People Based Measurement Attribution in Automotive
1. People Based Measurement
Christian Bartens Thomas Gage
Managing Director, Datalicious Gage Consulting
November 8, 2017
Automotive Attribution Summit
Boca Raton, Florida
4. Confidential and Proprietary5
How are my
media partners
performing?
How can I
optimize cross
channel tactics?
What media
mix is optimal?
What incremental
return am I getting?
17. Confidential and Proprietary18
Solution: People-Based Measurement
Purchase Path
§ Challenges related to building an exhaustive quality purchase path
that measures real people interacting across channels and devices
Purchase Attribution
§ Challenges related to establishing and distributing
the value that real people generated over time
TV Display Dealer Print Search Website Leads Sales Profit LTCV
$500 $2,000 $1,500 $4,000?% ?% ?% ?% ?% ?%
18. Confidential and Proprietary
The future of measurement
is identity management
§ To facilitate the stitching of touch
points across platforms, devices, etc
People can easily amass
hundreds of IDs across time
and different ad platforms
Deterministic and probabilistic
user ID matching is crucial
19
Purchase Path: People Have Many IDs
PII IDs
CRM ID
Google
Cookie ID
Facebook
People ID
iOS ID
Android
ID
IP
Address
[...]
19. Confidential and Proprietary20
Purchase Path: People Use Multiple Devices
Consumers increasingly use multiple devices and apps along the purchase path
This requires different measurement strategies to ID consumers across devices and apps
so individual touch points can be combined into one single cross-device purchase path
Datalicious study revealed that combining paths across devices lengthens the path by
38%, attributes 1.8x more sales to mobile equaling +142% percent for Facebook ads
TV Display Dealer Print Search Website $2,000
Device 1 Display
Device 2 Search Website $2,000
20. Confidential and Proprietary21
Purchase Path: People Use Multiple Devices
Consumers increasingly use multiple devices and apps along the purchase path
This requires different measurement strategies to ID consumers across devices and apps
so individual touch points can be combined into one single cross-device purchase path
Datalicious study revealed that combining paths across devices lengthens the path by
38%, attributes 1.8x more sales to mobile equaling +142% percent for Facebook ads
TV Display Dealer Print Search Website $2,000
Device 1 Display
Device 2 Search Website $2,000
Source http://data.li/2krpbPF
21. Confidential and Proprietary22
Purchase Path: People Still Visit Dealers
Physical branch interactions are still important touch points
Challenge is sourcing and then connecting the location data to the path
Various (mostly still untested) possible approaches
§ Walk-in counters and inclusion via Media Mix Model (MMM)
§ Latitude/longitude tracking via mobile ad servers, browsers or mobile apps
§ Proximity tracking or triangulation via in-store free Wi-Fi or Bluetooth beacons
Problem remains how to connect location data set to main path data set
§ Solution will be a combination of PII to device advertising ID and IP address matching
TV Display Print Search Website $2,000
Dealer
22. Confidential and Proprietary23
Purchase Path: People Inside Walled Gardens
Purchase
Path
Google
Facebook
iOS
[…]
Walled gardens opening
§ Facebook Clean Rooms, APIs
§ Google Ads Data Hub
§ Access via trusted partners
such as Datalicious, Equifax
Walled gardens closing
§ Apple iOS Tracking Prevention
– Technical R&D challenge
23. Confidential and Proprietary24
Purchase Path: People Inside Walled Gardens
Purchase
Path
Google
Facebook
iOS
[…]
Walled gardens opening
§ Facebook Clean Rooms, APIs
§ Google Ads Data Hub
§ Access via trusted partners
such as Datalicious, Equifax
Walled gardens closing
§ Apple iOS Tracking Prevention
– Technical R&D challenge
Source http://data.li/2kqPAwR
24. Confidential and Proprietary25
Purchase Path: People Interaction Quality
Just because ad impressions were served doesn’t mean …
§ They were not blocked – Ad blocking
– eMarketer reports that ad blocker penetration reaches 30-35% in the US
§ They were actually viewed – Ad view-ability
– Marketing Charts reports ad view-ability to be between 45-60% in the US
§ They were actually viewed by a human person – Ad fraud
– Datalicious study suggests twice as high as thought or 15-20% in the US
Filtering out ineffective ads can reduce ROI by >100%
TV Display Dealer Print Search Website $2,000
25. Confidential and Proprietary26
Purchase Path: People Interaction Quality
Just because ad impressions were served doesn’t mean …
§ They were not blocked – Ad blocking
– eMarketer reports that ad blocker penetration reaches 30-35% in the US
§ They were actually viewed – Ad view-ability
– Marketing Charts reports ad view-ability to be between 45-60% in the US
§ They were actually viewed by a human person – Ad fraud
– Datalicious study suggests twice as high as thought or 15-20% in the US
Filtering out ineffective ads can reduce ROI by >100%
TV Display Dealer Print Search Website $2,000
Source http://data.li/2yL25FG, http://data.li/2krKlNl
26. Confidential and Proprietary27
Purchase Attribution: People Convert Offline
What if all research and ads are online but the sale offline?
§ Reality is that online drives offline and vice versa
§ Cannot just attribute sales to online we can track online
Important to match offline sales to online campaign activity
§ Requires PII of sales to be matched to cookies and IP addresses
§ Combined cookie and IP address match rate between 50-60% for US
TV Display Dealer Print Search Website
$2,000
27. Confidential and Proprietary
Purchase
Path
Equifax
Consumer
Universe
Dealer
Car Sales,
Leads
3rd Party
Car Sales,
Registrations
Cleaning,
Matching,
Removing PII
Deterministic
Cross-Device
ID Matching
Economic,
Media Etc
Segments
Website,
App Events
VINs
PII
VINs
PII
Offline Person ID
Anonymized PII
Attribution
Modelling
Probabilistic
Cross-Device
ID Matching
Online Person IDs
Match Partner IDs
IP to ZIP+4
IP to Lat/Long
Timestamp
Offline Person ID
ZIP+4
Reports,
Dashboards
28
Cookie IDs
Device IDs
IP Address
Match Partner IDs
Advertising
Events
Cookie IDs
Device IDs
IP Address
Match Partner IDs
Ads SalesAttribution
Offline Person ID
Household ID
28. Confidential and Proprietary
Purchase
Path
Equifax
Consumer
Universe
Dealer
Car Sales,
Leads
3rd Party
Car Sales,
Registrations
Cleaning,
Matching,
Removing PII
Deterministic
Cross-Device
ID Matching
Economic,
Media Etc
Segments
Website,
App Events
VINs
PII
VINs
PII
Offline Person ID
Anonymized PII
Attribution
Modelling
Probabilistic
Cross-Device
ID Matching
Online Person IDs
Match Partner IDs
IP to ZIP+4
IP to Lat/Long
Timestamp
Offline Person ID
ZIP+4
Reports,
Dashboards
29
Cookie IDs
Device IDs
IP Address
Match Partner IDs
Advertising
Events
Cookie IDs
Device IDs
IP Address
Match Partner IDs
Ads SalesAttribution
Offline Person ID
Household ID
50-60%
90-100%
25-35%
29. Confidential and Proprietary30
What? You Cannot Match 100% of Sales?
Source http://data.li/2iAet4Y
Gallup
500/Survey/Day
Pew Research
1,500/US Survey
Say there are 40m
car sales per year
in the US of which
~17m are new car
sales so 25-35%
of that would be
10-14m sales
matched per year
or 27-28k per day
– a huge sample!
32. Confidential and Proprietary33
Purchase Attribution: Allocating Sales Credits
TV Display Dealer Print Search Website Leads Sales Profit LTCV
0% 0% 0% 0% 100% 0% $500 $2,000 $1,500 $4,000
20% 25% 25% 25% 25% 0% $500 $2,000 $1,500 $4,000
30% 10% 10% 10% 40% 0% $500 $2,000 $1,500 $4,000
A% B% C% D% E% F% $500 $2,000 $1,500 $4,000
Many different out of the box models but all wrong …
Only accurate way is a custom model …
33. Confidential and Proprietary34
Purchase Attribution: Model People’s Behavior
Hidden Markov Modell (HMM) proven to be most effective
§ Uses visible events to approximate hidden customer journey states
§ Does not make unrealistic assumptions about consumer habits
§ Accounts for negative experiences or barriers that detract people
34. Confidential and Proprietary35
Seems Complicated – Why Bother?
Because Forrester
suggests – and we can
confirm – that unified
marketing measurement
and resulting optimization
can easily yield 10-30%
media efficiencies
§ I.e. sell the same amount
of cars with 10-30% less
media spend
Source http://data.li/2krDzas
35. Confidential and Proprietary36
Reinvested Savings From Waste Drive Growth
Ads Sales Ads Sales Ads Sales
Spend Waste Reinvest Sales Incremental
Current Attributed Optimized
Waste Reinvest
Grow
36. Thank You
Questions
Christian Bartens Thomas Gage
Managing Director, Datalicious Gage Consulting
November 8, 2017
Automotive Attribution Summit
Boca Raton, Florida