Yes, I said it! So many of you are going gaga by visualizations; so blinded you forget the real reason for the work: Business Value. Some of you might even participate in contests of who can make the prettiest visualization. Some popular Data Science blogs even have a visual of the month section. Well, after all, they say a picture is worth a thousand words. That may be true of landscapes or love interests, but the visualizations data scientists produce for business people often aren’t worth the battery power it takes to display them on an iPad.
That’s not to say that a well-executed graph or Sankey diagram isn’t an amazing thing. If you are among the 1 percent of people who know how to use that chart, it may well convey a lot of information to you.
So how do we score? Business analytics have value only when they drive action.
Business people can’t act on visualizations alone to reduce churn, boost sales or optimize production. Visualizations don’t go far enough. A scatter plot that distinguishes high value and low value customers is certainly meaningful to data scientists, but often leaves sales and marketing executives asking the rather obvious question, “But what do I do with it?”
The real value therefore lies one step further along in the process: in recommending actions for real scenarios employees face. Visualizations, and this unfortunately applies to the majority of them that I encounter, are just pretty pictures without any guide to interpretation, explanation of narrative, or suggestions about next steps. As a data scientist it should be my responsibility to provide the story and then leverage my knowledge of the business along with that of domain experts to attach actionable options to the visualization.
Data science needs to be a bit less enamored of itself. I think probably 80 percent of data science energy is expended in choosing just the right analytic or machine learning technique for a given project. Those things are important, but they only yield about 20 percent of the final story: the visualization. If we really want analytics to make a daily difference in the business by integrating its output with operations, we need to put more effort into telling the rest of the story in ways that are meaningful to the people who will execute based on the findings.
This is not to lambaste the hard work of data scientists. We simply need to reconsider our thinking about visualization and business results. Data science needs to understand its target audience and deliver visualizations that aid interpretation. We need to make our analytics actionable; not visual.
For example, a data scientist can look at a scatter plot and derive meaning from it. Further down the line, a citizen data scientist—folks from across the organization who have some of the skills needed to prod at the data and find answers for themselves—might hold a somewhat less powerful magnifying glass to the results. That requires an adjustment or at least some annotation. But the end game is people and processes that need answers, direction, or a script to deal with the very real business of the business.
A data scientist can take data from the call center, information from POS systems, and clicks from the website and bring them all together to provide a churn prediction and customer lifetime value score. The next step is to create a pretty diagram that shows high and low risk. But the business can’t use that. It’s a report, not a game plan, and the distinction is critical. The visualization that goes that extra mile and tells the call center person the right thing to offer relative to the value of the customer under consideration is one that functions on a higher plane than what is typically offered today.
In the end, successful visualization boils down to four things:
- An understanding of the business problem to be solved
- An understanding of the people and processes that will be affected by the solution
- Data, analytics, and results
- Working with the business to turn analytics insights into something people and systems can act on.
There is a tendency to begin and end analytics undertakings with number 3. That’s understandable. Steps 1, 2, and 4 are hard and outside the comfort zone of many in data science. Nonetheless, that is where the real worth in analytics lies. The truth is the best visualizations isn’t a picture at all, it is a next best action script found in a client management system at the call center. Embedding analytic processes into working systems is how we score!
