Finding Customer Signals –Powering Advanced Analytics with Person-Based DataJohn Gim
SVP, Advanced Analytic Solutions
David Popkin
Head of Data Strategy for Brands
102
David PopkinHead of Data Strategy,
Brands
John GimSVP of Advanced Analytics
About Us
103
◦ Recap: The Cookie Problem
◦ Identity Resolution Unlocks Advanced Analytics
◦ The Transition from Segment to Signal
◦ Case Study – Identify Predictive Signals by Connecting Online & Offline Data
Agenda
104
COOKIENc92 87bc
COOKIE5Bmy 76Bc
Challenge: Devices
105
IDFAHa27 0285
COOKIENc92 87bc
COOKIE5Bmy 76Bc
JOHN SMITH42 Greenwood Ave
106
Attempts to measure impact make the limitations apparentKNOWN UNKNOWN
Social• Liked on FB• Tweets about
experience
See online or mobile ad • Impressions
and clicks• Publisher site• Campaign
Search• Search words• Referring URL• Site Activity
Email Capture• Open and clicks• Campaign• Refer a Friend
Purchase• Name, Address• Email Address• Products
Service Call• Phone Number• Issue• Disposition Status
Received Mailer• Birthday reminder• Offer tailored to me
See TV Ad• Impressions• Show• Campaign
UNIFIED VIEW OF CUSTOMER ACTIVITY
107
108
Omnichannel customer experience is the priority
3%
6%
10%
11%
40%
7%
14%
16%
18%
14%
7%
14%
17%
19%
9%
0% 10% 20% 30% 40% 50% 60% 70%
Programmatic buying
Mobile
Cross-channel marketing
Data-driven marketing
Customer experience
Please rank these five areas in order of priority for your organization in 2017 first
second
Econsultancy / Adobe Digital Intelligence Briefing
109
What challenges senior marketers
46%Integrating online and offline
“We’ve had to deploy and change our ways to make sure that we were present where the customer was getting inspired, because inspiration doesn’t always happen in the physical confines of a store.”
President of Online
US Retailer
51%UnderstandingCross-device behavior
Source: Forrester's Q1 2016 Cross-Channel Campaign Management Survey
48%TransformingReal-time interactions
BI Director
Global Bank
“Our interactions are mostly in the services area, but we’re looking to mine that data for cross-selling.”
Ninety-four percent of marketing professionals report that lacking “people-based” marketing capabilities contributes to their struggle with creating a complete view of cross-media exposure. Without this ability to link that complete view marketers aren’t able to accurately access customer insights.
110
94%
111
The Reality: No one has a complete picture
AGENCIES
?
112
Identity Resolution
Age: 21Male
Eli SimonJoe Glass
Elias J Simon143A 55th
New York, NY
Eli Smith(650) 307-2131
E. Simon55 Post
San Francisco(650) 307-2131
Eli Simon10 Main StArkansas
Eli Simon$75, Shoes
Device ID4500d
Cookie ID 5699zk7fb42
Cookie ID 9O582Lk3
ID G59x55Term: skinny jeans
Cookie ID 4978d8lk
Cookie ID: 89f73kpn
Sentiment
Sentiment
Cookie ID 48y9lk76
Cookie ID
Cookie ID
Cookie ID Cookie ID
Cookie ID
impression
impression
IDFA4500d
Identity Resolution
IdentityLink: Xj5m9b
113
Consumers are addressable
Data is actionable Technology is connected
114
Now, a unified customer view is possible
Unified Customer View
115
Accumulate engagement history and data with Identity
Airline Website Search Interest
Frequent Traveler Segment
Hotel Booking Partner Activity
Trigger Data (e.g. Weather)
IdentityFirst Party
Second Party
Third Party
Data Segment: “Golfers”
What type of signals can form the basis of a segment?
116
Carrie made a large purchase at Golf Galaxy
Adam installed RangeFinder
App Rachel watched
The Masters
Ben visits a golf course frequently
Emily subscribes to Golf Magazine
James bought sneakers
Alex installed the ESPN app
Sarah bought a TV
Chris went to Starbucks
Sample high-quality signal data sources available
Location Transaction Online Behavior
117
118
◦ Availability: Through data sourcing and advanced analytics, properly assess which data is available and useful
◦ Transparency: Proactive approach to knowing what data is driving composite scoring and avoiding algorithmic bias and black box solutions
◦ Competitive Advantage: Not relying on the same segment targets that are available to all competitors, but leveraging data associated to known and unknown user base
Root Data For Brands
119
Data Utility, Data Quality, Data SourceThe usefulness of data is highly dependent on the quality, which in turn is highly dependent on the source. With a more digitally connected society, there are opportunities to focus on more factually accurate signal data which can lead to higher data quality and utility.
Data Utility
Variable X is not very
useful
Data Quality
How good is the data?
Data Source
Where is the data coming from?
“The mind knows not what the tongue wants. […] If I asked all of you, for example, in this room, what you want in a coffee, you know what you’d say? Every one of you would say ‘I want a dark, rich, hearty roast.’ It’s what people always say when you ask them what they want in a coffee. What do you like? Dark, rich, hearty roast! What percentage of you actually like a dark, rich, hearty roast? According to Howard, somewhere between 25 and 27 percent of you. Most of you like milky, weak coffee. But you will never, ever say to someone who asks you what you want –that 'I want a milky, weak coffee.’”
– Malcolm Gladwell
The Challenge Of Capturing Actual Customer Preferences
The New Coke is one of the most famous research failures. Despite
thousands of sip tests and countless efforts to fine-tune the
taste based on the customer feedback, the New Coke was a
huge disaster.
The now acclaimed Aeron office chair received very low ratings in early tests.
Despite the ratings, the company decided to go on with manufacturing.
The rest is history: Aeron became one of the most iconic and best selling chairs in
the history of office furniture. And the irony: once the chair became famous, people started rating it much favorably.
It cost Walmart $1.85 billion to listen to what their customers
say. “This is the peril of listening to what your customers say
instead of what they actually did.” - Walmart Declutters Aisles Per
Customers’ Request, Then Loses $1.85 Billion In Sales
121
122
Experiment: Replacing or augmenting historically survey based, non-transactional data points with signal data
The Unsurprising Strength And Usage Rate Of Transactional Data In Statistical Models
52%of predictive modeling variables,
utilized across a variety of modeling types and targets, are derived from
transactional data
25% 11% 12%
Transactional (1st Party)
Demographic (Age, Income, etc)
Lifestyle, Preference
Other (Macro)
123
You know a lot about your customers/outcomesYou send us PII and customer attributes, we strip out PII and replace with IdentityLink. We send you 3rd Party Data tied to the same IdentityLink.
YOUR CRM DATA
FootballFan
Customer who Bought Car
Sports Fan
Browses Real
Estate
Buys Junk Food
Coffee Drinker
Drives a Minivan Customer who
Bought Car
Goes to the Movies
Drives a minivan
Browses Real Estate
Coffee Drinker
Buys Junk Food
124
IdentityLink lets us connect 1st and 3rd party attributes
Term Length Unsubscribed Service Complaint
Competitive Shopping
Drives a minivan
Goes to the Movies
Browses Real Estate
Coffee Drinker
Buys Junk Food Sports Fan
Service Appointment Email Open
125
Data science to determine which signals influence positive or negative known outcomes
Term Length Unsubscribed Service Complaint
Competitive Shopping
Drives a minivan
Goes to the Movies
Browses Real Estate
Coffee Drinker
Buys Junk Food Sports Fan
Service Appointment Email Open
v
126
Online-to-Offline Data Analysis
Customer who Bought Car
MODELING ENVIRONMENTCRM + 3rd Party data both
mapped to IdentityLink
Compare known outcomes from 1st party data to attributes in 3rd party data, all keyed off of IdentityLink. Use data science to determine which 3rd party data attributes are predictive to power future marketing and models.
Goes to the Movies
Coffee Drinker
Sports Fan
Browses Real Estate
Buys Junk Food
Drives a minivan
127
Top 10% more likely to convert
128
Case Study - Quantifiable Assessment Method
Firewalls & Privacy
Privacy Agreements
No 1 to 1 targeting
Technical Restrictions
Signal Data
5,000+ Files
1+ Terabyte of Data for test
7 Direct Sources
Data Value
Evaluating which sources and which data
points are most valuable
Power ScoreUtility in modeling
and prediction
Inform Score
Utility for planning, creative, and
strategic vision
© 2017 LiveRamp. All rights reserved. 129
• Aggregated age/income & other demographic data can fill gaps in standard appends (7-40% of fields can be NULL or unknown)
• Many confirmations and contradictions found across sources for ‘lifestyle’ attributes. TBD on creating an aggregated view across sources for model viability.
Case Study - Assessment Insights
• Aggregated geo data improves estimates of event traffic by understanding ‘true travel radius’ vs a generic radius assumption
• Combination of 3rd party web visits with competitive geo data improves defection/loyalty/imkpredictions where 1st party data is sparse or dated
• App level preferences, usage allow for more granular ‘product recommender’ engines
• Old customer records now have data associations that allow for identification of their journey stage by plotting timestamped data across sources. Primarily led through geo data and web activity.
• Plotting aggregated app-genre connections allow for an understanding of pairing/partnerships across verticals.
Model Improvement Comparative Data Interconnectivity
129
130
Fin.