This document discusses how Disney-ABC Digital Media brought A/B testing to their entertainment platforms. It provides background on their business and digital products. It then describes how they set up A/B testing in their organization and why it makes sense for entertainment. The document outlines a case study where they tested different variations of their full episode homepage and found that displaying the top current shows increased video starts by 4.94% compared to the control. They projected this small change could result in over $600k additional annual revenue.
1. Bringing A/B Testing to Entertainment
Disney-ABC Digital Media
The Optimizely LA Experience
8.19.14
Proprietary. Do not distribute without permission (Khai Tran).
Khai Tran
Sr. Mgr, Analytics & Customer Insights
Disney-ABC Digital Media
2. Building A/B Testing within Disney-ABC Digital Media
Background on our business
A case study
Final thoughts
2Proprietary. Do not distribute without permission (Khai Tran).
Why A/B test?
3. 3
Disney-ABC Television – WATCH Products & Platforms
Proprietary. Do not distribute without permission (Khai Tran).
Own, operate, and stream full-episodes, short-form, live video content and
games. Serve as a marketing channel for our shows and brands.
What’s our
Business?
Supported
Platforms
Supported
Brands
8. 8
Where A/B Testing Lives in our Org
WATCH Products & Platforms
Product Engineering Video
Programming MarketingAnalytics &
Customer Insights PMO
Data
"
Data Design
Requirements
Tagging
Data Integrity
Analytics
"
A/B Testing
Path Analysis
Case Studies
KPI Reporting
Customer Insights
"
Support Agents
Customer
Feedback
Escalation Mgmt
Voice of Customer
Proprietary. Do not distribute without permission (Khai Tran).
9. Why A/B Testing Makes Sense in Entertainment
9
1. Controls for external factors"
• Traffic is randomly split across test variations during
time period of test"
• Based on real users, not focus groups "
2. Launch and iterate faster"
• Eliminates guesswork in evaluating different options"
• Reduces time and resources "
3. Drive innovation and growth"
• Whether the test wins or loses, it enables to learn more
about your users"
• Speeds up innovation by minimizing risk and fear in
trying out new ideas
Controls for
External Factors
"
Originals vs. Repeats
Show Popularity
Seasonality
Windowing
Competitive Offerings
Macro Factors
Proprietary. Do not distribute without permission (Khai Tran).
10. 10
Driving Actionable Insights – Testing in the Larger Picture
Track Key
Success
Drivers
Analyze
User
Interactions
Segment for
Deeper
Insight
Test &
Optimize
Drive
Actionable
Insights
1. Track key success drivers
• Keep pulse of the business
• See if we’ve made a difference
"
2. Analyze user interactions
• Heat mapping, path analysis,
etc.
• Identify areas for improvement
"
3. Segment for deeper insight
• Sub-groups behave differently
• Solve for key segments
"
4. Test and Optimize
• Form hypotheses and tests based
on insights
• Validate different variations
based on cause-and-effect
Proprietary. Do not distribute without permission (Khai Tran).
11. 11
Full-Episodes Main Page"
"• Top page of our user path"
• High traffic page"
• High potential impact"
• Bounce rate: 20-35%
* Omniture: Jan – Jun 2011, FEP visits only (excluded iPad)
40%
35%
0.2% 1% 1%
0.5%
1%
2% 4%
7%
4%
3%
What’s Interesting
"35% of users clicked “All Shows”
in bottom nav. Why?
"
Insight: Users want to get
directly to the show that they’re
interested in.
ABC.com Case Study – What and How We Tested
12. 12Proprietary. Do not distribute without permission (Khai Tran).
Variation A: "
Control Version"
"
Full-Episode Home Page "
www.abc.com/watch
13. 13
Variation:
Display all shows with logos
"
Hypothesis:
"Will increase video starts because it:
" • Reduces click to all shows
• Easier to access non-recent content
"
Potential Concerns:
• Requires lots of scrolling.
• Choice overload may lead users to leave.
• Increases page load time.
Proprietary. Do not distribute without permission (Khai Tran).
Variation B: "
All Shows with Show Logos
14. 14
Variation:
Display all shows text
"
Hypothesis:
"Will increase video starts because it:
" • Is easier to scan and find shows
• Loads faster, requires less scrolling
"
Potential Concerns:
• Choice overload
• Too much text. Unattractive.
Proprietary. Do not distribute without permission (Khai Tran).
Variation C: "
All Shows with Text
15. 15
Variation:
Display top 8 noncurrent shows with link to “All
Shows”
"
Hypothesis:
"Will increase video starts because it:
" • Highlights most popular Fall shows
• Extends long tail for non-recently aired content
• Distinct from current shows displayed in the
slideshow
Proprietary. Do not distribute without permission (Khai Tran).
Variation D: "
Top Shows Fall Season "
(Noncurrent Shows)
16. 16
Variation:
Display top 8 shows that are in season at time of test
"
Hypothesis:
"Will increase video starts because:
" • These shows account for 80% of current video
views
• Easy to find and access these shows
Proprietary. Do not distribute without permission (Khai Tran).
Variation D: "
Top Shows Summer Season "
(Current shows)
17. 17
4.24% 3.74% 3.93%Video Starts 4.94%
Chance to beat 99.3% 98.4% 97.8% 97.9%
So Which One Won vs. the Control?
A
Control
B
All Shows Logos
C
All Shows Text
D
Top Shows
Noncurrent
E
Top Shows
Current
18. 18
0.5 Days
"• Define goals,
success metrics,
scope, etc.
"
"
2 Days
"• Assess dev needs
• Build test versions
0.5 Day
"• Configure
campaign in
Optimizely
Planning Design
Developm
ent
Campaign
Configura
tion
0.5 Day
"• QA
• Launch test
1 Day
"• Hypothesize test
variations and
design comps
Setup time – 4.5 days
7 Days
"• Advised by
T&T to run at
least 1 week
"
"
Ru
n
A/B
Tes
t
An
aly
sis
2 Days
"• Analyzed real-
time results
throughout
Run + Analysis time – 9 Days
A/B Test Details "
"
• Ran for one week (June 28 – July 5) "
• 660k+ unique visitors"
• 20% traffic allocation "
• Test groups maintained across sessions
A/B Testing Steps and Actual Time Spent
19. 19
"
Even with a slight page tweak, which appears below the fold, we can
produce a sizable revenue lift for the business.
Video Starts
Driven from
Homepage Revenue
146 MM $13.7 MM
+ 7.2 MM + $677 K
Lift from
Test
+ 4.94%
Projected
Contribution
Connect A/B Testing Results to Revenue Impact
Actual numbers have been masked, but results are in the ballpark. This test ran in 2011.
x $0.0938 =
x $0.0938 =
Avg. Ad Revenue
per Start
Before Test
20. 20
Image from Entrepreneur magazine (Dec 2012). http://www.entrepreneur.com/article/224967
Final Thoughts