Descriptive, predictive and prescriptive analysis

25 jan

What are descriptive, predictive and prescriptive analytics

You often hear about them in the context of data analysis, but what are their exact definitions and do they have an important role to play? Well yes, it matters! And the level of complexity increases exponentially as you move from descriptive to predictive to prescriptive.

1– Descriptive analysis to identify the situation

This type of analysis is primarily based on reviewing historical data and calculating key performance indicators (better known by the abbreviation "KPIs" for Key Performance Indicators). It allows you to see what happened during a given period and also to describe the current situation: much like an accountant would!

You are doing descriptive analysis when you:

  • Basically observe what happened for a selected period
  • Identify details that are the source of significant results
  • Paint a picture of the most significant past and present elements

2– Analysing with a crystal ball

It's the combination of the art and science of building predictive models where you use historical data to predict what will happen. For example:

  • Sales, based on historical growth and seasonality
  • New customers, based on your upcoming marketing campaigns
  • The number of transactions, based on the products offered and their price
  • Etc.

This analysis can also be used to plan and measure the impact of introducing a new product or opening a physical store.

3– Prescriptive analysis

This is the last of the three analyzes. It is very similar to predictive analytics because it has the same basic elements. Prescriptive analysis is a way of simulating results based on changing certain inputs in your model, that is, running multiple scenarios to find which conditions or operational parameters are best.

You need to have substantial data on hand to assess the many possible outcomes so that you can measure or note what might happen by doing something different. The result of this should be your best case scenario based on the criteria you provided.

If we take the example of the number of transactions, we would run the model several times with different price grids for the products and we would choose the scenario with the optimal price grid in terms of the number of transactions and profitability.

Hoping that this summary brings more clarity!