I’m going to be blunt – most customers (ie. the execs/business partners) don’t know what flavor of analytics they want. It’s not uncommon for an analyst to be told someone has use case for predictive analytics, when really what they’re asking for is a simple dashboard – which is closer to diagnostic analysis. For example, a simple indicator that flag when owners projects are nearing budget constraints…is this Predictive or Diagnostic analytics? If we are going by the definition, you’d probably argue it is “predictive”, because it will be estimating that something is going to happen in the future. However, the mere act of alerting is only answering the question, “Where do we focus attention amongst the numerous projects?”. Some would even over exaggerate to say that it’s AI if we implement an automated email alert feature with no human interaction. Which brings me to another common misconception, the buzz word ‘AI’ means Artificial Intelligence, but in reality, most people are thinking of Automated Intelligence. The automation piece can easily be handled by some simple programming script and isn’t an actual method of AI. Once the data analyst makes these clarifications, the customer realizes there is a lot more work to be done to identify valid applications for machine learning and predictive analytics use cases. More often than not, it requires many many examples for the model to learn from. Maybe even extensive interviews of what truly causes a project to go over budget… but that’s another story.
It is possible to include all the analytics mentioned above by restructuring the statement a little. Such as, “by monitoring supplier delivery (from internal and external sources), staffing capacity, and type of project, we can better predict a forecast to automate alerts for projects threatening to go over budget.” That statement is a much more formal customer request that shows the need for predictive analytics due to previous issues in supplier delivery and staffing capacity having an adverse effect on budget constraints for certain types of projects. The misconception I’m trying to convey with the previous example is that advanced analytics like; Predictive, AI, and even Forecasting, requires that the customer knows what questions they are trying to answer. Otherwise, the data analyst will have to shape the demand. To better show the relationship of the type of analytics and issues they answer, I’ve illustrated my attempt of an analytics maturity map that may prove useful to other data analysts or their customers. I’ve also included some common types of algorithms/methods used to generate the insights at those stages.
One major difference between my maturity map and others you can find on the internet is that mine is circular. I find that other resources online tend to portray that the different types of analytics is a linear progression… and separate entities. In my opinion, the various kinds of analytics are subsets of one another. It shows that if you aim to have prescriptive analytics, you can also be predicting, forecasting, and using descriptive analytics all at the same time, something not possible in a mutually exclusive diagram. Also, the algorithms/methods used in each group can work for its subcomponents. One might argue that – by definition of a ‘subset’- all diagnostic analytics are also predictive, I think that’s true because the potential is always there for each use case, whether you’ve enabled that capability yet is up to you.