Count Your Traffic

We've spent a lot of time discussing how to measure your Web site's effectiveness at achieving conversions, and each time we do, we talk about visitors to the site. As you have seen, you cannot determine your conversion rate unless you know how many visitors are coming. But how do you count visitors? That is what this section explores.

Web metrics experts refer to visitors coming to your Web sites as traffic, and the way Web traffic is measured has changed a lot over the years. You need to understand the fundamentals of traffic measurement because it is a critical part of tracking your success.

Page Views

The most fundamental metric in Web measurements is the page view, which is just what it sounds likea count of how many Web pages have been shown to your site's visitors. Web metrics reports always summarize page views for a particular time period, so you can see how many page views occurred in a day, a week, or a month.

By analyzing the trends of page views, you can see whether your views have gone up this month from last month, or from the same month last year. (For a toy retailer, comparing January's page views to those of last January is usually more instructive than comparing to the Christmas-inflated December numbers.)

More importantly, the page views of individual pages can be analyzed, so that you can see which particular pages on your site have been viewed the most, and you can analyze the trends for each page over time.

HOW PAGE VIEWS ARE MEASURED

No search marketing campaign can be measured without counting page views. But despite how simple it is for you to count the page views by your Web browser, it has taken years of evolution for Web metrics systems to accurately count them.

In the early days of the Web, most Web sites were measuring hits, the number of computer files that your Web server displayed to your visitors. At the time, this made some sense because most Web pages consisted of text from a single file. If every page consisted of one file, the number of hits was the same as the number of page views. It was easy to count hits, because your Web server writes a record into a log file every time it serves a file. Web measurement programs, such as WebTrends, emerged to analyze log files and produce reports summarizing the number of page views. So far, so good.

Unfortunately, by the end of the 1990s, hits had no meaning at all: Almost all pages were complex combinations of multiple text and numerous image files. The hits metric could tell you how many files were shown to visitors, but could not reveal how many actual pages were viewed. Existing Web metrics facilities tried to compensate for this situation with tortured algorithms that estimated "true" page views, sometimes by examining how close together in time the hits were retrieved and which user they were served to. So, if the Web log file showed that eight files (hits) were shown to the same network address in a split second, the metrics facility assumed this was one page view.

But lots of situations caused these algorithms to fail. Large companies with complex server environments served single page views from multiple servers: The company logo image file came from one server and the text from another. Now the metrics programs had to gather up all the log files from every company server to see that there were many hits to the same network address at the same time. And even then, the algorithms ran into problems because the servers had their internal clocks set slightly differently, so you could not tell how close together the hits actually were!

In addition, all America Online customers (more than 25 percent of many Web sites' visitors) appeared to be coming from a small range of network addresses from somewhere in Virginiathe addresses of AOL's servers. So, the metrics programs could not tell which hits were for one visitor versus anotherseveral at a time shared the same network address because they were routed to your site from the AOL server's network address.

So, the algorithms were failing both because they could not gather together all the hits from multiple Web servers with valid times, and because they could not tell which hits went to which visitors. It was clear that a new approach was needed. Variously known as "single-pixel tracking" or "Web beacons," this technique depends on you adding a one-pixel image file to each Web page on your site. When each page is served to a visitor, the hit for this unnoticed image is stored in your Web log file and can be counted as a surrogate for a page view.

Most Web metrics programs can use this technique today, and that ensures the accuracy of your page view metrics. If your metrics package still does page view estimates based on hit algorithms, you might want to invest in a better package.

Visits and Visitors

More interesting than analyzing page views is examining just who is viewing those pages. Most Web metrics programs can identify a series of pages that have been viewed by the same visitor, by using cookies. The Web metrics program "drops" a cookie file on each visitor's computer when she enters the site. This file contains a unique identifier not shared by any other computer's cookie. The Web metrics program reads the identifier from the cookie file each time a page is viewed and remembers which computer viewed that page.

Identifying visitors allows metrics programs to provide you with better information. Now, instead of merely counting page views, you can count the number of visits to your site. Visits, called sessions by some metrics facilities, let you see how many people are coming, not just how many pages were viewed by all people. For example, 2 Web sites might each show 100 page views, but 1 site has 4 visits with 25 page views each, whereas another might have 10 visits that average 10 views each. If you count just page views, all you know is that 100 pages were viewed. If you count visits, you know that 1 site got 4 visits, but they stayed and looked around for a while, whereas the other site got 10 visits from people who looked at fewer pages in each visit. This information is critical because you cannot get more conversions from more page viewsyou get more conversions from more visits.

In addition to identifying visits, you can also track the number of visitors (sometimes called unique visitors), because you can track whether the same visitor has returned to your site multiple times. So, the metrics report could show 100 page views in 10 visits by 8 different visitors, because a couple of those 8 visitors had more than 1 visit. This is important, too, because for many Web sites, people tend to visit multiple times before they convert; therefore, knowing the number of new versus returning visitors can tell you even more than tracking visits alone.

It is true that Web metrics programs accurately count all of these measurements, but it is critical that you use a single metrics facility across your entire site. Although small sites would undoubtedly do so, large Web sites with hundreds of thousands of pages might implement different metrics facilities for different parts of the site. If you are in this situation, collecting accurate measurements across your entire site is impossible.

To see why it is impossible, suppose that you have two metrics systems, one used in the sales part of your site and the other in the support area. In this situation, you cannot just add up the number of visitors in each system to get your total, because each system drops a visitor cookie when it sees a new visitor. Any visitor whose visit crosses both sales and support pages will be double counted. It will seem like two separate visits when it was, in fact, one. If you have even more than two systems, the problem is even worse. Double-counting your visits lowers your reported conversion rate and makes your efforts appear less effective than they are, because sales are reported accurately but the number of visitors that created the sales is artificially high.

If you must live with a situation where multiple metrics facilities are used for the same Web site, just keep in mind that the visit and visitor numbers will not be accurate when added up for the total site. It is still better to use those inaccurate numbers than to ignore them, because you can perform trend analysis on even inaccurate numbers because the inaccuracies are relatively constant. For example, if you add up visitor totals across three separate metrics systems, they will not be right, because some of those "visitors" are actually the same folks who headed into different parts of your site. Regardless, if you see that you added up 30 percent more visitors this month than you added up last month, you can probably conclude that something good is happening, even though you really cannot be sure that the improvement is actually 30 percent as opposed to 22 percent or 45 percent. It is definitely worth the effort to get to a single metrics system so that your numbers are accurate, but do use whatever data you have, no matter how imperfect. It's better than nothing.