Presentation emailteam
Transcript of Presentation emailteam
Mobile/Desktop User Analysis on Email Interaction Patterns
Outline1. Data and Preliminary Observations2. Link Click Distribution Analysis3. Section Click Distribution Analysis4. Analysis on different section type, size and
position5. Analysis on the same section format6. Summary
Outline1. Data and Preliminary Observations2. Link Click Distribution Analysis3. Section Click Distribution Analysis4. Analysis on different section type, size and
position5. Analysis on the same section format6. Summary
What kind of dataEvent Type Data Source Notes
Email Send Event/data/tracking/EmailSendEvent
Get the detailed structure send Emails like link, type, position.
Email Click Event/data/tracking/EmailClickEvent
Get Email click details like userid, device
Email View Event/data/tracking/EmailViewEvent
Get Email view details like userid, device
Preliminary Observations
• From previous data, we can know:
• 1.There are at most 34 links in the one email.• 2.These 34 links can be grouped as at most 7
sections.• 3.which user click on which email and which specific
links in the email.• 4.I count the click distribution on different links and
link position for both mobile and desktop• 5.We only focus on “digest email” in this analysis.
Outline1. Data and Preliminary Observations2. Link Click Distribution Analysis3. Section Click Distribution Analysis4. Analysis on different section type, size and
position5. Analysis on the same section format6. Summary
2.Link Click Distribution
Group the links by Section• We actually interested in the section in the email.• I group the 34 links. • There are at most 7 sections in one email.
SectionNo SectionName0 positions1 milestones2 shares3 profile4 endorsements
5 connections6 pymk
Outline1. Data and Preliminary Observations2. Link Click Distribution Analysis3. Section Click Distribution Analysis4. Analysis on different section type, size and
position5. Analysis on the same section format6. Summary
3.Section Click Distribution
Some section can be missing
SectionPos SectionName0 positions1 milestones2 shares3 profile4 endorsements
5 connections6 pymk
Original 7 sections
SectionPos SectionName
0 positions1 milestones2 shares
4 sections
are missiong
Section Click Distribution
Bias on previous analysis
• 1. On same position, section name is different• 2. One section name can be in different position
SectionPos SectionName0 positions1 milestones2 shares3 profile4 endorsements
5 connections6 pymk
Original 7 sections SectionPos SectionName
0 positions1 milestones2 shares
SectionPos SectionName
0 positions1 shares2 profile
SectionPos SectionName
0 profile1 endorsements2 connections
4 sections is
missiong
4 sections is
missiong
4 sections is
missiong
Bias on previous analysis
• 3. Section size is also different.
SectionPos SectionName0 positions1 milestones2 shares3 profile4 endorsements
5 connections6 pymk
Original 7 sections
SectionPos SectionName
0 positions1 milestones2 shares3 profile4 endorsements
SectionPos SectionName
0 positions1 milestones2 shares3 profile
SectionPos SectionName
0 profile1 endorsements2 connections
2 sections are
missiong
3 sections
are missiong
4 sections are
missiong
Outline1. Data and Preliminary Observations2. Link Click Distribution Analysis3. Section Click Distribution Analysis4. Analysis on different section type, size and
position5. Analysis on the same section format6. Summary
Bias• So we should consider differences:
1. Section name2. Section size3. Section position
• Original order is not optimal
• 1.As the section type increases the Uctr decreases.
• 2. 3Profile’s Uctr is higher than 2shares with section size 3 on pos 0.
• 3. The Desktop’s Uctr is higher than the mobile’s Uctr.
• Original order is not optimal
• With same section size and same section pos. Pymk and connection’s Uctr are higher than the endorsement
For endorsement:1. As the section
position increases the Uctr drops.
2. As the section size increases the Uctr drops.
3. Top position is important
Bias• Bias: For section endorsements with section
size 3, and section pos 1. we don’t know what is in the pos 0 and pos 2.
• For example:• We can have two different format:
1, shares;endorsements;pymk2, shares;endorsements;connections
Outline1. Data and Preliminary Observations2. Link Click Distribution Analysis3. Section Click Distribution Analysis4. Analysis on different section type, size and
position5. Analysis on the different section format6. Summary
Uctr Difference between setion name on Same Format
Format section name Uctr
shares;endorsements shares 0.070833333
shares;endorsements endorsements 0.027083333
shares;endorsements;connections shares 0.067493113
shares;endorsement;connections endorsements 0.022956841
shares;endorsements;connections connections 0.02892562
shares;endorsements;pymk shares 0.073609732
shares;endorsements;pymk endorsements 0.025819265
shares;endorsements;pymk pymk 0.02599861
Pymk and connection’s Uctr are higher than edorsement.Pymk increases the Uctr for the first section ‘shares”.
Maybe A better orderSectionNo SectionName
0 positions
1 milestones
2 profile
3 Shares
4 Pymk
5 connections
6 Endorsements
SectionNo SectionName
0 positions
1 milestones
2 shares
3 profile
4 endorsements
5 connections
6 pymk
Summary• 1. First link is very important, since it actually
contains more than 50% of all the clicks.• 2. The section order that we have now is not
optimal.• 3. On the Mobile data, the click distribution for
the first position is higher than the Desktop data, but the click distribution on mobile drops faster than the desktop data from the first position to the second position.
• Thank you!• You can find more detailed analysis on
Email User Analysis Wiki• Go/bomaEmailAnalysis
Pymk increases Uctr on first section
Format section name Uctr Format section name Uctr increase rate
endorsements;connections endorsements 0.069423175
endorsements;connections;pymk endorsements 0.071578619 3.01%
connections 0.032272702 connections 0.030748472 -4.96%
pymk 0.021006685
profile;endorsements profile 0.140718563
profile;endorsements;pymk profile 0.174781765 19.49%
endorsements 0.041916168 endorsements 0.027158099 -54.34%
pymk 0.025800194
• 1.As the section position increases the Uctr decreases.
• 2.6Pymk and 5connection’s Uctr is higher than the 4endorsements.
• This shows us the original order is not optimized
Example of email
module = endorsement
module = connections
module = pymk