Module 2 | Web Analytics

In the last module, we spent a good part of the time reflecting on topics of web analytics and metrics to assess website performance for business purposes. As we saw at the beginning of the module, web analytics is the “objective tracking, collection, measurement, reporting, and analysis of quantitative Internet data to optimize websites and web marketing initiatives” (Web Analytics Association – It seems to be a defunct organization in 2020), and a plethora of metrics have been developed in the last 30 years for that purpose.

 

Web Metrics: Occam’s Razor by Avinash Kaushik

 

As I was reading about different audiences, traffic channels, visitor behavior, and business outcomes, I ended up exploring the blog of Avinash Kaushik: Occam’s Razor. The number of topics and perspectives in this blog is beyond what I could read in a single week of coursework, though I believe all these topics are tidily discussed in his two bestsellers: Web Analytics, One Hour a Day and Web Analytics 2.0. Thinking of web metrics and Key Performance Indicators, three posts caught my attention:

 

Six Web Metrics / Key Performance Indicators To Die For 

Your Web Metrics: Super Lame or Super Awesome?

Best Web Metrics / KPIs for a Small, Medium or Large Sized Business

 

Two of Kaushik’s takeaways recommend (1) using metrics that describe properties of our business directly (as opposed to having the analyst infer the value of such properties) and (2) making outcomes (i.e., conversions, task completions, and business value) your ultimate performance measures (that’s where your business raison d'être lies!).

The third post was particularly revealing as an overview of web metrics and their analytical suitability for business intel. Although I missed references to the science behind and validation of the rationale given by Kaushik (these may be in the books!), his recommendations may well be a matter of research. Overall, Kaushik maps 13 web metrics based on two dimensions: business size and the object of analysis, i.e., acquisition, behavior, and outcomes.

 

If the goal of a business is to grow (in size, diversity of services, geographically…) I lean to think that keeping an eye on metrics for larger businesses (e.g., the percent of assisted conversions) could (1) provide a business with insights on how it is doing if it was a larger entity, and (2) help it learn how to use those metrics before business expansion and as facilitating means to enable such a goal. Knowing what to do to achieve a goal seems more advisable than waiting to get to that point (if that ever happens) and act in consequence. 

In practical terms, adding more or less metrics to a dashboard doesn’t seem to involve significant computer resources for the technical assistant or the analyst assessing performance, but redesigning your dashboard once your business grows could become a costlier option.

 

Web Analytics…Web Modeling?

 

Over the last weeks, I also got interested in the analytical capacity of dashboards and web analytics software, specifically, of Tableau and Google Analytics given their popularity in business intelligence. I notice a strong capacity of these platforms for exploratory data analysis (i.e., descriptive statistics and data visualization) with cutting-edge approaches for interactive and dynamic (real-time) visualizations. Having a professional background in geovisualization, I especially appreciated the study of color that these solutions have made to provide colorblind-friendly visualizations (e.g., see this post). 

The not-so-good news are, I didn’t get to find a lot of functionality in Tableau and Google Analytics for statistical modeling, say, hypothesis testing and forecasting. It seems like analytics gets simplified to summary statistics and non-quantitative pattern recognition. In fairness, a few tools are available:

 

New predictive capabilities in Google Analytics [July, 2020]

Predictive metrics in Google Analytics

Predictive audiences in Google Analytics

Anomaly Detection in Google Analytics

How Forecasting Works in Tableau [March, 2020]

Not really an unconditional stalwart of statistics, data mining, and machine learning, but I would say that we, humans, need modeling tools. It is not unknown these days that human brains are prone to see patterns where there aren’t (look at the clouds!) and don’t specially excel when processing information from multidimensional data spaces, so we got AI! 

 

 

As far as Google Analytics go, I wonder if the limitation comes from the fact that analytical services are online. Could you imagine all the Google Analytics accounts ad properties running simulations and neural networks for predictive purposes? But if the issue is the limitation of online processing, that kind of defeats the purpose of analytics, especially now that storing all your data on your own server is hardly a computer capacity problem.

I would love to see a day when the kit of modeling tools in dashboards show a level of quality and diversity at a similar level to the kit of descriptive and visualization tools. Would it be awesome to use an entire business database to model how changes in the website, products, or marketing campaigns could impact the interaction of clients with the site and eventually key performance indicators? Wouldn’t be insightful to decompose purchasing patterns into global trends, seasonality, and specific events (how about the weather?) to minimize your predictive error? 

To do that sort of analytics, forecasting, data mining, and simulation still remain to take over in Tableau and Google Analytics. And I don’t mean a handful of classical statistical tools without a lot of user’s capacity to set their parameters. I mean a versatile, dependable, and fully integrated modeling module that can prevent analysts from dreaming of MATLAB, SPSS or TensorFlow when wondering “What if…?”. 

I’ll leave it here as I think how I'd go about to implement a dashboard for business modelers and forecasters.