The Dog Ate My Tracking Practical Solutions for Valuing Mobile Traffic Soren Ryherd Working Planet
Jan 28, 2015
The Dog Ate My Tracking
Practical Solutions for Valuing Mobile Traffic
Soren Ryherd
Working Planet
#SMX 12 Jun 2014
Mobile Tracking Challenges
Image property of FollowYu, Inc.
Cross-Device Behavior
Tracking Breaks
• Javascript not supported on all phones
• Cookie persistence varies
• 90% of American adults have a cell phone
• 58% of American adults have a smartphone
• 32% of American adults own an e-reader
• 42% of American adults own a tablet computer
- 2014 Pew Charitable Trust Survey Data
#SMX 12 Jun 2014
Attribution Models – FAIL
because touchpoints cross devices thus breaking connections
Predictive Models (CPA based bidding) – FAIL
because mobile actions may not be tied to marketing source
Media Mix Models – SUCCEED
may be more difficult to create because of quantitative assessment
and holistic view required
Traditional Modeling Breaks Down
#SMX 12 Jun 2014
Media Mix Modeling
A Basic Media Mix Model
Mix Models attempt to predict unknown sales based on known marketing and advertising activity
U = C1*n1 + C2*n2 +C3*n3…+M
U = “Unknown” Monthly Sales
C = Known Channel Sales Activity
n = Channel Influence
M = Sales from additional sources (Word of Mouth, etc.)
Goal: Determine Cause and Effect Relationships Mathematically Based on Variability in Underlying Data
#SMX 12 Jun 2014
Case Study
Even when doing things right…
• Responsive Design
• Tablet Optimized
• Clear Mobile Calls to Action
• Multi-Touchpoint Tracking
• Attribution Modeling
• Profit-Driven Optimization
• (No call-in sales to confuse channels)
Mobile is still a challenge:
Phone-based PPC provides 20% of traffic, but >3% of PPC sales
#SMX 12 Jun 2014
Traditional Metrics Don’t Reveal Issue
0%
10%
20%
30%
40%
50%
60%
70%
80%
Unknown Link/SEO PPC Social
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
Desktop Phone Tablet
Traffic
Orders
By Device By Channel
Data from Oct13 – Jan14
Traffic & Orders
#SMX 12 Jun 2014
Combination of Device/Channel
0%
10%
20%
30%
40%
50%
60%
70%
Unknown Link/SEO PPC Social
0%
10%
20%
30%
40%
50%
60%
70%
Unknown Link/SEO PPC Social
Tablet
Phone
Desktop
Predicted Actual
Orders by Channel
Data from Oct13 – Jan14
#SMX 12 Jun 2014
Forming Hypotheses
Yes, it appears there is a problem in attributing mobile value
Identifying Signs:
• Over 76% of phone sales are not tied to any marketing source
• Phone sales tied to marketing source only occur in single session
• “Unknown” sales bucket is close to 30% of all orders despite multi-touchpoint tracking, with limited external traffic sources
• Orders by device proportional to traffic, but not by channel
Twin Hypotheses:
Phone tracking is breaking across multiple sessions and
Phone-initiated traffic is migrating to desktop/tablet to purchase
…Interestingly, cookie and tracking preservation is better on tablets
than desktop.
#SMX 12 Jun 2014
Understanding upper and lower limits:
•Upper limit bound by sales numbers
•Lower limits bound by performance
Common Sense Predictions
Felix Chien incorrectly-sourced phone orders cannot be more
than 7.6% of total sales under current media mix
Cross device sales (phone to desktop) are not likely to be more than that proportional to traffic by device – in this case a max of 20% of total orders
Potential for testing: Phone-based traffic may be providing 3x more value in phone-based sales than currently tracked to media source, and up to an additional 4.2x in cross-device sales for 7.2x higher potential value than estimated by strict CPA-based bidding
#SMX 12 Jun 2014
Geographic test
Pulse test
Cause & Effect Testing
Period 1 Period 2 Period 3 Period 4 Period 5 Period 6 Period 7 Period 8 Period 9 Period 10
Impressions
#SMX 12 Jun 2014
Adjusting for Time
0 5 10 15
Point of Investment
Sales Per Day From Initial Investment
1 6 11 16 21 26
1 6 11 16 21 26
Pulse length must take into
account sales cycle. If pulse is too short, it is difficult to determine trends because of overlapping “shoulder” data.
90% of sales in first 8 days 3 Day Pulse Cycle
7 Day Pulse Cycle
#SMX 12 Jun 2014
Results
On Off On Off On Off
Unknown
Link/SEO
PPC
Social
• 47% more mobile impressions were delivered during “On” pulses
• “Unknown” sales increased by 28% during “On” pulses, PPC by 10.7%
• Total sales increased by 16.7% for a total PPC spend increase of 6.3%
(Data from Feb 17 – Jun 4 ’14; extrapolated for final period)
Order Volume
#SMX 12 Jun 2014
Results: Valuing Mobile
Before
Phone-based traffic was
20.5% of total PPC clicks
2.5% of total orders were sourced back to phone-based PPC
After
Phone-based traffic is 30.2%
of total PPC clicks
Mix models predict phone-based PPC now responsible for 27% of total orders
Phone-based PPC traffic was being undervalued by >10x
#SMX 12 Jun 2014
Thank you
Soren Ryherd
Working Planet
131 Wayland Avenue
Providence, RI 02906
401-709-3123
WorkingPlanet.com
@sorenryherd
@workingplanet
Thank You