Self-Service Analytics and the Emperor’s New Clothes


There is a growing realization that asking business users to do their own analytics may not be a particularly productive thing to do. On the one hand we are seduced by attractive graphs and charts that use designer color schemes, but on the other hand people do have other things to do, and creating analysis that can be trusted is anything but trivial. Some of the vendors in this space are now actively saying that self-service analytics isn’t working – Logi Analytics for example.

So let’s have a look at this particular Emperor’s New Clothes – item by item:

  • Ease-of-use. Well yes, the new generation of visual analytics platforms do simple things with great ease, but alas meaningful analysis is not simple. In fact you will already know the simple things, and getting meaningful signal from data requires considerable skills, effort and time. The term ‘ease-of-use’ can seem like a bad joke when trying to analyze data from half a dozen data sources in an attempt to nail some anomalies. Meaningful analytics is not easy, and business users may not have the time or inclination to prepare, transform, clean, explore, discover and analyze data.
  • Trust. Human beings are pattern seeking creatures, and we will even find patterns in data that is known to be random. If a business is going to act on analysis it had better be sure that it is correct. But this propensity to find patterns where none exist in reality, throws doubt on simple, uninformed analysis. Sure, a trend may look impressive on a graph, but apparent trends, that are nothing but random noise, appear all the time. For example, in a business that has largely flat revenues, a six month unbroken upward trend in revenues will appear approximately every five years – and it will mean absolutely nothing. But such is the visual impact of such a thing, that management will be very tempted to invest in more capacity, only to see things revert to the mean over the coming months. If we are to trust analysis, then we need people with the skills necessary to separate the signal from the noise – and it isn’t a trivial task.
  • Productivity. Most business users spend a good deal of their time interacting with core business applications – ERP, CRM, supply chain – and so on. Having to switch out of these applications and load up a visual analysis platform is a fairly unproductive thing to do. Much better is the ability to embed visual analytics into these mainstream applications. This eliminates context switching, and provides a means of delivering pre-built analysis right where it is needed.
  • Integration. Visual analytics is just one of several forms of analysis used in businesses. While machine learning, AI, prescriptive analytics, and several others may seem like futures, they are not. Over the coming five years we will see a drift away from naive visual analysis, with more aggressive use of machine learning particularly, and the production-lining of the analytical process. In fact the real benefits of analysis will only become apparent once we have embedded the analysis into our apps, using it to automate some processes and make others much more efficient and effective. To this end we will need to embrace more sophisticated forms of analysis and integrate them into the production environment.
  • Action. If we poke around inside our data we will find many seemingly significant insights. As mentioned above, many of these ‘insights’ will just be accidents in the data, with no significance at all. The remaining insights immediately pose a problem however. How can we move from insight to action? This is not a trivial thing, since change often affects business processes. So in reality we need a full analytics lifecycle where insights are the first step, and changes to business processes take place in an orderly manner. Implementing changes may mean building new analytic models, new business processes, new reporting mechanisms – and so on. Some users of visual analysis platforms become so overwhelmed with their ‘insights’ that they suffer analysis paralysis – resulting in no action at all.

There is a growing awareness that the visual analytics feeding frenzy is probably past its peak, and so we can at last start to talk sensibly about what is possible and what is not. Suppliers such as GoodData, who also have a visual analytics platform, are expanding their reach to provide both infrastructure and integration with other forms of analytics to deliver an actionable analytics platform. Expect others to follow.

Perhaps the best resource users of visual analytics platforms can call upon is the book Signal by Stephen Few. It provides a no-nonsense, rigorous approach to extracting signal from noise using visual platforms. However, more generally, visual analytics is useful for data exploration and discovery, but other forms of analytics (machine learning for example) provide a much more robust mechanism for putting analytics into a production environment. Finally, it will come to pass that your visual analytics platform will eventually be AI driven. A taster of this appeared in BeyondCore, a visual platform that automatically extracted meaningful patterns from data. It has been acquired by Salesforce, and now does the same thing with Salesforce data. So AI driven BI is where we are headed, and will be commonplace within half a decade. Even so, we then need to ask who, or what, will be consuming the analysis – will it be man or machine. My bet is the latter of these two.