Schedule

 
Analysis Delivery Date Remark
Clustering done on time 
Contrast Analysis done on time
Survival Analysis done on time
Survey done on time
Path Analysis 8/2 futuristics tools
Fequency and Reach 7/27 regression analysis

 


Contrasts Analysis


1. Weekday only users

About 50% of the traffic by Netscape users (nscp edition) occurs during week days. This percentage is 28% for the home edition and only 4% for clubvaio. We may be able to increase our share of Netscape users visiting us on weekend. The contrast based on time detects similar patterns.

2. Browser

The swbell, pacbell, nscp and bellatlantic editions have less than 5% of traffic due to MSIE. It would make sense to have these Snap editions optimized for use with the Netscape browser. For the home edition, this percentage is 53%. Also, while 54% of traffic on the front door is due to MSIE, this percentage is only 35% for the /stock/quote page. Generally, there is less of MSIE presence in the money category or money pages.

Not surprisingly, only 2% of the users referred by netscape.com use MSIE. This percentage is equal to 61% for unknown referrals. Finally, if we look at the domain name, MSIE amounts to 89% of the traffic from aol.com (not surprisingly), and 3% from pacbell.net.

In conclusion, except for users from the netscape edition, the netscape browser is associated with "better" Snap usage.

3. User Age

By "old user", I mean a user who first visited Snap long before the start of the study. The pacbell, swbell and clubvaio editions have a higher proportion of active older users (about 55-60%, versus 30% for the home edition and 15% for the netscape edition). An analysis based on domain names confirms the high proportion of old users for swbell.net, pacbell.net, and also shows this effect for the cw.net domain.

Neoplanet edition: 13% of the users are "old", and they generate 26% of the traffic from this particular edition; this ratio (26%/13%) is sensibly higher than in other editions.

4. Region

Domain names are of course geographically correlated with regions (West for pacbell.net, North-East for bellatlantic.net). The proportion of users with a valid zipcode is much higher for pacbell.net than for aol.com or att.net.

There are also some big differences showing up in the advertising categories and directories. For the money category, the proportion of traffic known to originate from either N-E or West is about the same for each of these two regions: about 20%. Since we overall have less traffic from the N-E region, it means that people from the N-E are more into money than people from the West - possibly because the business hours for the stock markets are ~ 9-5 EST. We also find:

It seems that we could focus in getting more deals and partnerships with companies located in the North-East (to get more users from the N-E), because currently the West is over-represented.

5. Type of edition

Three types of editions: home, search partners (hmc, nscp, gte) and other partners. I am comparing here search partners with other partners.



Clustering Analysis


We have 15,579 sampled users. The sample has been split into 6 groups,
ranging from the one-time user to the very heavy user.

The sample was created between 3/19 and 3/25. Then, web usage was tracked between 3/26 and 5/18 (48 day period of time). The purpose of this analysis is to discriminate between "good" and "bad" users, based on behaviour.

1. User groups

Six user groups have been selected. The traffic breakdown after the sampling time period  (48 days worth of data) is as follows:
 
 
User Group Users in the sample Page views
One time 39% 0%
Very light 13% 2%
Light 15% 5%
Moderate 18% 17%
Heavy 12% 40%
Very Heavy 4% 36%

* One time user: didn't visit between 3/26 and 5/18.
* Very light users: visited only one day.
* Light users:  showed up either two or three days.
* Moderate users: showed up between 4 to 9 days.
* Heavy users: showed up between 10 to 27 days.
* Very heavy users: were on Snap at least on 28 different days.

If we look at the traffic on a single day during the sampling time period, we get:
 
 
User Group Users showing up Page views
One time 23% 16%
Very light 10% 10%
Light 14% 13%
Moderate 21% 20%
Heavy 20% 26%
Very Heavy 11% 16%

2. Results

Faithfulness is associated with:

  • Late hours for visiting Snap (10 pm to 4am, SF time).
  • Weekend users.
  • Fairly old users (not too old, not too young). By user age, I mean the time elapsed since the user has visited Snap for the first time.
  • Users from the West coast.
  • Users referred by email.com, speed.snap.com,pacbell.net.
  • Users visiting /main/finance, /mysnap, /main/local [very discriminating].
  • Users with Netscape browser (particularly version 3 Win95 U), not associated with the Netscape edition.
  • Users from the following editions: swbell, clubvaio, pacbell.
  • Users from the online survey are much more faithful than the average user,
    because of sampling bias  discussed earlier.

    Lack of commitment to Snap is associated with:



    Misc

    1. Survival Analysis

    2. Survey

    * Associations.
    * Correlation analysis.
    * Regression analysis.

    3. MemberId

    Analysing the user set consisting of the 1,200 sampled users (out of 15,000) who visited Snap on July 5 or July 6. The proportion of users with a memberID within each group is strongly connected to the user group label (the more heavy the user, the more likely he has a memberID).
     
     
    User Group users with memberId total users ratio
    One time 3 51 6%
    Very light 7 56 12%
    Light 6 93 6%
    Moderate 64 304 21%
    Heavy 120 423 28%
    Very Heavy 149 289 52%