What is Attribution?
This presentation will address the misconceptions about attribution solutions, and clarify how all-inclusive the term truly is. It will also cover how companies that implement proper attribution models can not only determine the most effective mediums in their campaigns, but can also establish behavior patterns in their consumers that will allow marketers to optimize their strategies, reallocate budgets as needed, and ultimately increase their ad spend to drive the highest possible ROI. The presentation will also include real world examples and stats that explain how these companies are benefiting from attribution.
2. What is attribution?
• Marketing activities = marketing touch-points
• Desire outcome = sale, subscription
2
3. Does it work?
"A lot of that is around being able to allocate our
spend where it's most effective."
3
4. Will it really change my decisions?
• Google AdWords Keyword: Job Positing Sites
• Last click attribution:
CPA $378 & ROI -84%
• Algorithmic attribution:
CPA $28 & ROI +229%
4
5. Inadequate Attribution Models
A simple customer path
TV Commercial
Mobile Search
Laptop Coupon Site
Tablet Website
Laptop Website
Desktop Website
Laptop Search
o Last-click: All exposures, except Coupon Site, assigned no value
o Last-device: TV, Mobile, Tablet, Desktop exposures assigned no value
o Digital-Media only: TV in this case, assigned no value
5
6. Complexity – Offline Tracking
• Offline Marketing spend & In-Store Sales
Online
Marketing
Touch point
35% Spend
TV
Direct Mail
Print
Radio
65% Spend
= Tracking
= Missing
In-Store Sales
& Phone Orders
65% Sales
User Visits
Website.com
User Closes Online
35% Sales
8. Complexity – Multi Device
Households have 5.7 Internet Connected
Devices on average1
1
Source: https://www.npd.com/wps/portal/npd/us/news/press-releases/internet-connected-devices-surpass-half-a-billion-in-u-s-homes-according-to-the-npd-group/
9. Case Study: Cross-device tracking, Indochino
Problem: Unsure if mobile campaigns were producing results
Result: Quantified how mobile campaigns were driving desktop sales
1. Discovered that
of multi-device users were
switching from mobile to desktop before converting
2. Optimized mobile based on overall results
“When we initially decided to
implement this, we were looking
forward to improving the quality of
our data, which we certainly did. The
additional reporting and device data
was a welcomed bonus! ”
9
13. Fixing GIGO - Cookies 1st vs. 3rd Party
• Third-party cookie is usually set by an
analytics vendor.
• First-party cookie is set in-house.
• First Party Cookies are regarded as the most
reliable method to measure visitor activity
Source: http://www.ogilvydma.com/2011/03/glossary-ana
13
14. Fixing GIGO -3rd Party Cookies Bad
• 3rd Party Cookies subject to 30% Greater
Cookie Deletion
1st Party Cookies
3rd Party Cookies
15. Fixing GIGO – Use 1st Party Cookies
• Make sure your attribution solution
uses:
JavaScript tag that serves a 1st party
cookie on your site
1st Party View Pixel to track display
impressions
15
16. Fixing GIGO - Device Fragmentation
• Email Address = The New Cookie…
• User ID
= The New Cookie
16
17. Fixing GIGO – User ID Solution
PII
PII
User ID
Name
email
ft@gm.com 6805
6805
6805
6805
Laptop
iPad
iPhone
PII
Fred
User ID
Device
17
18. Fixing GIGO – Cross Device 1st Party Data
1. Transaction – Get User ID
2. Login – Get User ID
3. ESP – Transactional Email – Get User
ID
18
19. Cross Device 2nd Party Data
• Convertro 2nd party client device mappings
• DoubleClick Android + G+ + Gmail
• Atlas Facebook data
19
20. Solving Offline Tracking
• Offline Marketing spend & In-Store Sales
Online
Marketing
Touch point
35% Spend
TV
Direct Mail
Print
Radio
65% Spend
= Tracking
= Missing
In-Store Sales
& Phone Orders
65% Sales
User Visits
Website.com
User Closes Online
35% Sales
21. Offline Step 1 – Identify File
Leads file tied to PII
1 In-Store Sales, Catalog, Direct Mail, & Soon Print
Email
john@gmail.com
jane@gmail.com
Name
Postal Address
3 1st Street, San Francisco, CA
John Doe 94105
520 Powell Street, New York, NY,
Jane Doe 10001
Order ID
In-Store Sale
1111
$440
1112
1113
1114
$133
$410
$640
22. Offline - Step 2 – Move File
Send file to LiveRamp
2
Secure FTP
Email
john@gmail.com
jane@gmail.com
Name
Postal Address
3 1st Street, San Francisco, CA
John Doe 94105
520 Powell Street, New York, NY,
Jane Doe 10001
Order ID
In-Store Sale
1111
$440
1112
1113
1114
$133
$410
$640
23. Offline - Step 3 – Activate
Associate purchase data with cookies
3 LiveRamp provides masked Walmart lead information to Convertro tied
to cookie IDs
Convertro Cookie ID
11212
12312
41928
123912
Order ID
1111
1112
1113
1114
In-Store Sale
$440
$133
$410
$640
24. Offline - TV Method #1 – Set-top Box Fusion
• Use set-top box data to understand Brand commercial
viewership
• Match set-top box data to look-a-like cookies
• Sync cookies w/ LiveRamp
• Import sources into Attribution Tool at user level
24
29. Offline - TV Drag Effect
Response Density Index by hour since TV Airing
0.6
0.5
0.4
0.3
0.2
0.1
0
0
Sunday
1
Monday
2
Tuesday
Wednesday
3
Thursday
4
Friday
Saturday
30. Offline Case Study: Dollar Shave Club
Problem: Where to profitable acquire customers on TV for online business
Result: TV works, but only on specific channels with specific creative
1. Reduced cost per spot by
2. Expanded its TV budget by
30
31. Attribution Model Evolution
• Single Click Rules Based - First click / Last Click
• Multi-Click Rules Based - U-Shape or Even Click
• Algorithmic Models
31
32. Convertro Algoritm
Probability of Conversion
1
0.9
Logistic Function
Events
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1
Exposure to Marketing
32
33. Convertro Algoritm
In Matrix Format (Example: Four
Unique Paths)
Y
Conversion & Frequency
Counts of Paths
X
Prob Conv
Convs
Total
Org.
PPC
TV
Email
Disp.
1%
10
1000
1
0
1
1
0
5
100
0
1
1
0
0
10%
10
100
1
0
0
0
1
3%
30
1000
0
1
1
1
0
5%
1
=
i
Resulting Weights
.10
.20
.30
.10
.30
33
34. Advanced Modeling: Event Chaining
• Addresses repeat purchases and multiple steps in
conversion funnel
• Attribute preceding events
First conversion gets credit for the next conversion
34
35. Advanced Modeling: Source Decay
• Estimate period over which source exposure effect
decays over time
• Plot for views shows that < 25% of conversions
occurred by end of 1 day
35
36. Model Validation
• Randomly splitting user click trails into training and
test sets (80/20 split)
• Train a set of models on the training set
• Then Run a prediction on the previously-unseen test
sets (with the actual conversion events removed
from the clicktrail).
44. Key Attribution Takeaways
• Be Holistic – Track online offline touch-points &
events
• Avoid attribution solutions that rely on 3rd party
cookies
• Leverage a User ID to track cross-device & have
more persistent tracking
• Use algorithmic attribution model
44
45. Key Attribution Takeaways
• Validate that your attribution model is accurate
against your data
• Attribution data is useless without cost (need CPA &
ROI)
• Feed attribution data into programmatic buying tools
• Determine optimal cross-channel spend allocations
• Determine what intra-channel optimizations need to
be made
45
46. Thank You!
NYC: 11 West 42nd Street
New York, NY 10036
HQ:
1453 Third St Promenade
Santa Monica, CA 90401
(888)308-9896
perez@convertro.com
www.convertro.com
46
Notes de l'éditeur
Redo to be left-right and less dense
Bigger problem than last click is “last device”
40% of touchpoints are in mobile but only 15% of conversions happen on a mobile device