Machine Learning Alleviates 'Blank Canvas' Syndrome

For an artist or a writer, a completely blank canvas might inspire the next Rembrandt masterpiece or Pulitzer Prize winner. For a business manager performing data analytics, a blank canvas might cause extreme terror.

Every data visualization tool on the market today looks and operates in a similar fashion.  They have a pick-box down the left of dimensional attributes and numeric metrics and a canvas onto which you would drag and drop those attributes and metrics thus constructing your visualizations. 

See examples from various vendors below, including Oracle.

However, there is an issue with all of these tools that is being referred to as a "blank canvas" syndrome. In this blog, we will discuss what it is and why is it something you would consider alleviating.

The Blank Canvas Syndrome

What is the issue with this blank canvas?  Seems like a natural and clean position to begin any analytics.  Let's take two different but equally capable analysts using the same data visualization tool.  Each analyst has their unique background, life experience, and professional experience.  If each analyst was tasked with finding the answer to the same business question, using the same tool and same data, would each analyst actually execute exactly the same steps and reach the exact same conclusion?  No, that's unlikely. 

How the analyst tackles the problem comes down to their experience, bias, and even their imagination.  When faced with a blank canvas, the combination of attributes and metrics selected and visualizations created becomes very different.  The end results are usually quite distinctly different.  When analysing an HR data set to understand attrition rates, our first analyst might think that salary is a key factor and create an answer around that while our second analyst thinks the issue is connected to with sub-standard offices that are not comfortable for employees.  Another issue that arises is that our analysts may become fixated on a spike or trend they spotted and consequently miss other critical factors that contribute to the overall problem.  Our analysts essentially can't see the forest for the trees.

Are our two analysts bad at their jobs?  No, it's human nature to fall back on your experience when faced with new challenges.  However, differing answers to the same question is a significant business problem.

Today, the trend is to become more productive and more efficient. Cutting costs by optimizing business processes is top of mind.  This rather slow, iterative, and wasteful approach for analysts to start their analyses to then render conclusions that might answer only part of the problem.     

Machine Learning That Explains Attributes in Context

Modern analytic platforms have machine learning embedded to alleviate the issues I described earlier.  Machine Learning removes human bias and imagination, considers vastly more information, is incapable of missing key signals and does everything in a fraction of the time.  Sure, the machine might not provide a 100 percent complete picture, and some human interpretation may be required, however, if our objective is to become more efficient with better accuracy, then machine learning saves tons of time—and time is money. 

The key is that machine learning is placed into the hands of the business professionals, the end users.  In our case, this means our two HR analysts.  There is no requirement to present the problem to a data scientist (and their team) who use extremely complex and expensive tools to power machine learning to generate results days later. 

What if you could instead, ask the visualization tool to explain an attribute in context of the other attributes and metrics in the dataset? 

Instead of starting with my blank canvas I'm immediately presented with two key drivers of attrition, Over Time and Job Role.  It turns out that salary and premises were both only minor factors – both our analysts missed the key business problems.

I'm also presented automatically generated insights like below.  With 93 percent confidence, attrition was high due with people that had a Job Level less than 2, have been with the company for less than four years and were putting in overtime.  Immediatley, I can see a tangible problem that I can address – without even starting to drag and drop stuff onto my blank canvas.

Finally I'm also presented with anomalies that are usually missed or too time consuming to figure out via manual analysis.  Here I can see that the machine discovered that our Research Scientists who are putting in overtime are leaving at a much higher rate than normal.

Alleviate Blank Canvas Syndrome

Today, blank canvas syndrome is accepted as normal.  However, starting analyses this way is inefficient, introduces human bias.  It also may lead to conflicting answers, or entirely miss the root cause of the problem.  Machine Learning is now in the hands of business users, empowering them to filter noise, quickly see the signals and share their findings. 

Have you started using machine learning with your analytics, or are you still just coping with how it's always been done?

Take a look at a complete video demonstration showing how to alleviate blank canvas syndrome using machine learning to explain attrition in an HR dataset.