Blackbeaks Blog….All things Analytics

In my last post I discussed a debate that has been going on in Web Analytics Demystifieds forum. I think Neil Mason has touched on an important point in his post, one which is reflected in later posts.

“Visitor and customer behavior is generally irrational and unpredictable at the individual visitor level. At the aggregate level, you lose out on the interesting stuff, the interesting stuff lies somewhere in between at the segment level. At this level, it’s more predictable, it differentiates and it’s useful.”

He has from an analysis standpoint hit the nail on the head there. The most interesting stuff is when you begin to understand what drives the individual to perform an action. While its unpredicatable, it’s very interesting in my view, it’s the closest we can get to actually interviewing the person, without talking to them. Segmentation is critically important, because by looking at the segment of people you’re most interested in you can begin to understand what their problems are. Aggregate data on its own is useful to flag what you should be segmenting but the least interesting in termsof overall analysis.

This brings me to how you could visualise the conversion process in terms of linear funnels, non linear micro actions leading to conversions and what I’d like to see in all web analytics tools in the future.

Firstly lets define a linear relationship and a non linear one.

The Linear Funnel

The linear funnel reports the conversion between point a and point b within pre-defined steps.

Shown here the conversion point is red and the steps (lets imagine moving from page to page) toward that conversion point are black. There is a pre-defined entry point and clear conversion point. The conversion point shown in red in this diagram is the macro action.

The micro actions are the points in between shown in black. These micro actions are the pre-defined steps in a linear conversion funnel.
Linear models such as this are very common in most web analytics tools and are undeniably useful in pointing out abandonment across a defined process such as a shopping cart.
By looking at the individual and how they navigate this scenario (or how they naviate away from this scenario) at this stage you can attempt to see where issues have arisen for them. Tools like Speed-Trap for instance will let you see where people have clicked and what people have typed into forms allowing you to determine whether there are problems. Most other tools are designed to test changes and see the effects.

When you look at an aggregate view you might be able to determine that you need to view people who navigated more than 5 pages for instance because then it might give you insight as to why the drop off happened between the first and second micro action.

The segmented data (those visitors who have viewed more than 5 pages) are the most interesting and you will probably see that more of those people have completed the scenario. Though then again you may not!

However there are problems with this. Where this model falls down is that they don’t measure the visitors that do not enter at point one (the first part of the process). On this scenario shown right/above if the visitor didn’t enter the website at the entry page measured at the top of this funnel they wouldn’t be recorded as part of the scenario.

Also it’s not a good visual representation of where visitors leave the process. It simply shows you the figures, it doesn’t say where they went in an obvious way, there is still a lot of work for you as an analyst to do, after you have determined where people are abandoning the linear funnel. Which brings us to the crux of the story, measuring a non linear process.

The Non Linear Funnel

The non linear conversion funnel reports the conversion action regardless of the steps, entry page or micro conversions taken to get there.

Lets visualise the diagram on the left. As before each black square represents a page and the red square represents the conversion. What this diagram is illustrating is that the visitor has arrived at a different entry point than previously and has navigated very differently to get to the conversion. In a non linear process this visitor would not be counted in the scenario and therefore you wouldn’t know about what her issues were.

The non linear process shows that she has navigated 9 pages in order to reach her goal. Each line represents a micro action which she has taken, none of which would be recorded in a linear funnel shown above. This in my view is one of the main problems with web analytics tools today. Most systems don’t allow you to track scenarios like the 9 step figure on the left.

Here is one that does.

Visual Sciences shows the whole website around the linear funnel. The Linear funnel model might have steps 1-6 defined previously and track how people traversed through those pages. Visual sciences however not only tracks 1-6, it will also track anywhere where visitors deviate from this path. It will also track if a visitor comes in on a different node than the main entry point shown by the largest line in this screenshot.

In the case of measuring the single visitor with this model you can see everywhere the visitor went easily and what issues they faced. Customer segments could be set-up to find out immediately how their patterns of browsing your website differ from the aggregate numbers thus allowing you to gain more insight into the personas you’re trying to persuade. The hub and spoke method that Shane Atchison talked about could easily be measured using this kind of tool. However the majority of customers don’t have Visual Sciences due to its price. (PS Note to WebSideStory; I’d like to HBX incorporate this kind of technology into it’s standard AMS offering).

So What Next?

What I was trying to get at with my last post about Marakov chains and web analytics was simple. Do we see other vendors incorporating the visual non linear process measurement described above into their systems, but more importantly would they also develop into predictive tools?

If we could predict based on data and the Marakov chain theory how people are best served by our websites, I think it would be a huge step forward.