Man, it seems like I just cannot get away from Gary Angel lately. He and I are engaged in some kind of crazy mutual-admiration society thing, which would normally worry me, but I know few people as into all this as Gary. He recently posted about “that darn engagement metric” where he expanded on some of my ideas and his response to my ideas and my response to his response to my ideas, etc. One thing Gary said really stuck out in my mind. Regarding my use of the visitor engagement metric to tell a story about the traffic that Marshall Sponder sends to my web site, Gary commented:

“This gradual building of a context around a measure is exactly how I think reporting actually works - and how analysis drives to actionable understanding.”

Exactly! The gradual building of context is what this is all about. The reason I’ve defined a visitor engagement metric is to provide another firm basis for the establishment of said context, another indicator on which we can draw to better understand a dimension or set of dimensions of data we collect.

Much has been written about the value of bloggers to business; it seems like you can’t open Business Week, Fortune, or even Newsweek anymore without having to read about the next big thing that bloggers and blogging are doing to change both business and society. But what can we know about the traffic bloggers send us? And what actions can we take based on that information? Let’s have a look …

Here I’ve rank-ordered the folks I list in my blogroll by percent of sessions they drive back to my web site. No great insights here based on my KPI “percent of sessions” but I suppose if I wanted to I could add Clint, Steve, Tim, and Eric Butler to my holiday card list.

Now I’ve added the session conversion KPI to my list of bloggers. Immediately I see two sets of actions I could possibly take: The first would be to send “much love” to Steve, Mike, Gary, Xavier, Manoj, Aurelie, and Tim for helping me pay for my children’s college education (all book proceeds go to my kids’s Fidelity 529 plans). The second would be to see what I could do to get the rest of the bloggers to say something like “Hey, go buy Peterson’s books, they’re great!”

Still, I should probably check to see first that these folks aren’t referring me traffic that later returns to the site and makes a purchase, right? I need to roll in a visitor-based conversion metric:

Ah ha, now I can see that I owe some serious thanks to Steven Jackson and the folks at the Blackbeak Blog! Better than one in ten people Steve has sent my way have made a book purchase, which is awesome. But it looks like I have a problem with bloggers like Anil Batra, Matt Jacobs, Robbin Steif and even Marshall Sponder. Zero percent visitor-based conversion to purchases on my site … must be some problem with how those folks are talking about me, right?

Oh, or maybe not, at least not in every case. I added a KPI for percent buy path sessions, basically the percentage of sessions in which a visitor at least starts down the book purchase path. Now I can see that Anil, Justin, and Robbin are all doing a pretty good job of getting people into the purchase consideration process, but for some reason those folks aren’t completing the purchase on my site. It’s not their fault, it’s my fault!

Damn.

But hey, maybe it’s still not my fault. Maybe even though the bloggers are sending me traffic that hits the buy path, maybe those folks aren’t really all that engaged with my site and content. Maybe the visitor’s they refer me are just looking at one page in the buy path and leaving, never to return.

Okay, or maybe not. Anil, Matt Jacobs, Marshall, and Clint are all sending me visitors that I consider to be “well engaged” with my site (my site-wide visitor engagement average is 30 percent.) Now I can see two clear action items:

1. I need to reach out to Anil and see if he and I can work out a deal to help further encourage his readers to completing the book purchase on my site. Anil, if you’re reading this, call me, we need to talk.
2. I need to reach out folks like Matt, Marshall, and Clint and see if there is some way I can get them to more passionately advocate for my books in their weblogs. Given that their visitors are more highly engaged than the “average visitor”, I have to believe their is an opportunity to sell more books.

But wait, I’m not done. In fact, I’ve only just begun to mine for the true opportunity here. But hopefully you can see, this gradual building of context is well-supported by each of the key performance indicators I keep in my arsenal, both simple metrics like “percent of sessions” and the more complex “visitor engagement”.

This post was a really long way of saying I agree with Gary about no one KPI driving a specific and easily understood action. All of our efforts are ultimately designed to help the online business better mine for opportunity and understand how that opportunity might potentially be leveraged. There are no easy answers, there are no silver bullets, there is no magic, nor mystery, nor puzzles …

There is only the gradual building of context …

Some interesting things I saw in the blogosphere the last few days:

• Robbin Steif interviewed me on behalf of the American Marketing Association for whom I’m doing a free webinar on web analytics in early March. Robbin always asks really good questions but sometimes my answers get lost in translation.
• Eric Enge has a really good interview with Jim Sterne. Jim is such a great leader of our industry and Eric Enge is really knocking ‘em down interviewing Brett Crosby, Dennis Mortenson (IndexTools), and Jim.
• Ian Thomas is busy telling us what Microsoft Gatineau will be without actually saying the words. Expect to see something outcome focused that is easy to use that provides people in a May/June timeline. Or, if you’re anxious to figure Gatineau out now, just grab the JavaScript code out of Ian’s blog and have a look-see.
• My very bright friend Ian Houston is starting a series breaking down the web analytics data model by focusing on the canonical data unit, the “event”. WARNING: Eat your Wheaties before you read Ian Houston. Ian makes perfectly logical statements like “refining the definition of Events to the concept of an Event Super Class where the sub-dimensions by type are not children of the Events dimensions but rather sub-classes of the Events class that inherit their properties and relations within the data model from the Super Class.”
• There are new jobs posted in my premium job board from PayPal, Stratigent, Cox Newspapers, Staples, and the World Wrestling Entertainment group.

Congrats to the Colts on yesterday’s Super Bowl win. I’m from outside of Chicago so my loyalties lie elsewhere but it’s nice to not see the freaking Patriots in the big game for a change.

While I have not had time to write Part V of my series on measuring visitor engagement, I wanted to take a few minutes to address some comments folks have made about the metric recently. It’s very encouraging to see folks like Gary Angel and Daniel Markus pushing the conversation about measuring engagement along as I can think of few more qualified to critique this work.

Gary Angel, who had very nice things to say about the metric, commented on how in some areas the metric is biased, specifically towards search engines and specific types of content. Gary is concerned that the Brand Index will unfairly bias towards search engines (given that one component is searches for brand-specific terms like “eric t. peterson” and “web analytics demystified”.) I examined this effect and it turns out that “branded searches” make up only a small part of the index for my site but Gary makes an excellent point, unnecessary bias should be removed from the index whenever possible. As such, in my current calculation I have removed this weighting from the Brand Index, redefining said index to only be direct sessions (non-search, non-referred.)

Score one for Gary.

Gary also commented that:

“… if I’m using my metric to measure the “engagement” produced by visitors who used a specific part of a site (like the blog or the press releases), it’s vitally important that my metric not include a strong built in bias toward one of the areas (like blogging). Some analysts might argue that this represents a flaw in the metric Eric proposes. I don’t think so. Every metric carries with it some biases – and no metric is appropriate to every situation.”

This is a good point, one that had been made by a handful of other folks who critiqued the metric early on. The problem I have with removing the Blog Index (ratio of blog reading sessions to all sessions) is the evidence that my weblog is a prime driver of engagement with my site and overall web analytics brand: Over the last 12 months, weblog subscribers are nearly 400 percent more likely to have returned to the site recently than non-readers; those visitors not subscribed to my blog (e.g., in Bloglines or Google Reader) but who are still reading blog content are 300 percent more likely to have returned recently.

Score one for Eric.

One thing worth noting, the way I am using Visual Site to measure weblog readership and subscription, this activity does not show up as traditional “page views” unless the reader A) reads the post on my web site or B) clicks through to the web site (at which time the post appears as a session “referrer”) — Visual Site is able to track external RSS and XML-based content using a non-page view event (something I call “reads”.) Not all web analytics systems afford their operators this flexibility so I thought it would be worth bringing up. This is part of the reason that the Blog Index needs to be a separate index, not part of the Click Depth Index as some have questioned.

But enough about Gary … Daniel Markus posted what I surmise to be a nice post about my visitor engagement metric at Marketing Facts late last week in which he called my calculation “the mother of all Web Analytics KPIs.” The post is entirely in Dutch and my Dutch is horrible so I wrote to Daniel and asked for a rough translation . While there were many good comments about the metric, they raised two concerns:

1. The calculation is complicated and difficult to understand.
2. There was some question of the utility of this metric, essentially calling into question the overall “actionability” (not a word) of visitor engagement.

Regarding the complexity of the calculation, as Gary has so eloquently stated any number of times, no indicator or metric is any use without understanding its components, its definition, and its inherent biases. Clearly the onus is on the web analyst to explain the metric and it’s definition to any audience they present engagement data, especially given the complete lack of formality around measuring “engagement” (at least until you started reading my posts on the subject.)

Given the complexity of the calculation, the latter concern is valid but one that misses the point of the metric. There are any number of loose definitions of “engagement” floating around in our community — duration, page views, average page views per session, sessions per visitor, etc. But none of these more easily understood (note: not easily interpreted) metrics, in my mind, captures the essence of an engaged visitor.

Visitor engagement has to be examined over diverse criteria, simple assessments simply do not work. To wit:

• To say that session duration is a good measure of engagement is fine, unless the visitor never returns to the site.
• To say that a high number of page views is a good measure of engagement is fine, unless the visitor runs up those page views in a very short period of time and was unlikely able to actually read content.
• To say that recency of visit is a good measure of engagement is fine, unless the visitor has only looked at your home page and left.
• To say that direct visits are a good measure of engagement is fine, unless those direct visits lead to short sessions of few pages viewed and the visitors never return.

I believe that the complexity of the calculation is where visitor engagement derives its value. For practitioners who are lucky enough to have access to a platform that can actually make this calculation and who are willing to take the time to explain to their audience what the metric measures and what its limitations and biases are, the metric can yield insights that would be unlikely to fall out of “traditional” web analytics.

I will leave you with an example of how I am deriving small insights from my measurement of visitor engagement.

Marshall Sponder is the WebMetricsGuru blogger and all-in-all a pretty nice guy. He and I had a little tiff awhile back over Avinash’s web analytics blogger index (something Avinash has stopped doing for some reason …) when I was less than complimentary about the volume of web analytics posts that he produced relative to his blogging in general. Examining traffic metrics from Marshall’s blog I would interpret the value of having a good relationship with him based on a set of commonly understood data:

Almost no volume and no books sold. Come on Marshall, let’s see a nice recommendation for Web Analytics Demystified already! ;-)

But wait, what if I have a closer look at the measured engagement of the visitors he’s been sending to my site:

While my “average” visitor to the site is only 24.2 percent engaged, visitors from Marshall’s posts are nearly 40 percent engaged with my site and, more importantly, of these visitors almost 10 percent are “highly engaged” (50 percent engagement or better.)

Marshall may not be selling books yet, but I have the nagging feeling if he tried even just a little, he could probably drive pretty good numbers given the engagement of the audience he referrers.

Now just imagine that you were running a million or billion dollar business, looking for new opportunities on the Internet. You have hundreds-if-not-thousands of sites sending you visitor traffic all day, every day. Maybe some of these people make purchases, but maybe you have nothing for them to purchase … how do you decide who to spend more time with and who to ignore?

Me, I’m going to write nice things about Marshall Sponder and if the folks from e-consultancy call me and want to do another interview, I’m taking that call right away! How’s that for a KPI defining an action?

For those of you keeping track at home, this is the fourth in what will likely be a five-part series on calculating an “engagement metric”. The first three posts are here:

• Part I
• Part II
• Part III

Originally I had postulated that an engaged visitor, at least on my web site, can be characterized as follows:

1. The visitor views “critical” content on the web site
2. The visitor has returned to the web site recently
3. The visitor returns directly to the web site some of the time
4. Some high percentage of the visitor’s sessions are “long” sessions
5. If available, the visitor is subscribed to at least one available site feed

Basically, the final calculation, one revised thanks to the valuable feedback of dozens of folks, is essentially the same with a few slight modifications. The final goals for my site, goals easily tweaked for any site, are as follows.

Well-engaged visitors will:

1. View a relatively large number of page views in a given session
2. Have visited the site in the last four weeks
3. Have relatively long sessions
4. Come directly to my site or come from a “Eric Peterson” branded search
5. Be reading my weblog in addition to non-blog content
6. Buy one or more of my books through my web site

As you can hopefully see, the first item in my original list (view “critical” content) has been softened somewhat. While the act of purchasing is necessitated by viewing critical content (my “thank you” page) ultimately I agreed with several reader comments that the a priori definition of visitor goals would skew the metric and reduce the metric’s ability to tell me about all of the content on my web site. Thanks to Victor and others for hammering this home.

Given all this, the visitor engagement metric is composed of six sub-metrics, each of which can be examined individually to provide context to the larger calculation. The six sub-metrics are:

1. Click-Depth Index: Percent of visitor sessions of “n” or more pages
2. Recency Index: Percent of visitor sessions occurring in the last “small n” weeks
3. Duration Index: Percent of visitor sessions of “n” or more minutes
4. Brand Index: Percent of visitor sessions originating directly or originating from search engine searches for terms like “eric t. peterson” and “web analytics demystified”, etc.
5. Blog Index: Ratio of blog reading sessions to all sessions
6. Conversion Index: In this case, session- or order-based conversion

Keep in mind, engagement is a visitor-based calculation, one designed to look at the lifetime of visitor sessions to the web site. So that the engagement of any visitor is a function of their lifetime of visits. Yeah, this assumes some stability in cookies so always use first-party cookies.

The final calculation is simply a summation of the component indices divided by the total number of components which yields a simple percentage:

If you’re looking across multiple visitors, you would read this as “the average visitor is just under 27 percent engaged, as defined by X, Y, and Z.” If you’re looking at a single visitor you can break engagement down on a session-by-session basis, watching for increases and decreases in the visitor’s engagement over time. In aggregate, visitor engagement becomes a very powerful but elegant key performance indicator that tells you a great deal about the make-up of your audience.

Once you decide that you need more information about the basis for an increase or decrease in visitor engagement, and assuming you have the right technology powering your analysis, you would simply visualize each of the core components over time:

As Clint commented in my last post, there is a surprising stability in each of the components, which is in my mind what you’re looking for. I want to see the variation show up when I examine engagement against my business-critical dimensions (referrer, campaign, page, search term, etc.)

When you analyze the visitor engagement calculation against all of your site visitors, you’re looking for a more-or-less normal distribution. This distribution is spiky because of the calculation, but if you’re able to drill-down, you should see something like this:

(The bars that exceed the visualization’s scale represent peaks that occur as visitors achieve 100% of sessions for 1, 2, 3, 4, and 5 of the engagement calculation’s core components. If you want to see this image at 100% scale let me know …)

Another way I can think about this is to use a scatter-plot, basically showing the same thing but easier to visualize differences as you drill-down into specific dimensions:

All of these calculations actually become relevant when you actually apply them to a dimension of data. Here, for example, is visitor engagement mapped to blog posts from my and Avinash Kaushik’s weblog:

Pretty cool, huh? I mean, it’s no great surprise that Avinash’s 2007 Web Analytics Predictions post has the highest visitor engagement score in this image when you think about all of the follow-up predictions his original post spawned. But boy-howdy, isn’t it nice to see that in a metric that you can understand and actually use?!

Here is the visitor engagement metric applied to some of the referrers to my web site:

No great surprise again that Feedburner, Technorati, and Wordpress are driving visitor engagement given that they are likely to be driving visitors maxing out their blog index score. But what about the folks at ROI Revolution, sending me visitors who are on average over 30% engaged, or Blackbeak and the folks at Conversion Chronicles, sending me visitors who are as engaged as my 27 percent site-wide average?

Arrrrrrr, indeed!

Finally, and I know that I showed this already but I just think it’s damn sexy, I can map visitor engagement against any geographic dimension in my system (continent, country, city, state, zip code, DMA, etc.) to see where I might want to focus my local marketing efforts in the future:

You two people in Midland, Michigan, get ready for an onslaught of Web Analytics Demystified promotions!

Oh, some random notes:

• You can add non-page view events (RIAs, AJAX, Flash, etc.) into the calculation easily. I don’t have much of that on my site but I have an “Event Index” calculation that can be added for sites heavily leveraging these types of applications.
• You can add the “social media index” that I discussed in my last post just as easily as you can add content-based indices for retail, customer support, business-to-business, or content.
• You can take or leave my idea of scoring the brand index against specific search terms. Visual Site gives me a really easy way to do that and I believe that engagement is very much a function of brand awareness, something notoriously difficult to measure in any practical way.
• Visual Sciences customers interested in deploying the visitor engagement metric should contact me directly via normal company channels. I have pretty much everything you need to get up and running with this in a ZIP file and I’d be happy to talk you through the process.

As always, I welcome your comments and feedback on the engagement calculation and anything else that comes to mind. In the next (and perhaps final) installment I will cover Clint’s inevitable complaint of “What the heck happened to all the great ’social media’ stuff?!?” as well as talk about some specific applications of the engagement metric.

UPDATED: I am a bad blogger. I referenced my friend Dylan Lewis in my original post and didn’t link to his blog. Dylan can be read at http://www.passionateanalyst.com/ and I encourage you to ask about the Platypus thing. My sincere apologies, Mr. Lewis!

UPDATED: I totally forgot about all the 2007 web analytics predictions! You can see a nice summary list posted to the Yahoo! group by Lars but this list does not include one of my favorite set of predictions from Craig Danuloff. Craig I think makes the boldest predictions of anyone playing the game.

It occurred to me last night that I’ve been so engaged with measuring engagement that some interesting stuff has passed me by. Let’s catch up over coffee, shall we?

• After being “tagged” by Gary Angel I tagged a few folks. So far Clint Ivy and Eric Butler have responded to my tagging, but perhaps the most interesting tag response comes from Dylan Lewis who postulates that the game of tag is either a thin disguise to increase our page ranks for searches for things like “web analytics and the grateful dead” or some intense navel gazing.
• The “death of the page view” conversation, while interesting, is starting to go too far when otherwise smart people begin to predict things like “we’ll no longer bother to collect pageviews by end of 2007.” While you can make the case for using unique visitors in comparative situations, I sincerely question statements like “[the] page is no longer considered a metric worth looking at.” Is it me?
• Justin Cutroni had a really good post a few weeks back titled “Web Analytics: It’s About Process” that I loved. I’m still well-engaged thinking about the processes behind the successful “doing” of web analytics at the Enterprise-level and have had several enlightening conversations lately. One of my favorite comments was “people often consider “process” to be a dirty word …” Ouch!
• There is a new job posted on my job board, which is slow to take off likely because I have been too busy to bug all the recruiters posting to the Yahoo! group to give it a try! If you’re in Utah and have experience with the local analytics technology, have a look at this posting.
• Finally, and those of us thinking about how “Web 2.0″ is going to be measured knew this was bound to happen, measurement tools are coming to Second Life. While I’m not a Second-Lifer (I barely get everything done in my first life) I am dying to see what kinds of metrics Electric Sheep are able to come up with. Can you imagine the KPIs? “Percent Avatars propositioning sex” and “Percent Avatars pretending to be adults who were probably eleven-year-olds” and the such. Seriously though, if you have access to these reports, I would LOVE to see them.

As usual, I welcome your comments.