Web analytics best practices

We have all seen the dramatic improvement in the capabilities of web analytics tools over the last few years, but how should organisations configure their web analytics tools and processes to drive the best insights?

In this E-marketing Essentials interview I talk to Gary Angel, a web analytics consultant with extensive experience of web analytics implementation best practice for many organisations.

Semphonic's team of analysts provide recommendations on web analytics implementations for the likes of American Express and O'Reilly.

• Read Gary's blog for more of his insights on web analytics best practice.

Q1. What are the major challenges in deploying web analytics in 2008 as you see them Gary?

[Gary Angel, Semphonic]: Tools, once the biggest headache and most severe limitation in web analytics, have gotten markedly better. On the other hand, deployments have gotten significantly more complicated "€“ especially around data integration.

Increasingly, our client-base is deploying a software as a service (SaaS) web analytics solution like Google Analytics that provides basic tracking and reporting and whose function also includes the creation of a clean web data set to be sent back to their company and analyzed internally using more powerful tools. This increasingly means that WA implementations are bound up with IT. Avoiding IT was the real driver behind the SaaS movement in the first place; so it"€™s kind of an ironic trend.

Big organizations are still struggling with quality control and standardization of metrics and implementation when deploying large numbers of sites "€“ especially internationally. Standardization is really important in that environment. It"€™s a lot easier to build and maintain one standardized implementation than a hundred different ones. It"€™s also more useful since in large organizations insuring comparability is essential.

Still, I"€™d say that the biggest challenges with web analytics really haven"€™t changed all that much. It"€™s still mostly about effectively using the tools and analysts to do more than generate reports. Real analysis that drives business change is depressingly uncommon. Most organizations have no plan around the analysis they expect to achieve, have no idea of the resources they should devote to analytics and have no concrete expectations for what they are expecting web analytics to provide.

Q2. With tools such as Doubleclick Dart and Atlas tagging we are moving beyond the "€œlast click wins"€ model for attributing sales to different digital media channels. Are you seeing any standard approaches or best practices to weighting assists from different media channels?

[Gary Angel, Semphonic]: Attribution remains a significant issue for clients. In our view, there is no one right answer to attribution. We all know that first or last doesn"€™t cut it. But it turns out that channels interact quite differently for different organizations. It also turns out to be nightmarishly difficult to produce coherent reports on channel interactions that capture anything like reality.

Building an attribution model usually seems to involve a significant company-specific deep-dive analysis followed by the creation of a set of business attribution rules that are applied to ongoing reporting. This is an area where we tend to emphasize Analytic Reporting (reports that have a model built-in using programmatic code) to achieve reports that capture the complex reality but can actually be digested.

Traditional web analytic tools have also gotten more sophistication in attribution "€“ CoreMetrics and Omniture (for example) both let users do more advanced attribution.

Q3. What is your view on the new Google benchmarking service?

[Gary Angel, Semphonic]: Everybody loves benchmarking data and with good reason. It provides a context for web data that is often sorely missed.

On the other hand, the issue of comparability is much, much more severe than people think. We see implementation after implementation that is fundamentally messed up. I have a simple rule that any data set for which no one has direct responsibility for keeping clean will be seriously flawed. There"€™s no one checking this information or validating that sites have designed their implementations in any thing like a comparable manner.

Think for a moment about the controversy swirling around traffic numbers between web analytics solutions and panel solutions and tell me why you"€™d expect essentially unchecked data from multiple non-standardized GA implementations to be useful?

I can"€™t think of too many other tools where the potential and temptation for abuse and misuse of the information is as high.

Q4. Web analytics tools are often limited in their capabilities for understanding and optimising the value of returning visitors. Any practical approaches that can work across different analytics tools.

[Gary Angel, Semphonic]: This remains a significant issue in web analytics for two reasons. First, there are the data quality issues created by cookie deletion. This actually impacts new visitor analysis more than returning visitor analysis (at least over fairly short time frames "€“ it does make attrition analysis nearly impossible) since cookies are deleted at will but when persisting are mostly reliable.

However, if the return cycle is long, then return visitor analysis without sign-in is essentially impossible.Even where the data is useable, however, this remains one of the bigger weaknesses in web analytic tools. Segmentation is probably the key feature that does drive organizations to focus on high-end tools and even these are often surprisingly weak.

Here are some thoughts on best practices to deal with what is essentially a large matrix of problems:

• Code the date when key events occur in variables in your analytic solution. This allows you to create segments based on key date ranges "€“ something impossible to do otherwise in most solutions.
• When doing visitor segmentation based on Voice of Customer (survey) data, be sure to capture the online id in your research and match the segmentation back to the behavioural data. Building behavioural segments on top of VOC data is the best way to make your overall visitor segmentation actionable in the online world.
• Use a first party cookie where possible.
• Capture customer specific identification (like email or username) as often as possible and use it to re-integrate customer data.
• Don"€™t be afraid to take the data back and do visitor segmentation outside the WA tools. There are many more appropriate tools for this in the greater Business Intelligence world.

Q5. Web analytics tools have never been good at making recommendations on how marketers should improve performance. Are you seeing examples of where this type of intelligence is built-in?

[Gary Angel, Semphonic]: Thankfully no. Anyone who has ever used tools that try to make recommendations (in areas like SEO for instance) will know what I mean here. There is simply no rational way to automate recommendations.

It"€™s true that in a highly structured framework like a multivariate test or a carefully constructed PPC campaign there are significant opportunities for machine optimization. But there is a very large amount of structure put into place around those areas before automated optimization can take place. That"€™s the exceptional case, not the rule.

What we would like to see is tools doing a significantly better job of tying information into systems instead of treating each metric as if it has a discrete meaning. When we build reports, we focus on a system like site traffic or conversion.

The reports are designed to show the relationship of variables within the system and how each factor is impacting the overall system performance. This is vastly more useful than just showing metrics but, at the same time, it is nothing like making recommendations.