4. Sense Networks
From online to real networks?
What’s next?
a network of real places
GPS
LBS
Location
Data
a network of real people
Online data is easy to get, what about the real world?
4
5. Sense Networks
GPS, LBS, and location data
Collaborative Marketing Advertising Search Social
Filtering Recommendation
SENSE NETWORKS
ANALYSIS, NETWORKS OF PLACES & PEOPLE, SEGMENTATION
GPS
LBS
Location
Data
VEHICLES APPS DEVICES CARRIERS
5
6. Sense Networks
The Old View of LBS Data
=
Single Ping @ Starbucks:
No personalization, no targeting
Can’t use a single ping, too much error in space & time…
It’s not just about when and where but also about who 6
7. Sense Networks
The New View of LBS Data
+ = Health & Fitness
Young Adult
Outdoorsy
Location history for understanding & personalization
Store data over space and time to overcome accuracy issues
Lesson: save your LBS data to segment your customers! 7
8. Sense Networks
Who Has Historical Data? User
Verizon T-Mobile
Sprint Vodafone AT&T Device Mfgrs
Carriers Nokia Samsung
Google RIM
Qualcomm
Publicis DATA Newfield
Bank of America WPP Polaris
GloPos Ericsson
Nielson
Google Omnicom TruePosition Alcatel-Lucent
Nokia Siemens Networks
Airsage
Other
Loc-aid Infrastructure
Pinch
Ad/Content WaveMarket 8
9. Sense Networks
Infrastructure Providers
2009 Q1
missing:
small &
enterprise
players:
Newfield,
Airsage,
ZTE, etc.
• nfrastructure looking for new competitive advantages
I
• long with Carriers: looking sources of top line growth via
A
• argeting, advertising, content delivery...
t 9
10. Sense Networks
Carriers and Prepay
• repay carriers know less and less about their customers
P
• ow: Churn management & segmentation w/o billing info
N
• oon: targeting, advertising, content delivery...
S 10
11. Sense Networks
Ericsson Ad Orchestrator
• nfrastructure solution to ad targeting/segmentation…
I
• ow to get the segments?
H 11
12. Sense Networks
MacroSense Segmentation
• ense Networks’ solution:
S
Convert massive per-user call/location data segments
SOURCE MONTHS PINGS USERS
Location & 18 10b 1.4m
4 8b 4m
Call Data 12 2b 4m
Sense Networks MacroSense
MDN WEALTH AGE CHURN TRAVELER
6462123442 200,000 46 6% 30%
Segmentation 9174341434 35,000 43 4% 20%
6468762413 150,000 31 11% 85%
12
13. Sense Networks
MacroSense Architecture
Raw
Features1…. Roller
Customer Normalizer
Call & Features2.…
Location Features3….
Data Features4….
Delta
Features5…
……
Demo
……
Tree Proprietary
Server …… Non-PII Profiles
……
……
User
SIC ……. Learning
Segment
Tree Engine
Output
Server
13
14. Sense Networks
GPS and location data
10+ million devices giving (lat,long,time,acc)… lingua franca14
16. Sense Networks
CitySense: where is everyone
• itysense: real-time density of users at every street corner
C
• oisson models find most active bars/restaurants
P
16
17. Sense Networks
Next: where’s everyone like me
Need to have a network of people
Each dot
is a user
Dot’s color
is user’s
social
cluster
17
19. Sense Networks
Network of People
Hard to say if User A is like User B…
User A User B
… don’t just look if they co-locate physically
… check if they overlap semantically (network of places) 19
20. Sense Networks
Network of Places
is place A like place B?
Look at each place’s Flow, Commerce & Demographics
20
21. Sense Networks
Network of Places: Flow
Look at flow A to B
Markov transition
Apply MVE on graph
Color code
clusters in graph
21
22. Sense Networks
Encoding a Person’s Lifestyle
9 example users’
lifestyle matrices
no PII information
compute pair-wise
similarity from
matrices
=
how much two
people co-locate
semantically
… can then use machine learning to segment, predict, etc. 22
24. Sense Networks
Network of People: Segments
“Young & Edgy”
• Out every night in young,
racially diverse, low income
neighborhoods
“Weekend Mole” “Mature Homebody”
• Out occasionally on • Rarely goes out, typically
weeknights, typically spends nights in mature,
middle-aged, Latino, middle- white, higher income
income neighborhoods neighborhoods
24
25. Proprietary & Confidential
Segmentation: Standard Output
Convert call, location & lbs data into Claritas type segments
e.g. Pepsi wants to send ad to only “Young&Edgy”
Move lbs into established advertising industry…
SEGMENTATION
01 - Millionaires: Millionaires, as the name implies, collects America’s most
successful achievers and old money. It ranks first for both median and per-
capita income, salaries, self and investment income, home values and net
worth, and ranks second behind the Urban Brahmins in higher education and
Raw professional/managerial occupations.
Customer MACROSENSE 02 - Country Clubbers: Country Clubbers are one rung down from the
Millionaires and ranks second on all measures of income and affluence and
Call & third in college and postgraduate education. These expensive suburbs are
packed into our three great metropolitan strips, dubbed “Bos-Wash”, “Chi-Pitts”,
Location and “San-San”, with the bulk (40%) in Bos-Wash. Much of this wealth is new
money, earned and freely spent by captains of business and technology.
Data
03 - Turbo Boomers: Turbo Boomers rank first in the large “baby boomers” age
group of 35-44 and are concentrated in the rapid growth cities of Atlanta,
Washington DC, Dallas, Denver, Los Angeles and San Francisco. They are
heavy hitters, highly educated and employed in executive and professional
occupations ranking second in marriage and fourth for household income.
…
25
26. Proprietary & Confidential
Segmentation: Car Buyer
Recommendation & Marketing based on the Network
Identifying Active New Car Shoppers
High end large cars for Low end small cars for lower
wealthy, middle age families income, middle/older age
with kids consumers
26
29. Sense Networks
Discussion: New LBS Opportunity
• BS and Location more about segmenting customer
L
• tore your data to understand your customer
S
• se MacroSense to convert LBS history into segmentation
U
• nfrastructure (Ericsson, Alcatel, NSN,…): best data sources
I
• arriers (AT&T, Sprint,…): want growth, ads, targeting
C
• dvertisers (WPP, Nielsen,…): want segmentation
A
• ense Networks: segments from location & call data
S
• nables online business, segmentation, personalization
E
…from Offline LBS data