Posted by: Pete Horadan

So I’ve laid the foundation for the idea of unique signatures, now let’s look at them in more detail.

On the access side, we start by creating a Device Signature for each machine that accesses an application. We track over 80 unique attributes of the machine, and utilize a proprietary statistical method for understanding the patterns and the evolution of change to the patter for a given machine. Using this approach, we can tell with great accuracy, which machines accessing a system are unique and create associations between devices and visitors.

Another of our algorithms—keystroke dynamics—can actually get to the individual level if required.  Keystroke dynamics works in conjunction with our Device Signature, analyzing visitor-specific typing rhythms. While typing a string of characters, every person exhibits a series of keystroke timing events, or a “biometric signature.” The measurement and comparison of visitor-specific typing rhythms is called keystroke dynamics. Keystroke dynamics derives a unique Biometric Signature of the visitor—essentially enabling systems to infer individual people using the system, independent of location or system. 

Accurate contextual information has been the focus of our latest technology break-throughs here at Scout Analytics.  Using our experience in creating complex algorithms that produce a highly accurate device and biometric tracking, we’ve developed new algorithms that take the incoming IP address and translate that information into insights about the business or legal entity associated with the user.  By linking IP address information with firmographic and geolocation data from online databases, a Network Signature is derived that can be used for segmentation of behavioral profiles.

The predictions produced by behavioral analytics are only as good as the quality of the data behind it.  We’re focused on doing it right.

Behavioral Analytics Accuracy, Metrics, Tracking

Posted by: Pete Horadan

I talked earlier about the insufficiencies of current tracking approaches and their impact on behavioral analytics.  Let’s now talk about how it should be done—how we do it here at Scout Analytics.  Basically, our technology is centered around a unique signature concept.  We analyze dozens of unique attributes and utilize patent-pending algorithms to derive a unique signature for the device, network, even the individual user (biometric). 

You can think of signatures sort of like a genome—while there are overlapping characteristics—each carries a unique and identifiable pattern and sequence.

Of course, this isn’t easy or everyone would be doing it.  As you can imagine, machines, networks and even people change in different environments and over time—you install a new version of Flash, or change to Silverlight.  The cast on your broken right hand changes how you use the computer.  You change network carriers.  The trick is to have algorithms that take into account enough factors and have enough intelligence to sense the change as a progression of a previous signature, rather than a different signature.

We have algorithms for recording and tracking a Devices Signature, a Network Signature and a Biometric Signature.  Each of these unique signatures works to improve the accuracy of tracking and context during analysis.  So next time I’ll give a little more insight into each.  The development of these techniques is something that truly sets us apart, and it’s what makes Demand Rating™ work.

Behavioral Analytics, Metrics Accuracy, Behavioral Analytics, Metrics

Posted by: Pete Horadan

Last week, we looked at the serious acurracy issues associated with using cookies to track visitor behavior and engagement.  This week we’ll explore the accuracy of another often-used factor, IP address. In short, IP address provides even less information about users than a cookie. 

The accuracy issue starts with the fact that one user may have many IP addresses.   For an example, let’s look at me. I access different sites (e.g., TechCrunch) from home through my personal broadband, while at my coffeehouse through their broadband, or through the network at my corporate office. When I’m on the road, I broaden my access points to include hotels and airports. When using IP address to count visitors, the information collected for me might look something like this:

IP Address                    Domain                         Location
24.16.13.198             comcast.net                Woodinville, WA  
208.54.4.23               tmobile.net                   Issaquah, WA
70.103.221.189    scoutanalytics.com              Issaquah, WA
12.204.178.67          marriott.com                  New York, NY

Of course this is incorrect. What appears to be four unique visitors, is really only me. 

The flip side is true as well, namely one IP address may have many visitors.  The common example is found when visitors behind an office firewall may all appear to come from the same address. In this case, paid-content providers have no visibility into individual visitors or their different behaviors. So, when trying to analyze unique visitor counts from IP addresses, the problem is almost intractable.  Does a given IP address represent one visitor, many visitors, or just a fraction of a visitor?

On their own, IP addresses have serious limitations in terms of establishing unique visitors or tracking behavior, seriously over- or under-estimating the true picture. Improperly used, they can lead to faulty analysis.  IP addresses are best used to provide hints which I’ll discuss more in future posts.

Behavioral Analytics, Metrics Accuracy, Behavioral Analytics, Metrics

Posted by: Pete Horadan

So we’ve been making the point all along that Demand Rating™ provides a new lens on subscriber loyalty. Realistically though, the usefulness of behavioral analytics or web analytics are gated by accuracy.  So how do you get accurate online metrics?  What’s really knowable?

Let’s start with usage tracking.  The traditional approach is through the use of tracking cookies.  A cookie is a small piece of data that is stored on a device by the application. Each new browser that accesses an application is given a cookie, and the application relies on that cookie to track user behavior. The issue, of course, is that the cookie method of tracking is wildly inaccurate, and it’s simple to see why.

At the highest level, tracking based on cookies don’t track individuals at all—they track cookies.   Users access applications through a variety of browsers and machines—from the office, at home, from their mobile devices.  Cookies can and are often reset by the user or even automatically by the browser whenever it shuts down.  It is practically never true that an individual visitor will utilize a single cookied browser for access.

In fact a recent study at Penn State shows that “cookies are about just as inaccurate in estimating unique visitors as unique network addresses,” overestimating audience size by 7 -30 times! So, do you really want to know that 90 cookies accessed your application?  Or that three people did?  The whole flawed cookie process results in a huge overstatement of usage and a total inability to track actual people. 

In a future article, I’ll talk about a better approach that avoids these problems.

Behavioral Analytics, Metrics Accuracy, Cookies, Metrics