Marketing to the 'Anonymous': Using Cluster Analysis to Fuel Behavioral Segments

The Value of Clickstream Analysis
Clickstream is the traditional strength of web analytics. Analyst new and experienced understand the ubiquity of page views and visits. Arguably all three are the core components of any analytic package from Coremetrics to Clicktracks. We've all made a living off of this kind of data, as it was the very first piece of informaton that could be mined from logfiles. But that was a very long time ago. How useful is clickstream analysis today? Do these historic models still stand the test of time? Can you reasonably market to anonymous users? Or does our insistence on doing so explain the anemic conversion rates we routinely record?

Graph 1.1 (Traditional Visitor Bar Chart from Omniture's Discover 2.5)

A New Perspective on Unique Web Visitors

The mature web analyst understands that in order to market to consumers we must 'personify' our web visitors. The familiar edge of bar charts across spreadsheets may be enough to appease a pedestrian use of analytics, but it's a very mild distraction from the real question.

What can we learn from online behaviors? How can we profit from these visitors? How do we derive a communication strategy around visits? What vehicles may we use to profit from these customer paths? Are there patterns in the data that help us to segment millions of visitors into meaningful structures beyond the web analytic framework?

Graph 1.2: Typical Process Funnel based on a step through process of a shopping cart/Discover 2.5

A Slightly Different Take on Web Profiling
Personifying your web visitors as unique is the first step in creating behavioral profiles. One may develop meaningful customer profiles by leveraging attributes such as keywords, time of day, entry page, recency, clicks, orders and the like.

True, web analytics does a fair job of creating buckets that can be analyzed and applied to process models, but the problem is that web analytics was not meant to maintain complex historical attributes at a visitor level. That kind of data quickly becomes indiscernible. But if you can identify a unique key for each visitor you have a formula for communicating with a customer, rather than continuing to measure and market to anonymous visitors.


6d2b2bb62df2739dc791b0ccd64611bc = unique visitor ID x 10,000,000 visitors/10x web analysts.

Its absolutely not a leap of faith to take a unique visitor ID and append every piece of available data to it, but clearly after a few hundred records over a 3 - 6 month period, this could become taxing to even the best web analytic tool. Besides, web analytics can not discern this level of detail to identify the level of customer information we ultimately require.

Moving Beyond One Size Fits All Marketing
So how do we go from marketing to technographic segments (segments defined by entry pages/paths, etc) to influencing customer segments? Fortunately for mere mortals, machines (SPSS or SAS) can perform much of the invasive surgery required to make sense of this data. All of that precious clickstream information can be gleaned from our unique visitor and organized into clusters. Now we can take a look at customer reach by applying the COMM (creative, offer, messaging, monetization) principle for marketing optimization. The question then becomes which messages are appropriate to each unique customer segment and how do we measure the lift between theses discrete groups.

Using Cluster Analysis in Web Analytics
The primary goal in cluster analysis is to find meaningful structure in exploratory data and yes...this would be particularly true in web data.

Cluster analysis is an exploratory, analytic technique used to classify observations into finite and, ideally, small numbers of groups.

So this: (illustration pulled from SPSS/Visualizing Clusters/cluster analysis demonstrating distance measures based on response pattern similarity)

Can be used for this: (illustration from Omniture Discover 2.5/Segment Builder/Claritas Prizm)

In Summary
From linear behavioral segmentation to cluster analysis through SPSS and then back again.

There are all new techniques to be used in Web Analytics. Our industry is still so very new, but we must continue to borrow over techniques and approaches from other, well established industry processes.

13 Defining Observations About Web Analytics

I can only describe the following as simple anecdotes that I have observed to be true over time.

I have found it helpful to document and to refer to them from time to time. In 3 months, I am certain there will be others.

Enjoy.

• Web Analytics is change management. And at the end of the day, if its truly your passion, web analytics is performance management. If you aren't changing anything, don't bother to report on everything. Seek improvement above all things. Don't define success by your peers. They may have a very different way of determining it.
• Be thoughtful. Think COMMSS. Creative. Offer. Message. Measurement. Segment. Score.
  
• Web Analytics is application development. Nothing on the web lives very long without growth. The web grows in the areas you feed it. Therefore, you must consistently monitor the integrity of the data. A credible analyst depends on it and is ultimately bound to it.
• Web Analytics is diagnostic. I have often used web analytics to confirm or disprove my own observations.
• Web Analytics is not reporting. You'll find yourself outsourced shortly if you truly believe so.
• Web Analytics is not strategy, but is the linchpin of any successful channel roadmap. Don't confuse web analytics for strategy, although it is often used to define it. Channel under performance is the norm because of this fundamental misunderstanding.
• Web Analytics is profitable. The last 3 analysts I interviewed were looking for salaries well north of 100k! None had more than 5 years of experience. Several had 3.
• Web Analytics is Hard. Yes, it is. The data offers possibilities limited only by your imagination. The more you use the data to manage up, the more difficult analytics becomes.
• Web Analytics is sell. There are probably 5% of all web analysts that can comfortably stand in front of large crowds, diverse backgrounds and varied business models....and smile..confidently delivering that impact message. Don't fool yourself. Web analytics is sell. Know your tools. Understand analytics. Experience the data, but sell it and in doing so, you sell yourself.
• Web Analytics is optimization and optimization is math, whether or not you realize it.
• Along the same lines, optimization should be the color of money. But financial data data exists outside of web analytics! Don't confuse revenue (net present value) for profit. Don't confuse conversion with success.
  
• Web Analytics is not conversion. Conversion is profit. You must follow conversion to its end and that can never be found in a Web Analytic tool.
• Web Analytics is too linear and too limiting to those with data experience. Vendors neatly package these branded applications. The data is sampled and formatted in ways that make it easier for the marketer to manage and yet, my experience has always been that the most powerful drivers to web analytics come from areas well outside of these applications.

Feel free to rip the cover off of Web Analytics.

Measuring Landing Pages through Web Analytics

Recognizing Homepage Noise

So much of what exists on your website is noise. Just count the number of messages you currently have on your homepage.

Banners, imagery and competing offers distract customers from taking the desired conversion action. It's no wonder conversion metrics are so poor!

Putting generic messages in front of customers meant for everyone is inefficient. It's simply irresponsible marketing. Sort of like placing water into a leaking bucket; your message is quickly dispersed. When looking to influence customers action, Brian Eisenberg talks about employing the WIFM principle (what's in it for me), but most of our site experiences can be summed up as 'WSIEC' (why should I even care?).

Creating Effective Landing Pages

By funneling your customers, you create a more focused and inviting customer sales experience. Similar to breadcrumbs, landing pages are more than just stripped down versions of existing copy. Landing pages can be a highly effective way to lead customers down predetermined information pathways capable of qualifying customer intent and/or influencing conversion.

The Role of Web Analytics in Landing Pages

In web analytics, studying and modifying landing pages is often key in improving favorable customer scenarios, i.e downloading a white paper, requesting client contact or completing a purchase.

Conversion Defined

Remember that when customers meet their goals, ultimately your business does. Conversion only occurs when you profit from the actions of your customers.

Landing pages should routinely be tied into marketing campaigns, special offers, or newsletters.

Here are some components you should think about when creating an effective landing page.

• The Hero Shot (what the customer gets for taking part in the conversion process)
• Value Prop or Benefit Statement
• The Message (includes the qualifying statement, i.e the reason you are here)
• Offer
• The objection (believe it or not, not everyone wants to give you their credit card)
• Call to Action (make it stand out, make it painfully obvious, outline the next steps)
• Trust identifiers

If your current landing pages aren't driving conversions with regularity (double digit conversion rates) perhaps that should be your first web improvement project. In the hands of an experience web analyst, landing pages can become effective conversion tools.

7 Rules for Keeping Web Analytics Relatively Simple

Okay, let's be honest, 6 out of 10 web sites fail miserably according to a published Forrester report way back in 2003. Not surprisingly, some things change while others remain the same.

The unfortunate part about web site management is that Web sites fail.

But web sites often fail because they are simply mismanaged. Even after 10 years, there's still a lack of understanding around the channel.

So long as conversion rates hover at 2.5% there's still a lot of stars to be earned. Site managers still can't readily identify key performance metrics on their site.

Are you recovering from your 3rd redesign in two years?

Sorry to say it, but you aren't managing your business. Don't feel too bad though you are in the majority.

You can prevent this from occurring by simply focusing on Optimization in '08. Stop recovering! Stop building!

Getting More Out of Your Website through Website Optimization and Conversion

Let's define Web Site Optimization as the framework for identifying, analyzing and correcting potential problem areas on your site.

The smart site manager restricts these "problem areas" to activities that either generate profits or reduce spend, i.e improve the percentage of completed forms, increasing the number of site registrations, decreasing home page abandonment, or increasing the number of newsletter subscriptions.

Creating a Better Framework for Managing Your Website

It was really less than a decade ago, that logfile analysis was the only way to manage web sites, but in fact, most logfile analyzers were developed to measure server activity, not business performance.

Web analytics has changed how we measure the online business, creating new opportunities to better understand customer interactions.

If you really want success. Define it. Then work towards it.

Below, I've outlined what may be considered a framework for better managing and optimizing your site.

Rule 1 Prioritization
What are you trying to optimize? Start small. Develop your objectives early so you don't waste a lot of valuable time. Prioritize those objectives around activities that provide the greatest value and where possible the least amount of effort. If you try to optimize everything, you won't accomplish anything.

Rule 2 Apply the marriage of metrics
Develop key performance metrics around what you want to improve. Its okay to use universally accepted KPIs to run your business, but eventually you'll discover those that are unique to your needs.

Rule 3 Understand Your Own Business Model.
You'd be surprised at how often site managers don't know this. There are really 4 business models; Lead Generation, Content, Commerce and Self Service. Many websites are complex hybrids of all four.

Rule 4 Establish Site Benchmarks and Set Targets
After a few weeks of collecting data and defining problem areas, you'll need an understanding of where you are, before you can determine where your business can go.

Rule 5 Change Management
All of the data and performance metrics in the world won't help you if you don't take action. Don't let the sheer volume of reports you can run be your only success metric. Be aggressive. Stay focused.

Rule 6 Get the Right Web Reporting Tools for the job.
Find the right tools for the right job. There are many site management tools with strengths and weaknesses. Look to acquire those tools that truly fit your understanding of the business.

Rule 7 Hire Experienced Talent
Hire smart. They won't come cheap. Experienced analysts can generate salaries north of 100,000.

Look for web analytic professionals to measure and manage your site with at least 3 to 5 years of experience. Ideally, that person should be as familiar with reporting tools as he or she is with managing projects.

What's the Difference Between Multivariate Testing and Behavioral Targeting?

One of my colleagues recently asked me the question, "We've invested this money in behavioral targeting, but don't we already have multivariate testing software"? What's the difference between the two?"

Tough question, but I used the following analogy to explain it.

Let's start with Multivariate Testing. Consider any city block filled with restaurants. During a given night patrons will populate them based on the level of activity and/or personal food preference or even reputation to name a few. The perceived interest from a population is registered through multivariate testing. Each time a customer walks through the door, a vote is registered for one restaurant over another.

Over time, MVT will understand which restaurant should be recommended to a specific group (a segment of visitors) based on the level of activity and or overall traffic generated. Ultimately, MVT compares the results of this targeted group over another sampled group(s) to determine the overall performance lift.

Based on observing a group "over time", MVT will suggest a specific "champion" restaurant (most often selected) over all others.

In truth, MVT collects a number of data points, including page referrer, keyword, browser, time of day, etc, but it cannot use that information to leverage decisions on its own. It does not maintain that level of sophistication. For lack of a better term, it is dumb.

Eventually MVT will require human intervention from the marketer to further take advantage of usage patterns that come to light in the response data, i.e. creating new segments to optimize testing efforts.

Comparing Behavioral Targeting

We've established that multivariate testing looks at a given population of visitors over a specified length of time. Conversely, behavioral targeting tends to look at individual behavior, comparing it to others that may have performed similarly in the past.

Now back to our restaurant analogy. Behavioral targeting looks at this same issue of preference, but from a slightly different perspective. It may not be enough to put you in any restaurant. It may be more important to put you in the restaurant that's right for YOU at that time.

Behavioral targeting attempts to narrow the list of possibilities by taking into account your previous preferences, the number of people in your party, whether or not you are with friends or you are on a date. It may even consider whether or not you are interested in French or Thai or even if you just want appetizers or a full entre.

And yes, it may even test price sensitivity to determine how much you and others that look like you are willing to spend.

You may be inclined to believe none of this matters, but the truth is that it does. Many of these factors are evaluated by behavioral targeting in real time, without human intervention. It is constantly evaluating all of the possibilities in order to display the creative message that generates the most clickthroughs, conversions, or average order value.

The problem with both of these approaches is that they require thought. Dare I say strategy? Marketers have to become better at tailoring messaging and varying use of creative. Sadly, this is the exception rather than the norm.

The secret to online success is that there is no secret. Work is essential if you want to generate remarkable results. Next I'll talk about how both can be exploited and how the two can be synergized.

Be sure to have a look at my diagram (PDF) which outlines fundamental differences between multivariate testing and behavioral targeting.