Analytics

Diagnosing Session-Level Data for a Streaming Service in Seconds

By Laura Shores - July 1, 2020

Take a moment and imagine yourself as a data analyst at a large online streaming service. It’s a highly competitive industry with intense pressure to compete for viewers’ time and attention,  and one that thrives on insights from streaming analytics.

In your day-to-day role, you’re feeling the pain as your streaming service captures an increasing amount of audience data, but you’re on a never-ending “data request hamster wheel.” Given the amount of data captured for every session, it’s impossible to quickly and comprehensively answer these repeated questions from stakeholders.

From working with our customers we know this isn’t a hypothetical — it’s a very real problem that analysts from many industries face, especially in the data-heavy streaming industries. But what if you could automate the manual, rote work of data exploration, and surface the most interesting, useful, and hard-to-find facts from your complex data? Let’s take a look at how you can use Sisu to find answers in your dynamic session-level data faster and inform decisions on everything from programming to ad lift. 

Diagnose your most complex metrics in session-level data

The average session-level dataset includes hundreds of columns – everything from household demographics and content viewed to session minutes and dayparts. That adds up to billions of possible reasons for change and it’s difficult to know where to start. 

Normally, you’d start with a hypothesis of the factors you think are impacting your metric. At Sisu, we believe it’s best to begin with what you know, which is your key business metric, and let our platform uncover the factors that matter so you don’t miss any. So let’s start with an analysis of session duration. 

Sisu simplifies the process of diagnosing this complex metric. To begin you only need to declare two things: where the session-level data lives and the column that defines the KPI. In this case, that’s an existing table (public.siview_sessions) and session_duration as our KPI, and we’d like to understand why it’s not increasing. 

(Product images use mock data.) 

Typically, you’d have to manually pick through the data to select the factors you think might affect this metric. This is time-consuming and introduces a certain level of bias into the analysis. Instead, Sisu will automatically identify the features in the data that are most important to the KPI, and gives you the choice to deselect factors that prove to be less valuable. 

For this query, we want to understand how the metric performs over time. We’ll select  session_date and all that’s left to do is to “Get the facts.”

Check millions of hypotheses in less than six seconds

Behind the scenes, Sisu is testing hundreds of thousands of individual sessions and millions of possibilities in the data to explain what impacts session duration. This is faster and more comprehensive than any tool (or team of analysts) and eliminates the arduous task of manually slicing and dicing data. As a result, you get more time for valuable analysis and collaboration with the business.

Focus attention on the highest-impact populations 

Now we can see that average session duration shows a moderate 2% decline month-to-month (from about 53 minutes down to 52.3 ). A dashboard will only show a relatively flat curve, which doesn’t give you a lot to go on. With Sisu, we can see trouble lurking below the surface. Let’s take a look at some of the facts below: 


In these facts, you can understand with surgical precision exactly what’s driving the metric – or harming it. For example, right out of the gate we see that viewers in Canada who are watching
Stranger Things demonstrate a significant increase in session duration (going from less than 40 minutes to over an hour). Additionally, this group is driving an overall increase of 1.2 minutes to the KPI. That’s an impact that really matters. Without these Canadian fans of Sci-fi Eggo Waffles, session duration would have plunged another 2 points. 

Fast results like these enable you to quickly and precisely identify interesting subpopulations and to prioritize the attention of your streaming service based on the groups that are impacting your key metric. 

Drill down and proactively compare populations

Finally, we know a single fact like this never stands alone. In addition to understanding what’s happening to this particular subpopulation, we need to examine how this group compares to other similar populations via the population comparison table. This will save you from hours of building custom queries and manual visualizations – we’ve done the work for you.

In the table below, the platform compares how different shows and regions are performing, and how each impacts the overall KPI. In this example, the table validates that there is something special about the Canadians watching Stranger Things, and solidifies that this group represents an aberration and not a greater trend for the streaming service. 

And that’s it! In just a few seconds we uncovered the key factors driving changes in the session_duration metric and we now know which subpopulations are performing better than others.

Need help finding more comprehensive answers in your data, faster? Get in touch for a custom demo or head to our customer page to see how customers like Samsung and Upwork use Sisu to find answers and take action on their data. 


Read more

Fast Data Demands Faster Analytics: Three Tips for Accelerating Analytics

During our recent webinar, "The Future of Analytics: Faster data demands faster analytics," we discussed three tips for accelerating analytics by automating fast data."

Read more

Designing Better Datasets for Diagnostic Analytics:
A Sisu How-to Guide

Read more