The previous post in this series on the “Convergence of Digital Analytics and Database Marketing” was a pivotal one. In it, I described the details of Semphonic’s Two-Tiered Segmentation scheme. A two-tiered segmentation starts with a traditional audience segmentation (Visitor Type) and then sub-divides it with a second layer based on visit intent (Visit Type). In that post, I showed a sample Two-Tiered Segmentation and explained how Two-Tiered Segmentations should be at the heart of almost every Web analytics activity. I explained how KPIs take on meaning only when removed from a site-wide context and placed within a Two-Tiered Segmentation matrix; how that same matrix forms the basis for a good A/B or Multivariate Testing Plan; and how the matrix should be the framework for Management Reporting and for your broad site analytics projects. In short, the Two-Tiered Segmentation replaces nearly all of the standard Web analytics best-practice with a completely different (and much better) paradigm.
In that post, I used a fairly straightforward example of a Two-Tiered Segmentation based on a Financial Services example. I’ve found, however, that sometimes people in an industry different than the example given will tend to read too much into the choice. They’ll think that while a method might apply to industry X, it isn’t relevant to them. Sometimes, that’s even true. Not every method is generalizable across every site type.
Two-Tiered Segmentation, on the other hand, is generalizable - not to every site type (and I’ll discuss in a later post cases where it isn’t appropriate), but to the overwhelming majority of site types. I’m going to go broad-brush and say that 99% of the sites that Semphonic deals with (enterprise sites) can, should, and must be studied with a Two-Tiered Segmentation.
One of the very first Two-Tiered Segmentations we delivered was in the hospitality industry. Unlike my Financial Services example, where the Web site supported users with fundamentally different relationships to the business (Advisors, Plan Managers, High-Wealth Investors, etc.), that isn’t as common or important in Hospitality. A good Hospitality Visitor Type Segmentation might start with something as simple as this:
Each of these basic Visitor Type Segmentations can and should be further sub-divided, but I’m going to focus on the Existing Customer group because I think it’s the most interesting.
There is no one right way to create a segmentation scheme. In sub-dividing Customers, I might reasonably choose dimensions like Existing Relationship, Demographics, Travel Personas, or Potential Value. All of these are plausible and for different analysis purposes potentially interchangeable.
On the other hand, the best answer usually isn’t “All-of-the-above.” The more sub-segmentations I create for Visitor-Type, the more complicated every report and analysis becomes. If I over-divide my population, I can begin to hide the forest behind too many trees. In this case, I'm going to further segment based on existing relationship:
Why this focus on online vs. offline and Loyalty Program? This segmentation represents a fundamentally different population in terms of our expectation of online success, it captures a key business initiative (to move offline customers online), and I expect that it would most appropriately drive our high-level database marketing and testing strategies.
I want to dwell on that last point for a second. In an earlier post, I explained why capturing Meta-Data in measurement implementations is essential to driving good analysis and database marketing. The overwhelming majority of consultants and experts doing Omniture or other Web analytics implementations are just implementation specialists, they don’t know anything about real analysis and they regularly miss all of these capture opportunities.
The exact same situation holds for KPI developers. The vast majority of KPI developers never do (and have never done) any real analysis or database marketing. They treat KPI development as an activity designed expressly to support reporting. That is part of its function, but it’s not the whole story. The way you segment your population will have a profound impact on the thinking of the whole organization. It needs to set the table appropriately for all of your subsequent analysis and marketing opportunities. If I had chosen Travel Personas as my next segmentation, I’d have implicitly endorsed a very different database marketing strategy than if I choose Online/Offline followed by Loyalty Member.
The extent to which your work in creating a Web analytics infrastructure, designing KPIs, and building a Management Report drive or hinder your ability to move up the analytics maturity curve is poorly understood. They are deeply related.
At Semphonic, we like to show how the basic Maturity Curve:
Is better represented with additional steps and a significantly different slope:
But even more fundamentally, we try to point out that the WAY you move up the lower stages (-3,-2 to -1 and 1) has a profound impact on your ability to go higher.
You can get to Stage 1 without good Meta Data capture and without KPIs and Reporting based on a Two-Tiered Segmentation. But it will be much, much harder to move up the curve from there or to do worthwhile work at the stage.
I’ve often talked in the past about the “analysis barrier.” The challenge organizations seem to face moving beyond reporting to analysis. The barrier isn’t an illusion, but it exists largely because organizations have taken the wrong path (site-wide, “actionable” KPIs and technology implementers) to move up to Stage #1.
If you set the table with pitchforks, it’s hard to eat a good dinner.
Organizations have tried to dodge the analysis barrier by aggressively pursuing Multivariate Testing. It doesn’t work. They just end up repeating the same mistakes – focusing on site-wide tests and banal creative experiments.
To break the analysis barrier, you need to make sure your entire program is setup the right way FROM THE BEGINNING.
So when I pick Online vs. Offline as my next segmentation level, I’m making a profound business decision. For another company, in other situation, with a different set of web properties, I might pick very differently. Nor is it obvious or provable that I made the right decision in any particular case. This is the "art" of Segmentation and only successful use in practice justifies (without proving) any particular decision.
In practice, we'd typically extend this Segmentation with a value dimension, but I’m going to stop here and turn to the 2nd Tier of the Segmentation, visit type.
We use a combination of existing Site Use Cases, Site Walk-throughs, Opinion Research data, and Behavioral data to create the 2nd Tier. With most large enterprise Web sites, there are a surprising (even disturbing) number of different visit types. Web sites often cover multiple lines of business, multiple business functions, and multiple corporate endeavors. You simply cannot accurately capture this reality with 3 or 4 basic visit-types.
In most two-tiered segmentations, we identify 20+ different types each of which account for at least 1% of site traffic.
At a simplified level, however, we are generally able to group the visit types into a smaller set of uber-types:
Every single one of these Visit Types has a distinct behavioral signature, a distinct set of applicable site content and functionality, a distinct set of KPIs and relevant metrics, and a distinct set of testing and database marketing opportunities.
Some of these visit types are easy to identify. Corporate visitors (job seekers, press, etc.) usually leave an obvious content footprint. Job Seekers will look at a great deal of content on the Website, but they will ALWAYS look at job information. On the other hand, no one BUT job seekers will ever touch jobs content.
Other visit types present difficult challenges when signaturing. For most Hospitality sites, it’s extremely difficult to tell the difference between a visitor trying to find the phone number for a hotel that’s already been booked and a visitor looking to make a booking at that same hotel. The basic behavioral trail(Home Page, Area Search, Property Detail, Exit) is often identical, and only the integration of offline data can possibly solve the problem.
It’s a critical problem to solve, because although the behavioral record is identical, the success measure and the opportunity are not. Your chance of selling a room night to an existing booker looking for a phone number? Forget the slim – it’s nonesky. When you include those visits in your Look-to-Book rate, you’re shortselling your site in a big way and making yourself vulnerable to serious errors and misinterpretations. As customers use your web site for more post-booking support, you can misled into believing that your Conversion Rate is falling. In effect, you'll misinterpret an improvement in your online customer support with a deterioration in your online booking efficiency. That's a disastrous error to make but it's almost inevitable with a site-wide KPI approach.
Am I beating a dead horse here? Good, because that horse needs to be six feet under. Rigorous visit categorization is the KEY to good reporting and analysis.
Putting all of this together yields the full Two-Tiered Travel Segmentation:
In the gold boxes, I’ve shown the top success metric for each combination. The ONLY Visitor/Visit Type intersection that yields the same success metric for every audience type is Potential Booking sessions. Yes, Look-to-Book is indeed important. On the other hand, it’s ONLY important for a specific type of visit. Mix in all those other visit types, and all you have is, as the Grinch once put it when imagining the sounds of Christmas morning, “Noise, Noise, Noise, NOISE!”
Analysts should hate noise the way the Grinch hated Christmas. Noise makes for fuzzy thinking, bad decisions and analysis paralysis.
I’m hoping that all this seems to land somewhere between easy and obvious. Obvious it is, unless you’ve been listening to too many Web analytics experts. Easy it really isn’t. Not only is there great art in getting the best dimensions in your segmentation, the process of identifying and categorizing visit types is challenging indeed. Before I take that up, however, I’m going to describe Two-Tiered Segmentations for industries that may not seem quite so straightforward: Media Sites and Public Sector sites.
[BTW: A few weeks back I did Emer’s Silly Series Profile. I was on vacation when it hit but I’ve been meaning to point it out. If you want to find out how my early experience in politics scarred me, the most important part of a superhero’s ensemble, the profession I should be pursuing, or the many uses of philosophy, I encourage you to check it out! These are a hoot and I was both dangerously and disingenuously honest (see my thoughts on "deeply shallow").]
Wow, what a detailed series of posts. I've been advocating a segmented look at customer activity for a long time, with limited success internally. Site Wide KPI's and 'Best Practices' are the most overused (and irrelevant) terms in Web Analytics. I find it very interesting how executives and other managers will pour over detailed reports breaking down offline sales data into a multitude of segmentation's, but when it comes to the web they revert to such generic terms and requests. I've seen a lot of money being spent on reports which end up being pretty meaningless.
What has been very helpful from reading this series is your encouragement in structuring the two-tier approach. Most of what I've done previously has been a single tier. The challenge I have is putting together a tier hierarchy for our web data which approximates some of the segmentation we use in our brick and mortar business. They won't and don't need to be exactly the same, as they have two different functions, but one of the great learning's I've come to appreciate over the last several years working with web metrics is the need to explain things in terms/concepts other people throughout the company already use whenever possible. It's a great feeling when you're trying to explain something to a manager and you convert the 'web' talk into company talk and see the light go on in their heads all of a sudden.
Regards,
Cleve
Posted by: Cleve Young | May 10, 2011 at 11:41 AM