On The Trail

Some of the things keeping me busy lately are a bunch of site re-designs. From a web analytics perspective this means new strategies, choosing appropriate KPIs, creating new reports and, of course, tagging the new sites appropriately and making sure the tags work as intended. I use a few tools to help me with the tagging tasks, here they are in no particular order:

• Live HTTP headers – This is a Firefox plug in that shows all the HTTP request and response headers that are sent when a page loads. It has a nice filter feature that allows you to just show the headers you want to see (some pages might request 50 images/stylesheets/script files so it can get kind of messy without a filter). This is VERY useful for debugging AJAX and Flash tagging. One drawback is that it doesn’t parse the querystring, you just see the whole shebang on one line. That said, it is usually not too tough to confirm that the key info is being sent.
• Fiddler – The same deal except for Internet Explorer. This one doesn’t have the nice filter feature, although you can write scripts to extend it’s functionality. That’s on my to-do list, but for now I just search through the list. Also, to see HTTPS requests you need to install something else, it is explained on the Fiddler site.
• WT Debugger – The WebTrends debugging script. Drag this link on to the “Links” area of your browser. Press the button when you are on a page with a WebTrends tag and you’ll see everything that has been sent to the SDC server, nicely formatted.
• Our homemade page checker spider. We wrote a little spider that will crawl a site and look for specific text in the source of each page. We use this to look for missing tags. It’s still pretty shaky. I’d love to hear of any similar (free/cheap) programs out there that do this.

That would be the smell of a stale blog. Two things have kept me from writing: summer and being insanely busy. Summer in Minnesota is not to be wasted behind a keyboard! I’m still insanely busy, but summer is on the wane. So, it’s time to get back at it!

TrackMeNot came to my attention the other day. It is a plugin for Firefox that sends a random search query to the major search engines every few seconds (default is every 12 seconds). It was inspired by the release of the AOL search data. If you haven’t heard about that you can search the data here. Pretty cool, but scary eh? I wouldn’t want my search records collected like that, even if my name wasn’t directly attached. TrackMeNot is meant to game this system by producing so much noise that the signal is lost. The tool is pretty simple, it looks like it took about 10 minutes to program.

TrackMeNot has no direct impact on Web Analytics (unless you work for a big search engine), but it should serve as a warning: Be careful with your data! Think about how it would look in public and be very careful about collecting and processing anything that would compromise your visitor’s privacy. The scary thing about the AOL data is that many people were easily identifiable even without their names attached. It would take another 5 minutes of programming for the creators of this program to do the same thing to most web analytics data. I don’t think there is a big demand for something like this right now, but If there are a few high profile unintentional data releases that would change pretty quickly. And that would make all of this much more difficult.

The first Minneapolis version of Web Analytics Wednesday is being held on June 14th — only a week away! We’ll be at the View at the Calhoun Beach Club. Not coincidentally, the WAW will start after the MIMA event upstairs in the Calhoun Beach Club. We’ll be starting at 7:30, so be there or be square. Or, more likely, be there AND be square.

If you want to go please RSVP here: Web Analytics Wednesday. (Registration is free!)

UPDATE: I wrote this a week ago and thought I pressed “publish”, but I guess not. It didn’t show up on my site and I only just discovered it… Anyway, better late than never!

I just got back from Eric Peterson’s talk at tonight’s MIMA meeting. I loved it — because his main thesis is that companies don’t spend enough on Web Analytics in general, and web analysts in particular. Can you say “job security”? Keep preaching the Word, Eric! Much of the data to back up this thesis was from his Jupiter days. It can be seen in a few archived webinars, I think this one has most of it.

He did have a few points that I hadn’t heard before. One was the concept of “The New Usability Framework”. Eric defined this as a marriage of automated a/b style testing, website clickstream analysis and survey/customer feedback with traditional usability best practices. This is one of the things we aspire to do at Ciceron (although we usually don’t have the budget for automated testing). We certainly didn’t have a snappy name for it like “The New Usability Framework” — get it in our marketing copy, STAT!

The other thing that stuck out for me was Eric’s answer to a question about how to most effectively present data. (I happened to have spent a good chunk of the day thinking about this so my ears perked up). He related a story about an analysis team that presented ideas in the form of a news story or a press release. Headline, copy, supporting graphics, all printed out on paper. That is a really good idea and I can think of quite a few instances where this concept will make our reporting more effective.

It was also an excellent networking event. I finally met my local WebTrends rep face to face and a bunch of others in the Minnesota Interactive Marketing scene (imagine that at a MIMA event). I’ll have to start going to these things more often!

Today’s ClickZ article by Shane Atchinson (Branding Versus Direct Response: Sound Familiar?) was thought provoking. He asks why web analysts currently focus so much on the direct response metrics instead of focusing on long term customer relationships and branding. The answer (as I’m sure he knows) is simple: it is easy to measure direct response style metrics. Things get fuzzy pretty quickly when you venture into branding.

That said, I agree with his point completely. We should focus more on the effects of branding. Such a large percentage of overall ad spending goes there, who wouldn’t be interested! The problem is defining success in a way that can actually be measured. It’s easy with direct response: $X in sales or XX leads. But what is a credible measure of brand loyalty? And at the end of the day how can brand loyalty be tied back to actual sales numbers? Shane didn’t address this question, but it is the key to his proposal.

As you may have noticed, I didn’t get around to posting anything about days 2 or 3 of eMetrics. I figured it was more important to spend the time participating in the conference than it was to write about it. That said, the conference and the thoughts it stirred up will be providing me with quite a bit of content here for a while. If you’d like good reporting and commentary of this year’s eMetrics check out these thoughtful posts by Eric Peterson and Robbin Steif (in both cases these are the last of several posts, so keep clicking).
One thing that has been rolling around in my head for the last week was a discussion I had about branding KPIs during the “Critical Few KPI Discussion”. I mentioned to the group that we always report the proportion of branded vs. non-branded searches even though there isn’t really anything we can do based on that info. But, as soon as I said it, it occurred to me that we could tease out an actual KPI from the search data. The trick is that some people come to the site from a search engine twice. If both of these searches are available then you have a measure of the visitor’s brand awareness before they visited the site, and after they have visited the site. I can think of a bunch of different ways to calculate it, but I think this is the best way:

Site Branding Impact = (% Branded second searches)/
(% Branded first searches for the segment that searched more than once)

This meets Eric Peterson’s requirements for a KPI, namely that it be a ratio of two numbers and that it measures something that you can actually change. It also turns out this number is pretty easy to get from WebTrend’s visitor history file, I’m sure other tools can do it as well.
Now, I realize that this is an imperfect metric. People who really remember the brand may remember the URL instead of doing a second search, or they may bookmark the page, or delete their cookies or whatever. The search engine rankings for branded and non-branded terms will change over time and skew the results. The big “but” here is that there are precious few other branding metrics that qualify for KPI status. Jason Burby wrote a ClickZ article about it today and I would argue that none of the measures he listed could really be boiled down into a single number. Given that there are few alternatives, I’ll take imperfect.
How can you impact this number? Well, two of our clients will be releasing new site designs within the next month. This seems like an ideal time to compare the site branding impact of the new sites with the old sites. I’ll let you know how it works!

Xavier Casanova, founder of Fireclick (a web analytics tool) has moved on to a startup called Perenety that is developing a application that makes it easy to share big files (like videos, images, work files or whatever) with friends or whoever. He also publishes an excellent blog that he doesn’t update nearly enough anymore. He talked about a ton of great stuff, but I’ll boil it down: point one was viral marketing. He showed the hockey stick growth curves of several blogs (including his own). His point was that you have to “seed” your traffic for a while until you get enough “sneezers” to start the viral growth. (I love the term “sneezers”, and I’m going to steal it and use it shamelessly). The sneezers won’t sneeze unless there is something cool or remarkable to talk about. So far so good with blog popularity, but what does this have to do with his startup? Well, for a startup like his to get funding these days it needs to grow virally and quickly. They ran a quick survey to see what people thought was cooler: A program that allowed you to access your files anywhere or a program that allowed you to quickly send big files to your friends and family. They were describing the same product, just with two different messages. Both were received well, but “survey says” #2 is much cooler — so it was more likely to have a successful marketing campaign. This seems like it may not be a big deal, but it changes the focus of the company from business users to the mySpace crowd.
So here is another case of analytics having a huge impact on a company.

More tomorrow!

Highlight #1 of 2 for the first day of eMetrics:

Stacey Coopes from FordDirect.com is doing amazing things with her web analytics data. I wish she had had 3 hours to talk because each slide of her presentation could have been expanded into a full length talk. Here are two bits that really stood out: Simple, quick tests are more likely to pay off than major redesigns. Example: They did a major redesign of a part of their site and it fell flat, but changing the wording on a button increased conversion by up to 20%. We’ve done exactly the same things on a few sites with similar results. The take away is that simple is good and low risk, complicated is hard and high risk.

So that was neat, but the coolest bit of her talk was a very simple graph. FordDirect.com allows you to “build your car” or otherwise indicate what features you want. The graph had the percentage of website visitors who indicated an interest in a particular option (eg. V6 engines) vs. the actual sales of that particular option. It was an almost perfectly straight line. In other words the website clickpath is an almost perfect predictor of how well Ford’s various models and options will sell. If that wasn’t cool enough, Ford is actually using this to plan their production. Web analytics matters at Ford — to the tune of several billion dollars.

I’m currently in sunny Santa Barbara at the eMetrics Summit. Day one has just about wrapped up, all that’s left is a little eating and drinking on the lawn. I came to the same conference last year, and before the venue, speakers and food (especially the food!) are excellent. There were many excellent talks today, but I think the highlight is putting faces to the names of people I see writing in various venues throughout the year. The networking here is excellent, and I’m a guy who would rather do pretty much anything else. The open, collegial atmosphere is the best part of the conference and it’s really why I came back.

I could write many pages about the talks, but I’ll limit it to two highlights.

The highlights turned out to be rather long, so I’m going to put them in separate posts, so you’ve probably already read them since they’ll show above this one. Oh well!

A recent ClickZ article by Shane Atchinson has caused a ton of good discussion on the Web Analytics news group In a nutshell, Shane Argued that traditional linear conversion funnels are fairly useless because few people browse a site using a common conversion path. He argues that a hub and spoke model is much more accurate, wherein the hub is a particular page/section of interest and the spokes are the paths away from it. Some spokes move towards a conversion (this is good), while some move away or are dead ends (bad).

He is right, of course. Linear conversion funnels are great for shopping carts where you have to complete a series of steps in a specific order, but that is really all they are good for. Looking for common multi page paths in pages that are not part of a defined process is a pretty big waste of time. The hub and spoke model, whether your hub is a single page or a group of pages, is usually a much more productive use of time because it reflects the reality of what is happening in most cases.

Actually, that brings up a good refinement of Shane’s argument. The “scope” of the hub and spoke can and should change. At the most macro level your hub and spoke model shows how people are getting to your site. The hub is your entire site, the spokes link out to the internet. Zooming in a bit you can see how people move from your entry pages towards conversion, or to dead ends (hubs are the entry pages, spokes are where people go from them), The final zoom in would be figuring out how people end up on your conversion page. In this case the flow of the model is reversed, the hub is the conversion page and the spokes are incoming links from other pages on your site. The metaphor we use internally is that of ripples on a pond. The conversion is where the rock hit the water, the concentric circles are the various levels of “zoom” as you move farther and farther out.

This is just a model and it doesn’t work for every site; there will always be exceptions. On some sites the traditional linear conversion funnel fits perfectly. But I would argue that on most sites a hub and spoke model of traffic analysis makes much more sense.