Flash Tracking: What Goes in Doesn't Always Come Out

Most of the web analytic tools out there claim that they can be tweaked to accommodate all-flash sites. They can. But with all they might hype about being able to track a Flash site just like a regular site – whether using a custom solution like Omniture’s Action Source or through embedded JS image-requests, I’ve found recently that there are certain key metrics that become untrackable in all-Flash sites. Rightly or wrongly, these metrics are also those which business managers often pay the most attention to.

Take the favorite metric “Time on Site”. Sometimes this is reported – as in WebTrends – as a key dashboard item (“average time on site”). Sometimes this is broken down into buckets, as in Omniture’s SiteCatalyst (“less than 1 minute”, “5-10 minutes”). These metrics will populate in a Flash site, but the numbers are often useless, and here’s why: Time on site is calculated by most web analytics solutions by subtracting the time-stamp of the first page-load from the time-stamp of the last page-load. Inherently, even on HTML sites, this means that single-access visits are not counted, and that the time spent on the last page viewed isn’t counted. On publishing sites this last fact can be very misleading – if I drive people to get quickly to some content page and they spend 20 minutes reading and digesting that content, but then save it to their favorites and leave, web analytic solutions think that’s a very short visit.

Now take an All-Flash site. If I’m using JS image-requests embedded in the Flash to record user activity (buttons clicked on, tabs, links), but there are no page-loads (and hence no re-loading of the universal tag), the web analytics solution thinks it’s a single access visit. Guess what – no time on site. Only if I get to a different HTML page which runs the universal tag will it register anything at all.

One major online retail site recently launched an All-Flash version and custom tagged it in Omniture to record a very complex (and laudable) variable structure. During the all-HTML phase of the site, they had stats on average time on site which were in the 2-3 minute range. The Flash site launched, and suddenly this became 4-5 minutes. Initially happy, they then realized that those 4-5 minutes comprised a tiny fraction of site visits, because it was computed only when someone clicked on one of the Footer links which were still in HTML.

Another metric which can be severely affected by all-Flash sites is referring domains. Here I admit that the only experience I’ve had with this issue is Omniture, but I hope someone reading this might know of examples in other web analytics solutions. The referring domain gets recorded on the universal JS tag, which continues to count “instances” of the referring domain until the tag is run again. Thus “instances” of a referring domain are vastly over-inflated on all-Flash sites. Luckily, the data can be de-duped in Data Warehouse or Discover, but is not ideal for quick reporting purposes, since “source of visit” reports are often one of the first things marketing managers look at.

Anyone know of other metrics affected by Flash tagging solutions?

Paul Legutko, vice president of Analytics for SEMphonic, oversees web analytics and SEM analytics projects. Legutko holds a bachelor of arts from Harvard and a Ph.D. and master's degrees from the University of Michigan.