So we load up our shiny new visual analytics platform and start slicing and dicing data, creating charts and graphs using the latest designer color schemes, and we might even add a dash of animation. And because we now have so many sources of data, the combinations of dimensions and measures is pretty well endless. As a result we start to see correlations and trends where none had been visible before. These ‘insights’ come thick and fast. Only there are two issues that spoil the party. Are those ‘insights’ really insights, and what do we do with them?
To answer the first question. It has long been documented that human beings invariably add their biases when analyzing data. If we want to find a trend, sure enough we will find one. Whether it is really a trend depends on a deep understanding of the business, and whether that rising line is in any way justified. If not, then it is probably nothing more than noise dressed up to look pretty. So we have to be very careful when digging around in data. We should understand that correlation is not causation, and that many of the ‘insights’ we gain might be nothing more than randomness made to look respectable.
The second question – what do we do with the insights – is complicated. Let’s examine what might need to happen to make an insight actionable:
- Business managers must be convinced that the insight is real and can be trusted. It becomes necessary to create a story, presenting all the evidence, with explanations of why the various charts and graphs are the way they are. This might seem like it should be a straightforward process, but let’s remember that change usually involves treading on someone’s toes. So there will be resistance from those who will need to implement change, and detailed scrutiny of what is being presented, with ruthless exploitation of any holes in the evidence being presented. Think carefully before you poke a stick in the hornet’s nest.
- If there is broad agreement that the analysis is correct, the next step is to decide how business processes need to change. The change might be trivial, or it might go to the heart of several business processes. If it is the latter then it may take months to redesign processes and get agreement from all involved.
- Once implemented, the changes need to be monitored, and so new reporting and analysis procedures need to be put in place. Fine tuning might need to be made, or heavens forbid, we might find that matters have not improved in the way we expected. New analysis, with new ‘insights’ might suggest additional changes. It might be best to keep quiet about that one.
It should be clear from this that analysis is really just a small part of the insight-to-action journey. While suppliers of analytics tools flatter business users that their insights are important, the reality is somewhat different. Most insights will not be insights at all, but random noise making itself look respectable. Those insights that have been interrogated and found to be true mean nothing unless business processes can be changed, and this is where the real hard work is to be found.
Before we obsess over charts and graphs we should ensure that the results of analysis can be implemented in a controlled, documented, measurable manner. To this end standards such as Decision Model and Notation are being introduced, as a way to integrate decision logic into business processes. After all, making the results of analysis actionable generally means that people make different operational decisions.
Right now we are fascinated with the shiny new things – just because they are new and visually attractive. The real work consists of having people who know when randomness is trying to pull its tricks, so that analysis actually means something, and having the processes in place to transform significant insights into action. Anything else is just a waste of time.
Within half a decade we will have AI powered BI, and so the task of finding meaningful insights in our data will become easier. We will also have the methods and infrastructure to quickly move from meaningful analysis to modified business processes. Until then we need to be cautious and expect that actionable analytics will become more of a reality as the whole process becomes automated.