Predictive and Prescriptive Analytics in Practice

As part of the data and analytics practice, leaders use both predictive and prescriptive analytics techniques individually to find an answer to questions as part of any decision-making process. The challenge many business decisions are multi-dimensional and have different aspects.  Therefore, using an individual technique is only part of an answer, and combining them requires different techniques, skillsets, and tooling. These traits of business decisions make it hard (expensive, time-consuming, and lack of skills) for organizations to automate business decisions that derive business operations with a big impact on business objectives.

Predictive Analytics

Predictive analytics enables organizations to look into past data and anticipate the future with confidence as part of the data and analytics practice. One of the techniques in predictive analytics is Machine Learning (ML). Machine learning is an umbrella of multiple techniques and algorithms that enables organizations to classify, predict and forecast based on historical data. Also, there are techniques as part of ML that enable organizations to better understand their data, such as clustering algorithms. This enables organizations to find abnormalities, similarities, and outliers in a data set.

Prescriptive Analytics

As part of the data and analytics practice, prescriptive analytics help organizations address the questions of “what should be done?” or “what can we do to make a given outcome happen?”.

Prescriptive analytics also is an umbrella of multiple techniques that allows organizations to create recommendations for a specific outcome. Compute and calculate values based on many different parameters for different circumstances. Create a decision outcome based on different situations. Rules and optimizations are part of prescriptive analytics which each has multiple techniques of modeling. For instance, leaders can model business rules using decision table, natural language, and tree/subtree depending on their personal preferences and their technical skills.

Combining Predictive and Prescriptive Analytics to Solve a Business Problem

Many business decisions in organizations are multifaceted. This means a single technique of either predictive or prescriptive analytics will not address the business challenge to automate the business decision. A multi-dimensional business decision requires a combination of both techniques.

They require to consider multiple dimensions and multiple aspects of the decisions. In this scenario, Decision Graph, e.g., the decisions requirement diagram (DRD), will help in shaping the business decisions from multiple aspects and viewpoints.

Once the unified model for business decisions is created using a decision graph, each decision is implemented using different techniques, i.e. rules, ML, process, etc.

As you can see in the above decision graph model, the base price for the premium calculation uses an ML-trained model. And, then there are multiple other decisions in the graph based on rules using decision tables for discounts, COVID vaccine, etc.

The benefit of this combined approach is that in most cases, it is not possible to find enough quality data to retrain an ML model; specifically, if the situation is recent, then there are not enough data points that existed anyways. Therefore, a better solution is to create a base model for predictive techniques to create a prediction, forecast, etc., and then apply more specific situations using prescriptive models based on business rules, computational logic, etc., to reflect the situation and specifics of the scenarios.

The Big Picture

In an enterprise, when we look at complex business decisions holistically, this combined predictive and prescriptive analytics approach enables them to deliver values quickly and frequently. The reason is by enabling organizations to automate the whole business decision rather than one piece of it in a reusable, iterative manner. And add the next best action (prescriptive analytics, i.e., decision engine) to the solution to identify what should be done for the next step and create the whole orchestration to carry out the actions.

In this approach, we will fulfill all 4 stages of the decision cycle. The benefit of that is to create an autonomous, self-sufficient unit of a decision that can be reused across multiple business processes, applications, systems, and so on, and it will drive the business operations. Opposite to when the result of a predictive model or a BI is fairly isolated, not integrated, and does not change business operations and experts’ behaviors!

Last updated May 5th, 2022 at 12:09 pm, Published April 21st, 2022 at 12:09 pm

Arash Aghlara