Further Thoughts on Data Warehousing

Late last week I got together with the Beyond Web Analytics guys again and did a podcast with Jason Rushin from Quantivo on the recent series of posts regarding data warehousing of web analytics data. As I thought about that broadcast and the three blog posts (Avinash, Jason and my own), I felt like the biggest missing piece was any really substantive discussion of why you MIGHT actually warehouse data. Adam Greco was pressing for this on the podcast and I think he’s right - it's pretty much missing.

I’m fairly satisfied that my last post is persuasive in responding to the arguments against warehousing analytics data. But one of the drawbacks to writing in response to someone else's post is that you end up using that framework and not necessarily arguing your own case.

So today, I wanted to discuss the class of questions, analysis problems, and marketing opportunities that are better supported in a custom warehouse than in a traditional web analytics solution.

Questions a Warehouse can Answer

When web analytics professionals start talking warehousing, there’s a natural suspicion that what’s really at stake is just a bunch of fancy tools designed to solve esoteric problems. So I think it’s important to lay out the types of questions that a warehouse might address that aren’t readily answered in a traditional web analytics solution. Taking a look at the questions may give you a better sense of whether you really need a warehouse or not.

Here's a list of questions I came up with that I think are either much easier, or only possible, to answer with a data warehouse:

• Do I have more visitors increasing in usage or decreasing in usage?
• For visitors who decreased in usage, what are they doing less of?
• What did visitors do in the 1st (and following n) weeks after registration?
• What did visitors do in the session after a purchase?
• If a visitor is coming to site more often than before, are they more likely to purchase?
• Is it better for a visitor to view product features in their initial visit to the site or is it better if they view product features in a subsequent visit?
• Are visitors who view multiple different products in a session less likely to convert than visitors who view similar amounts of content but focus on a single product?
• When, in their session, did visitors tend to use Tool X (say internal search) – and did the tool perform better when used early in the session or late in the session?
• If a visitor came to my site on PPC branded search in May, what did they do in June?
• For visitors who are most interested in Sports, what is their second most common content interest?
• Is it better if visitors viewed Page A and then Page B or is it better if they viewed Page B and then Page A.

As I hinted, it’s not impossible to answer all of these questions with a web analytics tool. But most of them will require special tagging and some pretty clever analysis – and most will also require a full enterprise web analytics solution like Discover, NetInsight or Marketing Warehouse if you are to have any chance at all to get at them.

Do these seem like esoteric questions to you? Are these the kind of questions you’d expect to have to hire a team of professional consultants, re-tag your website, and then struggle with the tool to get at?

The point I try to make when I talk to people about data warehousing is how basic and important the questions you can’t otherwise answer really are.


Targeting Opportunities a Warehouse can Support
Data Warehousing isn’t all about answering questions. As an old database marketing guy, I firmly believe one of the biggest advantages to warehousing online data is the vast array of targeted marketing opportunities it opens up.

A warehouse creates targeting marketing opportunities in two ways. First, a custom warehouse is much easier to integrate with your outbound marketing systems. But the story isn’t just about integration. A warehouse can easily support targeting rules that are challenging or impossible in a web analytics solution.

Here’s a short-list of targeting methods (you can generalize to whatever your business is) that are virtually unsupportable without a custom warehouse:

• Select visitors who are most interested in Lawn Care Products based on view behavior
• Select visitors who are most interested in Lawn Care Products based on purchase behavior
• Select all visitors and assign a code to each that shows the category they are most interested in so that I can match offers
• Select all visitors who demonstrate 20% lift or more when offered a discount
• Select all visitors whose interest in Lawn Care Products seems to have declined in the last X months
• Select all visitors who have a new interest in Lawn Care Products when we have track-record for them and no previous interest is shown
• Select all visitors who have viewed the Toro 1500 page more than 3 times and haven’t purchased
• Select all visitors who have spent more than 15 minutes in the last week looking at Lawn Care Products and more than 20 minutes looking at outdoor recreation equipment

It’s pretty much this simple: if you want to target visitors by how much they care about or do of something, whether their interest is increasing or decreasing, and whether their interest is new or old, then you need a warehouse. As far as I can see, that’s pretty much anything you might actually want to target on.


Integrations best done in a Warehouse
For many warehouse efforts, integration is the cornerstone of the justification. Doing integrations isn’t done just have to a single place to put data. When you integrate data, you open up new questions, targeting opportunities, and analysis methods that otherwise don’t exist.

Web analytics tools (at least the enterprise ones) do a pretty good job of integrating certain types of information. Email integrations are both easy and well done. Online Survey integrations aren’t too bad either. Beyond that, however, you’re likely to struggle with integrations. And there are a number of important integrations that simply aren’t suitable for SaaS web analytics solutions:

Customer Integration: Combining your customer records with online behavior
What you can do with it:

• Find out how your best customers use the web
• Find out whether particular relationships drive visit types and site success
• Find out whether Tool X is useful (and used) by customer types
• Determine whether site visits drive incremental customer lifetime value
• Figure out whether online/offline promotions cannibalize the opposing channel
• Determine the offline impact of web site behaviors
• Target customers with view behaviors different from purchase behaviors
• Target customers with offline purchase behaviors that haven’t translated into online behaviors
• Target customers whose total relationship is waning

Call Center Integration: Combining your call center data with online behavior
What you can do with it:

• Determine if the web site supports real call avoidance
• Measure the value of online customer support pages
• Identify topics not well covered by the online channel
• Target groups of customers who might use the web site to self-service but haven’t

Tracking Study (panel) Integration: Combining your panel tracking data with online behavior
What you can do with it:

• Determine the long-term impact of web site relationships on attrition
• Measure the brand-impact of the web site
• Determine if changes in the business environment have made web branding less impactful
• Target the types of visitors with potential attrition issues

Econometric, Spend, and Environment Data
What you can do with it:

• Measure the correlation and factors that drive web site success and determine how much marketing efforts can actually drive outcomes
• Determine the optional media mix for mass media campaigns
• Understand the relationship between site awareness and brand awareness
• Time the best moments and channels for specific types of customer campaign

Some of these (like customer integration) are so basic and obvious that’s it almost redundant to discuss their virtues. Others, like the integration of tracking studies and econometric data are less obvious but still provide numerous opportunities for important learnings.

Of course it’s possible, with the enterprise web analytics tools, to stuff almost any kind of data into them. Unfortunately, they simply do not provide flexible data structures, flexible analysis techniques, or a robust data model for actually taking advantage of or studying any of this external data.

Types of Analytics Unavailable Except in a Warehouse

Which brings me to my last topic – the tools and analysis techniques that you don’t usually have available to you in most web analytics tools but that are generally possible in a warehouse. I’m not going to argue for why each of these is important or compelling – I trust that most enterprises will have experience of a number of these and can reasonably judge for themselves how valuable these tools/methods are:

• Correlation
• Variation
• Regression
• Factor Analysis
• Cluster Analysis
• Decision Trees
• Data Visualization
• Scoring
• Time Sequence and Time Series Analysis
• Predictive Modeling
• Grouping

Etcetera, Etcetera, Etcetera. I really could go on and on here. But, in the final analysis, where I started is much more important than where I finished. It’s the questions you can answer and the actions you can take that really matter. All of these techniques and tools are enablers. They are important for what they can accomplish in the hands of an analyst - not for what they are.

At one time or another, we’ve worked on projects that involved nearly all of the questions, opportunities and tools I listed.

We’ve done full-scale call-center integrations for some of our biggest clients. And the questions and targeting opportunities I put there were borrowed right out of our  experiences.

We’ve done econometric modeling for companies too, figuring out what market conditions and marketing mixes were effective in driving web traffic and conversion – and believe me – we didn’t do it using a web analytics solution. And we’ve asked and answered nearly identical questions to every single one I put in my initial list – sometimes many times over.

These are, for the most part, fundamental questions that drive directly to important business decisions. We’ve tried (and sometimes succeeded) in answering these questions within enterprise web analytics solutions too. But it can be frustrating, expensive and difficult to do. Give an analyst the wrong tool and the wrong platform, and a good analyst may still get the job done. It will just take far longer and the analysis will likely be riskier and less precise. And I should emphasize the word "may" because we've failed plenty of times when we just couldn't get the tools to do what we wanted.

So if you have a lot of the kinds of questions or needs I’ve discussed above; if you’ve reached the limits of what you can accomplish with your Web analytics tool; it may well be time to consider whether the risks and expense of custom data warehousing still outweigh the potential benefits. Building your own data warehouse for online data is no joke; it's all too easy to fail. But if you've reached the limits of what you can accomplish with your existing solutions, you should have most of the knowledge and experience necessary to succeed. And in today's market, there are a plethora of interesting, innovative and very workable technology solutions that can help make you successful.