Descriptive, Subscriptive, Predictive, Prescriptive
Big data and analytics is very similar to writing a story or a report where the author addresses all of the big W's (and H) of any event. The more information that can be derived from the story, the more value is added. The following article describes the four phases of analytics.
- What happened?
- Why did it happen?
- What will happen?
- How do we control what will happen?
According to Dr. Michael Wu, the chief scientist of Lithium Technologies, more than 80% of analytics are descriptive. This includes most dashboard/personalized analytics software solutions that condense data into more useful nuggets of information. When working with Big Data, it may not even be possible to describe all of the data collected so just a sample of the collected data may be described.
Descriptive analytics may be termed exploratory data analysis by a data scientist or statistician. This is the first and most important step in the problem solving process but it should not be thought of as the final product. It does not solve the problem but rather sets the framework for asking the correct questions.
Descriptive analytics may include calculating averages, variances, percentiles, aggregated tables and a wide variety of creative visualizations. By the end of the descriptive phase of the project, we should be able to describe what occurred.
Subscriptive or diagnostic analytics builds on the questions that were formed during the exploratory data analysis step. Using various statistical methods, we can begin formulating and testing hypothesis surrounding why things occurred.
For example, an analysis of variance (ANOVA) may be performed to determine if certain factors have an affect on the outcome of interest. The effect the factor has on the outcome can be estimated and even simulated based on the changing state of the factor. The insights derived during the subscriptive analytics phase can indicate what information will be useful during the following predictive analytics phase.
Using information and hypotheses developed during the descriptive and subscriptive analytics phases, predictive models may be developed to begin answering questions regarding what will happen in similar scenarios in the future.
This can be as simple as fitting a trend line as seen in the figure to the right or be as complex as developing machine learning algorithms to forecast future data. The immense value behind predictive analytics is the ability to foresee potential challenges or successes in your businesses operations and being able to preemptively act accordingly. This leads us to our next question regarding the ability to guide future outcomes.
Prescriptive analytics is used to recommend various courses of action based on simulated possible outcomes to these recommendations. According to Dr. Wu, prescriptive analytics predicts multiple futures based on decisions made now.
This information can be used to optimize business processes. In any business process, you can control certain factors and then forecast the resulting effect on desired outcomes.
The ability to gain foresight and control over business operations can grant your organization powerful advantages. These methods have applications in sciences, engineering, social media, and all business so when looking for an analytics solution, make sure you are receiving the full analytics package.