A Common Field Is Not Enough – Additional Considerations in Analytics
A recent white paper published by Tableau, entitled 6 Best Practices for Creating Effective Dashboards (freely available on their website), included a useful interview with Unilever’s Director of Analytics. Rishi Kumar discussed the rich variety of data available in the manufacturing sector, from production and sales figures to behavioral data such as the content of a consumer’s basket of goods.
What is not treated in the paper, however, is how these best practices in manufacturing translate to analytics on the service side of the economy. In the case of IneoQuest, we are in the business of selling accountability – to our customers, our content providers, and marketing clients – and it is worth pausing to note how this informs our approach to analytics.
Dr. Kumar is right to highlight how joining various datasets on a common attribute is a useful jumping-off point. What the article omits, however, is just how critical the choice of a meaningful starting place is. For example, using the same dataset, a marketing executive may have different requirements than a network operations chief, and both may have distinct needs from a manager in customer service. Same data, but different views.
This is all inspired by an issue encountered by HBO during the season premiere of Game of Thrones. HBO GO, the firm’s streaming service, experienced outages, excessive buffering times, and access issues. Customer frustration predictably lit up social media, but worse than the problem itself was the fact that an elementary application of analytics would have mitigated the outrage. Using a relational database, IneoQuest products correlate Program Name and User ID, and along with industry metrics like Session Establishment Time and Bitrate help network operators and broadcasters quickly identify affected users and take the appropriate action (be it via an apology email, a statement credit, or similar remedy).
In some respects, the case in video experience monitoring is similar to the use case at Unilever, and naturally the instinct is to join on product, just as Unilever does. While a manufacturer may ask how different varieties of soap are selling, a content provider might wish to analyze by a specific video asset. But this begs another crucial question: are we conducting our analysis at the correct granularity? Do our customers need us to go up one level and evaluate an entire channel (HBO or ESPN)? Are they asking us to provide analytics at a lower level and report on a given bitrate variant (1080p, 1080i, etc.)? There is value waiting to be unlocked at each dimension.
More questions will come in future posts, including a treatment of the importance of meaningful metrics and the crucial nature of processing and presenting relevant data in ways which extract actionable insights. The long and short of it is this: the most brilliant programming mind can attack the most robust data set, but without the requisite foresight and understanding of a user’s needs, the customer (and their clients) will never be better off than when we started with millions of rows of raw data.
