Descriptive Analysis – indicates what has happened for who.

There is a strange perception that more complex analysis, produces more detailed results and that more detailed results will naturally return to larger actions and bigger returns. Bonamy Finch would recommend that Descriptive Analysis is a better place to start as it provides critical context for your more complex analysis, allowing them to be much more interpretable and clear to understand.

Descriptive Analysis techniques are used to describe historical data sets, essentially organising raw information into groups so that any patterns are easily discerned. They do not however, provide insight into causation of these patterns, and so it is commonplace that Descriptive Analysis is combined with additional analysis techniques in order to fully inform decisions.

Examples of Descriptive Analysis include:

• Arithmetic tabulations
• Data aggregation
• Mathematical measures (such as mean, medium, mode and range)

Diagnostic Analysis – indicates why something has happened.

Diagnostic Analysis aims to understand why something happened. It does this by comparing and contrasting data sets and assesses potential patterns, identifies correlations, and aims to determine causality. As manual manipulation of big data becomes less feasible, machine learning and artificially generated algorithms are being used for Diagnostic Analysis.

Examples of Diagnostic Analysis include:

• Multiple Regression Analysis (identifies the best weighted combination of variables to predict an outcome.)
• Conjoint Analysis (identifies an optimal combination of features in a product or service, by weighing them on a scale)
• Probability Measures

Predictive Analysis – indicates what will happen.

Predictive Analysis is the art of taking historical data and using it to predict the probability of events yet to happen. As the volume of data available has rapidly increased, so too has the accuracy of statistical models. Today Predictive Analysis often incorporates machine learning, AI and complex algorithms which allow for the representation of interactions between multiple variables. While this branch of analysis has advanced, it is still reliant on the right strategy being in place. Analysis might tell you the likelihood of a customer to convert, but it won’t give you the strategy on how to make this happen. Predictive tools have been unknowingly misused, which results in misleading results which may not become apparent for weeks, months or even years later.

Examples of Predictive Analysis include:

• Linear Regression (which focuses on the relationship between a dependent variable and an independent variable(s) as a predictor)
• Discrete Choice Analysis (model choices made by people among a finite set of alternatives)

Prescriptive Analysis – indicates what will happen if certain actions are taken.

Prescriptive Analysis could be considered an extension of Predictive Analysis and is the most advanced of all data analysis techniques. Where Predictive Analysis looks at potential outcomes, Prescriptive Analysis extends this by attempting to predict possible outcomes based on particular courses of action. For example the probability of churn decreasing if you offer a royalty scheme vs discounting products.

In order to achieve this level of complexity, Prescriptive Analysis is reliant on technology and while this is rapidly advancing, we are a long way away from models being able to full anticipate the changing nature of human behaviour. It is also key to identify the appropriate type of Prescriptive Analysis for your problem. Examples include:

• Predictive Analysis + Rules (combines predictions with business rules)
• Optimisation (determines the ideal usage of limited time and resources in situations that have different levels of uncertainty)
• Game Theory (a theoretical framework for conceiving social situations among competing players)