"Understanding the full Customer Journey" or "Building a 360 degree Customer-View" are the kinds of things that CEOs and Digital Strategists love to talk about. But it's all just pretty pictures and hot air unless you can figure out how to get the information in one place and, even more important, figure out how to manage and represent it coherently when you bring it all together. It's challenging enough to build a data representation for a single, complex channel like the Web. Creating a usable and understandable representation for five or more channels seems daunting indeed.
So the real trick is putting that information together in a warehouse in a way that actually makes sense. In my recent posts, I've been describing our techniques for representing Web analytics data in the warehouse. By using a Two-Tiered Segmentation and a set of RFM fields for each visit-type, we are able to capture most of the interesting information from a Web touchpoint in a very compact and convenient format.
However, everything I've described so far has been very specific to the Web. That's not terrible - Web data integration is at the top-of-the-list for most CDW and Analytic Marts these days. But even as the Website has grown to become vitally important as a customer touchpoint, it's been reduced in the sense that more and more touchpoints keep springing up. How should we think about modeling Mobile Apps, Facebook, Twitter, and Communities, not to mention offline staples like Call-Center, Branch, or Mail?
The framework we've developed for handling Website data is surprisingly appropriate to most or even all of these sources. In the Two-Tiered Segmentation, the Visitor Type is likely to stay constant across all channels. It's the Visit-Type segmentation that's really distinctive in our approach to Web data and that same concept can be applied to almost any kind of touchpoint.
In addition, the Recency, Frequency, Success model (essentially RFM) that I've suggested for the likely metrics within each Visit Type are applicable to almost any channel. Consider the following Visit-Type Segmentations:
In each case, I'll assume that what we store in the CDW is a set of RFM metrics for each Visit Type within each channel for every single customer. It's a fair number of fields per Customer, but it's still a VERY compact representation of the customer journey.
If we do set things up this way, we could easily answer all of the following questions about the Customer Journey:
- Which customers do Market Alerts on the Web but don't use the Mobile App at all?
- Which customers are currently doing planning on the Web but haven't had a Branch Consultation?
- Which customers used the Call-Center for a Change of Account but didn't try to do it on the Web?
- Which customers failed on the Web in a Change of Account Status but didn't call the Call-Center?
- Which customers are shifting from Branch to Web but still using Branch significantly?
and so on.
It's powerful stuff. In essence, we can answer any question about channel usage by function, success, recency or frequency. We can easily see the order in which visitors did something (because we have recency for each touchpoint) and we can see when they failed in one channel and moved to another. We can even see when WE think a visitor succeeded in a channel but they moved to another channel with a seemingly similar action.
This isn't magic. The underlying assumption is that we can track a customer as such across these channels. In the financial services world, that's quite likely to be true. In some other verticals, it quite likely isn't. In those verticals, you'll have to maintain separate customer records by channel and only merge them when a key joining opportunity emerges. Even so, it's vastly easier to do that when you've reduced the data from a channel down to a single, easily represented customer record.
The Two-Tiered Segmentation model generalizes surprisingly well. Or maybe it isn't all that surprising. All we're really saying is that any touchpoint can reasonably be described by the following data points:
It makes perfect sense when you put it this way.
The type of touchpoints that may be most problematic are ones that are either mono-focused or customer-inscrutable. Display-Ads, for example, don't fit the model in an obvious fashion. What did the Customer want to do when getting a display impression? Probably anything but get the impression!
Nevertheless, I don't think the model breaks-down so much as reduces to something simpler. In the case of Display, I'd probably be inclined to model all Display Ad impressions as a single touchpoint type (Browsing the Web) and use the RFM metrics underneath that.
If I did that, I could answer questions like these:
- Did viewing a Display Ad make a Customer more likely to open an additional account?
- Were prospects who saw a display ad after visiting the Website more likely to return to the site or open an account?
This seems to me pretty much what we hope or expect to get out the analysis of channels like Display as part of the Customer journey. So while the Two-Tiered segmentation isn't really necessary inside the channel, neither is it fundamentally broken. Some channels simply are mono-focused so that the Visit-Type dimension collapses into the broader channel. In the old days (but probably not today), ATM visits might have been similarly mono-focused.
What about social media? It actually fits very well into a Two-Tiered Segmentation. I might, for instance, choose to model Twitter like this:
I'll assume that the Success Dimension is a sentiment score and that we have our typical RFM model beneath each touchpoint type. With this data model, I can answer some really cool Twitter usage questions quite easily:
- Which Customers comment more about us than about our competitors?
- Which Customers are highly involved in industry conversations but not our conversations?
- Which Customers are more positive about the industry conversations than our conversations?
- Which Customers have the highest Financial Services mindshare?
- Which Customers are growing positive and interested in a competitor?
Each tweet is treated as a touchpoint (Visit Type). Cool stuff - and a powerful advance over a simple Twitter aggregation. One thing this model doesn't address is influence, but I think that's a topic for another day. Representing influence in the data model is fairly trivial (though I think it could be captured at the Visit Type or Customer-level or both) - the hard part is figuring out what influence means and that's really a question for Social Media analytics.
Let's sum up. The Two-Tiered Segmentation isn't just a really good Web data model. It turns out to be an excellent representation for almost any type of customer touchpoint. It can be applied to touchpoints as diverse as ATMs, Twitter, Mobile Apps, Call-Center and even Display Ads. For most of these, it drives significant value beyond a traditional representation and by providing a uniform framework for handling ANY customer touchpoint, it makes for a simple, compelling model in the CDW.
Two-Tiered Segmentation is the perfect framework for making that vision of the full customer journey and 360 degree view of the customer a reality. Not only does it generalize every touchpoint into a common framework, it provides a uniform metric foundation applicable to ANY and EVERY touchpoint. With a data model built around the Two-Tiered segmentation, you can answer a spectacular range of interesting questions about the customer journey. Best of all, you do so easily, with standard SQL or BI tools in a very performance efficient manner, and you provide an easily understood customer record that helps guide the marketer toward effective and interesting uses of the data. You just can't ask for more of a data model than that!
[We're getting pretty close to sold-out at X Change, so this is like "Last Call" on Friday night. If you're hoping to come out, please do register this week! I know the Conference is still almost a month away, but we have a pretty cap on attendance and we're getting very, very close.]
[Click here for a summary of this extended series on Digital Analytics and Database Marketing]
Comments