blue-green background with a light bulb made of arrows and bullet points


Eight Great Data Science Problems to Solve in Digital Analytics

Mike Murphy

The digital analytics industry, while growing substantially, is not without some unsolved issues holding it back. Gary Angel, a Principal in Advisory Services, Digital Analytics at Ernst & Young, discussed tackling these important, often-ignored problems in data science as they apply to digital analytics at the 5th Annual Digital Analytics Philadelphia Symposium, held on October 29, 2015 at Drexel University.

Angel discussed eight largely unsolved problems:

1. Understanding Visitor Intent

Knowing what a visitor wants to do is no easy task. Understanding visitor intent can shape every decision in digital marketing, from content to personalization. Data must be viewed in the context of user action sequence, time, and pattern. Angel stressed that current statistical analyses fall short of providing this context.

There is a fair amount of analysis on the web about this issue. One strong suggestion from Adobe is to use short, personalized customer surveys; another suggestion is to use remote usability testing.

2. Behavior Elasticity

Analysts may not be focusing on use case elasticity and association analysis to see how easy it is to pull a user into certain behavior. For example, a purchase use case is different than a support use case (such as signing up for paperless billing). He recommended a stronger recognition that not every use case is the same.

Angel elaborates on this point and suggests a testing method to address in in this blog post.

3. Topological Data Analysis

Angel pointed out that analysts don’t use topological analytics to map digital properties. That is to say that site structure creates a choice architecture that provides a significant influence to user behavior. Analytics doesn’t often control for the site structure. They should factor in the “lay of the land” in quantifying user behavior, and take a close look at the surface characteristics of whatever data is in the system.

This approach has garnered some attention in the social media area, particularly in light of the connectivity of social media use.

4. Content Opportunity Cost

Adding, removing or changing content on a site shifts the distribution of attention and clicks. Analytics through assortment optimization (used in retail to determine customer behavior on store shelves) can solve this problem, but is underutilized in the digital realm. Angel stated that the digital realm is not all that different from a store shelf, since web site design and shelf design are both assortment choices.

Youthful companies such as Celect are working to fill this void with custom-built analytics solutions that account for opportunity cost.

5. Multi-Product Merchandising

Multi-product merchandising compounds the prior problem. While merchandising may increase clicks and sales for one product, it may not be positively affecting overall sales. That could be good, or could be bad, Mr. Angel said, but must be viewed in context. Optimizing product aisle and search pages to shift consumers from one product to another is a tremendous opportunity.

A paper by MIT data scientists explores this issue in further detail.

6. Voice of Customer Representation

Voice of customer representativeness is an issue because while online VoC is a critical driver of intent analytics, but analysts cannot easily determine how reliable it is.

Commentators recommend that companies invest time and resource in determining how to analyze customer feedback, so as to ensure representativeness of results.

7. Site Marketing VS Site Content

The interrelationships between site marketing and site content are traditionally ignored by campaign analytics and site analysis. That is to say that content changes both inside and outside of a web site may affect data and should be factored into analysis.

This article explores the important of tailoring site content that matches the user experience at various stages of the purchasing process, even if it means that the content will lack some uniformity.

8. Incremental Attribution

Incremental attribution is an issue because measuring lift or decline across multiple channels is difficult without segmentation and variable analysis.

This last issue was addressed by another speaker at the Symposium, Professor Elea Feit.


Angel suggested that digital analytics tools today can solve these problems, and there is a tremendous opportunity for professionals to do so.