My talk at CDO Vision on the tools needed to understand the places where your customers engage, and the techniques needed to move them forward in the buyers journey.
2. 2
Ensuring consistency of message across channels
Using online data to optimize the offline experience
What will be critical in
shaping your digital
marketing strategy over
the next 5 years?
Source: Adobe / EconsultancyMarch 2017
Optimizing the customer journey across multiple touchpoints
Learning new techniques disciplines and skills for new channels
Understanding how mobile users research and buy products
Understanding when/where consumers use different devices
Using offline data to optimize the online experience
70%
66%
58%
50%
46%
45%
39%
Big Brand Marketing Priorities
A Quest For The Right Message, In the Right Place
3. 3
0 5 10 15 20 25 30 35 40 45 50
None of these
Managing health records online
Playing a game online
Using a social networking app
Banking online
Making a purchase from a retail website
Online Activities
Shown: % Prioritize Function over Privacy
EU US
• 57% of online shoppers are
comfortable sharing
information as long as it is
for their benefit
• 64% of respondents said
they’d prefer the
personalized experience
• 73% of consumers prefer to
do business with brands
that use personal
information to make their
shopping experiences
more relevant
Data, Personalization vs. Privacy
Source: eMarketer April 2016
Consumers Accept The Risks (if done respectfully)
7. 7
Deterministic
• Device matching relies on user
information captured during login
events
• 95+% certainty
• Proprietary relationships with a
telecom providers, publishers,
retailers, that provide anonymized
bridging data to identify the same
consumer across multiple devices
• Limited by scale, often walled
Probabilistic
• Begins with a “truth set”
• Lower confidence score
• No reliance on cookies / PII
• Better over time and with larger sets
• Statistical analysis through:
• Device Proximity Data
• Browsing Patterns
• Time-Based Clues
ABC’s of Cross Device Targeting
DEVICE ASSOCIATION GRAPH
8. 8
Before Using Audience Data
Remove Redundant “Labels to Minimize Noise
Audience data
• Data tends to be binary in nature – either a user
possesses a given label, or does not possess a
given label.
• Large numbers of such labels. Thousands, tens
of thousands, or hundreds of thousands are
common.
• Diversity and quantity of of attribute data,
demand distilling the essential structure of
Internet audiences into coherent and simple
clusters with similar attributes.
(5% blocked by users) (45% blocked by users)
9. 9
Remove Redundancy before Creating “VisitorID’s”
Source: Clustering methods. (From Python documentation)
• Most methods require an additional
estimation procedure, to find the
optimal clusters count.
• Some are non-deterministic and use
random initialization or other
randomization during optimization
procedure (2 different runs with same
input parameters can return different
results)
• As big data optimization complexity
increases dramatically, so in practice
we will get different results often.
Recommend 2 stage clustering
• Non Stability - a small change in input data can bring about a dramatic change in final results.
Data can change for reasons like, acquiring data, data obsolescence, or using a subsample..
10. 10
Website
Analytics
Mobile
In-App
Media
Touch
points
-Offline
-CRM,
-Loyalty
-Email
Device Association Graphs Redshift
Test Ground for
Attribution Models
Customer
Data
Platform
Dynamic
Creative
Media
Endpoints
Offline
Attribution
Web
Content
Email
Marketing
Data Layer Aggregation Campaign Activation Execution
Retargeting
Content
Personalization
“Look-a-Like”
Audiences
DMP
Behavioral /
Demographic
Facebook
Custom
Audiences
Processing / Segmentation
Google
-Segmenting
-Clustering
-Visit Stitching
Batch or API
Enterprise
Tag Management
(Data Layer)
Customer Data Platform
1st Party Activities 3rd Party Activities
11. 11
1
Done by CDP: Algorithms stitch
together unique “Visitor ID’s” to
create a unified view of
customer included map any
associated device graphs
2 3Visitor profiles are established
that include interest, propensity,
devices, & channel preference
Bayesian models continuously
monitor performance,
evaluate which ads need to
be optimized out
Done through partnerships (DMP):
Rules engine displays advertising
copy based on “Visitor ID” of
customer A/B Testing of Ads
(Optimizely).
JavaScript aggregates &
standardizes data coming from
html pages and mobile apps.
Done by ETM: Metadata for urls,
visitors, timestamps, from
different vendors are formatted
and sent to system of record.
Creating And Serving the Unified View Of the Customer
DATA
TAG & CAPTURE SERVE & OPTIMIZESEGMENT & STITCH
Holistic Omni-Channel Visitor Stitching, Targeting & Personalization