Why we'll use Google Universal Analytics over Mixpanel and KISSMetrics

Agreed, these are edge cases. They did create a lot of questions for me though, and made the whole thing rather confusing as a user.

As to how much of an issue these edge cases represent, I find it hard to get a real sense of it. I guess it really depends on the situation, what you want to measure and the user experience you offer to your visitors.

As someone who has built several top 1000 trafficed websites over the past decade here is what the publishing industry definitely needs out of an analytics program.

1. Please give me a report that can prove that my user traffic is real.

2. Please give me a report that can prove that the traffic is healthy.

I know that I can get this from analytics now, but it needs to be the focus.

For a decade I've competed against content websites that for the most part game seo traffic, build click traps and generally pollute the Internet with secondary source content. I've always had fairly large audiences on my sites, with healthy 50% returning visitor rates. However, when it comes to getting ad dollars, I always lost to competitors who had much larger volume mostly because they were either buying meaningless inbound links or using some other scam like click trap "we recommend this hot girl talking about prostate cancer" photos to goose their numbers. Meanwhile we'd create quality content and my sites would have hundreds of comments, while theirs would have very little. It didn't matter that my audience was more engaged, advertisers bought volume.

I just need something that I can show to an advertiser (or even better, that they have access to and can compare) that says... hey, this website isn't a constructed fabrication made to fake volume and take your money you sucker. This is a real website.

A lot of the industry right now is based upon buying links from aging front door portals (Yahoo, MSN, AOL) which still do ungodly amounts of traffic with a mostly Internet illiterate audience. Sites buy these links, convert them into CPM click traps on their targeted magazine sites and sell their inventory to advertisers who don't know that the whole thing is shell game. They think they're buying ads on a hot new site with explosive growth.

Do you have more information on how I can see a live view of the practices you discussed in your last paragraph? I was under the impression that traffic from the top portals was costly and not exactly suitable as a component in an arbitrage play like you mentioned.

I started saving all my page views in a postgresql database. Schema is pretty simple.

I have the following tables:

    sessions
      session_id (uuid type)
      created_at
    
 
    page_views
      page_view_id
      session_id
      created_at
      site_id
      path
      query_string (hstore)
      user_agent
      referral_url
      ip_address
      user_id
      http_method (get, post, etc)
      details (hstore, used to tag page views/actions)
   

This allows me to simply query all my page views against data in my live database. I can see the path a user took to place an order. I can easily integrate a/b tests. If someone uses a coupon on the site and we want to see if they later came back and viewed/purchased more, we can easily write a sql query to figure that out. We can simply figure out lifetime customer value, even if not logged in. If we're getting a large amount of traffic from a certain affiliate, we can alert our staff.

It's really awesome to be able to have your data in the same place. Having analytics data spread out to GA made it difficult to match that data against ours. If we need to scale out to multi-terabytes, postgres_fdw will make querying against the analytical database simple.

Since we're also tracking affiliate purchases to pay out commissions, I also have another table that that stores additional information about a page view if they came from an affiliate site (click id, the affiliate network, etc).

Here's the plpgsql function I use for saving the sessions and page views: https://gist.github.com/joevandyk/f63523cdd1a3aa75d0ec

Yeah, we do that kind of stuff as well. At least you know what your data means. But when you start getting millions of hits a day, you won't necessarily want to spend some time scaling your system... In that case leaving it to the pros and focusing instead on your product may prove the most sensible move.