel mix
Steps to Determine the Right Media Mix
Incremental Lift on
Conversion Metric
This slide needs to show this equation:
Score = - 0.911891 + 0.095298 (Education_Group:1-College) + 0.150753 (Education_Group:2-HighSchool&Unknown)
A re-articulation of the Deterministic Analysis build/path
A better way to show the metrics in the bottom slide
Right now, many marketers determine their media mix ahead of a campaign with constraints on each device and letting them run independently, without optimizing across the various devices. [CLICK]
But with Moment Scoring™, you can let the AI determine the right media mix. The real magic is that it is constantly changing and adjusting because it’s not about the device - it’s about the moment.
So over the length of the campaign, if we could look into some different points in time, you could see how the media mix is updated in real time. One day, tablet might be doing best so the biggest spend is there. Then, a few days later, desktop might be best.
It’s incredible to know that we no longer have to think in silos anymore.
In the Data Cloud , there is high volume data coming in
Disqus – enables companies to facilitate conversations across discussion boards
Able to understand behavioral interests
Within our network we see . . .
High value audiences out
To date 550 audiences that we've developed
To help you better understanding our segments, we'll share with you how we create new taxonomies and the data points we leverage to really understand our customers