tirsdag 16. februar 2010

Biometrics firm confirms: User counts for websites are 2-4 times too high

For years, comScore Web analytics experts have said that cookie-based user counts are wrong, often by a factor of two or more because most Americans use more than one computer to access the Internet.
Now, a tool meant to better identify, track and measure Web users through clever biometrics has backed up that claim.
Scout Analytics, a startup based in Issaquah, Washington, just east of Microsoft’s Bellevue campus, will issue a report Wednesday that says the cookie-based user per month metrics used by many small online publishers — for example, the Google Analytics tools we use at VentureBeat — overestimate the number of unique visitors to sites by a factor of two to four. That means many sites’ actual readerships are only a quarter of what they seem to be.
Scout has a fascinating product: They insert a Javascript program into a Web page. While the page is in front of you in your browser window, the program can measure your keyboard typing patterns: How long you hold each key, and how long between individual keystrokes. To be clear, Scout doesn’t record what you type. It quantifies the rhythms of your typing.
Scout claims that everyone has personal typing rhythms that can be differentiated in as few as 12 keystrokes. A typical site login is sufficient to identify if there are more than one people sharing a username and password.
More important, Scout’s software can also tell that the same person hit a website from five different computers over the course of a month, or that three people are sharing a single login and password.
Scout VP of strategy told me in a phone call on Tuesday that the average number of computers used by most Americans during a 30-day period is 3.5. The count increasingly includes a smartphone with a browser. As I wrote for Slate in 2006, we couldn’t tell if Slate had 4 million readers, 8 million readers, or some entirely other number. Different methods and different companies gave us wildly different counts.
Analytics consultant John Lovett of Web Analytics Demystified said in a prepared statement for Scout, “Virtually all measurement techniques have some rate of error, but online marketers who have a heavy reliance on cookies need to know this method has astonishingly low accuracy.”
Lovett means that a cookie can’t tell if different people are using the same computer, or if multiple people are sharing an account. This is often the case: a laptop, a desktop and an iPhone all being used by the same person.
Scout summarized their findings in the pending report:
From the device signatures of each visit, Scout Analytics was able to identify nearly 600,000 unique devices. Further correlating the biometric signatures from the visits, Scout Analytics identified more than 175,000 unique individuals approximately 45,000 of whom were unlicensed. When each distinct paid content product was analyzed, Scout Analytics found the number of users to be two to four times overstated by a theoretical 100% reliable cookie (i.e., even if cookies were never deleted).
Scout, founded in late 2008, has about 25 employees and is privately funded.
[Chart illustration by Paul Boutin. It's not VentureBeat traffic. I made it up.]

Ingen kommentarer:

Legg inn en kommentar