Ensuring Your Visualizations Answer Questions Instead of Creating Confusion
With data visualization tools like Looker and Tableau, creating compelling stories through data is easier than ever. But how do you ensure that your data is actually answering questions rather than creating confusion and unproductive questions?
A Picture Paints a Thousand Data Points
The key is to stay focused: the beginning, middle, and end of your visualization process should center around the main takeaway of your analysis. In our examples below we will be using Looker.
First, the insight itself will help choose a style of visualization, e.g. timeline, map, correlation, etc.
For example, I’m trying to understand how a feature change on my e-commerce site in December 2017 affected the behavior of my visitors. Let’s say this feature change adds a required sign-in before a user can see their cart. The key to this insight is the time-based feature change, so I want to compare visitor behavior before and after the change. My visualization will be time-based, with time on the X-axis and my behavior metrics on the Y-axis.
While time-based visualizations are common, there are tons of interesting, compelling viz types out there. For a detailed look at different types of visualizations, check out Looker’s visualization guide.
Do Some Housekeeping
Now I want to reduce the noise, and ensure that my viz highlights the most important information. Crucially, you should keep in mind who your main stakeholders are. For instance, if my audience wants to understand how this feature change affects sales numbers, I might highlight how revenue changes over time.
However, in this case, my main stakeholder is a Product Owner whose main goal is to get visitors to the next step in the conversion funnel. In order to highlight this KPI, I’ll show the % of users converting to the next step.
Be Concise, Be Clear and Reveal the Whole Picture
Next, I want to ensure that my visualization is not showing confusing incomplete periods or false trends.
It appears that the % of visitors getting to the next step decreases dramatically in the last month. This looks concerning, but since I’m completing this analysis in early January 2019, the latest month is showing only a small amount of data, skewing the results. Looker has a great filter feature to help users avoid this common problem. See in the filter pane how I’m able to set the time period to 12 complete months, meaning that my visualization will not include the current incomplete month.
Now that my visualization isn’t showing an incomplete period, I can look at the trend line more accurately. Again, most visualization tools have great trend lines features, and Looker is no exception.
What Story Are We Trying To Tell
So what’s the story here? Where’s our takeaway? Remember, we’re trying to compare the behavior before and after the feature change in December 2017. Let’s make this more clear, and assess the trends before and after the change.
No we can see that before the feature change, visits to the next step were increasing, whereas after the feature change, visits to the next step were decreasing. Let’s dig deeper, and make sure that this trend is consistent over different customer types.
If I split out my visitors into two different segments: existing customers and new customers. I see that existing customers tend to continue on to the next step, whereas new customers will drop off of the page. This is a valuable insight to bring back to the Product Owner.
Finally, I want to make sure that my visualization answers questions, rather than creating more questions. If I don’t annotate this graph, my stakeholders won’t understand the main insight: that requiring sign-in is a major pain point for new users.
Now the insight is clear, concise, and actionable!
Conclusion
The type of data we are working with should dictate the appropriate visualization to use. Clarity is key. Time series data works extremely well with line graphs.
Know your audience. Different business users and stakeholders will want very different things from a visualization.
Making sure not to be misleading is paramount. Far too many visualizations contain confusing or misleading data, and often incorrect results. Always validate against your source data and ensure you don’t give your audience a misleading story.
Make your insights clear, refined and concise. Annotation boxes are a great way to give short sharp detail on insights discovered.
Ensure that your visualization is answering questions, not creating more confusion! This will increase the desire for other valuable questions to be answered.
It’s important to know and leverage the power of your BI tool of choice, stay tuned for more information on how The Analytics Academy is leveraging the power of Looker in articles to come.