Business analytics is a set of methods (and tools that implement those methods) that use data to drive the analysis of systems and organizations, can be divided into four types [1]:
- Descriptive analytics: the goal is to picture the processes and statistics that drive the business in a meaningful fashion. Process maps and dashboards are excellent examples of descriptive tools.
- Predictive analytics: the idea is to use historical data and machine learning (ML) methods to predict the behavior of key performance indicators (KPIs). The main goal is to be able to detect problematic parts in the process early and eliminate waste before it occurs. More advanced predictive methods and tools not only provide a prediction but also give the user an overview of the main root causes that drive inefficiencies and problems. This ability became known as explainable AI (XAI) and is now a desired (and almost mandatory) feature of modern predictive tools.
- Comparative Analytics: this is a new type of analytics proposed recently in Baron (2020) [1]. The idea is to compare the current system to various alternative systems. For example, if we take a procurement process, we may wish to compare our current process to an idealized version of the same process. This is known in other domains as conformance or concordance checking. Comparative analytics may help us discover deviations from the ideal process. Moreover, they allow the users to uncover the impact of changes and deviations on key performance indicators. Lastly, they are a steppingstone to the holy grail of business analytics, namely, prescriptive analytics, because when comparing the systems, one ties together system inputs (such as the number of purchases made, and vendors used for purchasing materials) and system outputs (such as delayed shipments and late payments).
- Prescriptive analytics: the most important type of business analytics, in my view. These methods and tools produce recommendations, optimized tasks, and changes that would improve the underlying processes in a data-driven and model-based fashion. They must be data-driven to provide evidence-based analyses, and they must be based on a model to enable what-if? calculations and optimization.  For example, if other types of analytics combined would discover and quantify the fact that Vendor1 is causing trouble in supplying a product in a procurement process or be able to predict a late payment in an order fulfillment process, then prescriptive analytics would actually tell us that we must change Vendor1 to Vendor2 and recommend changing payment terms to prevent the order from being paid late. These recommendations are the result of an optimization process, which relies both on historical data and on a model of the underlying system
Since I claimed that the fourth type of analytics is the holy grail, and the main target that we should try to achieve, one may ask: can I simply apply prescriptive methods without bothering with the other types? The answer to this question, in my opinion, is in the negative. One must go through a constant cycle of Descriptive Predictive Comparative Prescriptive without ever stopping on one particular type. Without discovering the process and showing the main KPIs on dashboards, we cannot really know where the problems are and where our predictive focus should be. Without predicting the behavior of the system (and uncovering the root– causes that drive the KPIs in an analytical ML-driven fashion), one cannot compare systems under various changes. Lastly, without comparative analytics, one cannot quantify the relationships between changes to the inputs and the outputs a crucial connection that drives prescriptive analytics. In other words, the four types of analytics, are essentially four steps in a constant improvement cycle that should never end.
However, most process mining and business analytics tools available today mostly focus on several (but not all) of the components from the four I mentioned above. There are great dashboard tools that spend their entire effort on descriptive analytics. Other tools focus on prediction, without trying to understand the impact of various changes and interventions on the KPIs.
I believe that only a well-rounded and wholesome tool that provides access to all four types of analytics can be robust, sustainable, and useful to businesses in the long term.
[1]Baron, Opher, Business Analytics in Service Operations ”Lessons from Healthcare Operations (August 14, 2020). Naval Research Logistics, Forthcoming, Available at SSRN:¯https://ssrn.com/abstract=3673903¯or¯http://dx.doi.org/10.2139/ssrn.3673903
Arik Senderovich, PhD