Prescriptive analytics is a data- and model-based process of understanding what is occurring, then making well-informed decisions with the insights we glean. As a methodology, prescriptive analytics commonly leverage tools such as machine learning or artificial intelligence to understand the systems impacting outcomes, then graph analysis to interpret and communicate the results. By using these data-driven methods, it’s possible to understand data sets that are too large for humans to analyze manually, and to make careful decisions based on an understanding of the processes rather than relying on instinct or habit.
How to Do Prescriptive Analysis
- Pre-process the data: Pre-processing the data often involves removing outliers, reformatting the data and addressing gaps in the data.
- Use the data to drive the model: Using the data to drive the model often means training and testing a model from a tool such as Scikit-learn, then using the model to predict the results.
- Interpret the results: Interpreting the results often leverages techniques such as graph analysis to understand what the model results say and then conveying the story to others.
How Does Prescriptive Analytics Work?
Prescriptive analytics is a means of using technology and large data sets to make better decisions. It is commonly leveraged by businesses to understand their current operating environment in order to make strategic decisions. Prescriptive analytics supports these goals by examining large data sets to understand what is happening, build a model to explain what is happening and suggest the best path forward given the current understanding of the data.
Prescriptive analytics typically leverages machine learning and artificial intelligence techniques to understand the data set. These tools are capable of identifying patterns in large data sets, then extrapolating patterns to different conditions in order to evaluate the impact of different decisions. We can constantly update the models by retraining them on new data sets to continuously improve the models’ understanding of the problem and provide better recommendations to stakeholders.
The data inputs to the model are determined by the question the user asks the model to answer. Common examples of data inputs include information about possible scenarios, past performance, current performance, environmental factors believed to impact performance and available resources.
What Is Prescriptive Analytics Used For?
Prescriptive analytics is fundamentally used to answer a question then make recommendations or decisions using the new understanding of the situation. This approach can answer any question so long you have adequate data. Examples of typical questions include:
- “What does this mean?”
- “What should we do?”
- “How can we make X happen?”
Prescriptive analytics is a useful tool for evaluating both the probability and impacts of different outcomes. In this way, prescriptive analytics helps an organization prepare for possible outcomes, particularly the worst-case scenario.
The main strength of prescriptive analytics is that it uses computer models to analyze larger amounts of data than the human brain can handle. As a result, the model can identify trends that humans would miss and helps us develop a more nuanced understanding of the data. In this way, prescriptive analytics help us make data-informed decisions, rather than jumping to ill-informed conclusions based on prior experience, hunches or gut instinct.
Examples of Prescriptive Analytics
Some typical uses of prescriptive analytics include:
- Finance: Identifying and preventing fraud
- Insurance: Limiting risk by evaluating the likelihood and impacts of negative events
- Manufacturing: Increasing efficiency by identifying and resolving causes of inefficiency
- Marketing: Increasing customer loyalty by identifying and resolving pain points
Some specific examples include:
- The California Independent System Operator (CAISO) predicts the electric supply and demand to identify times when the grid may not be able to meet demand and incentivize people to conserve energy.
- Google analyzes internet usage patterns to identify advertisements that are likely to appeal to users.
- Property management companies can model predictive control for products, such as water heaters. These models predict the products’ energy consumption given the upcoming weather patterns and typical occupant behaviors, then adjusts the control strategy for the building to minimize the operating cost while satisfying user needs.
Disadvantages of Prescriptive Analytics
Prescriptive analytics is not a panacea that can magically produce useful information without a strong understanding of the situation. Since the models are capable of answering specific questions, the users must know what questions to ask and how to interpret and respond to the results. Simply passing a large quantity of data to the model will not yield a valuable result.
Similarly, the models utilized in prescriptive analytics are data-based and thus subject to the GIGO (Garbage In, Garbage Out) concept. A model trained on garbage data will yield garbage results so thorough data pre-processing is key.
Descriptive vs. Diagnostic vs. Predictive vs. Prescriptive Analytics
As a methodology, prescriptive analytics looks at what happened in the past and helps prescribe a path forward based on that data. Descriptive, diagnostic and predictive analytics all work a bit differently.
Descriptive analytics analyzes historical data to better understand changes that have occurred in a business or process. Specifically, it addresses the question “What happened?”
For instance, we can use data such as price, revenue, number of customers and user behavior to understand why revenue may have increased or decreased recently. Addressing these questions helps us understand an organization’s strengths and weaknesses — what is or isn’t working. If an organization understands these data points, they can better leverage their strengths and address their weaknesses to drive improved results.
Diagnostic analytics attempts to address the question “Why did this happen?” Using diagnostic analytics, we can connect causes to effects by looking for data-based connections. In order to examine the relationship between causes and effects the user must supply very large data sets describing each possible cause. For example, if Disney wants to understand what caused a decline in Disneyland ticket sales in one year they may need data showing historical weather, economic conditions, global health, Disney popularity, ticket sales for newly released movies and anything else that may have caused the downturn in attendance.
If applied effectively, diagnostic analytics can provide great insight into the best ways to run an organization or process. The outputs from diagnostic models provide relationships between choices made by the organization and results, thereby informing the user of what does and does not work well.
Predictive analytics attempts to answer the question “What will happen next?” This process uses historical data to create an understanding of the existing trends and impacts, then predict what will happen in the future. The understanding of how trends impact results enables us to evaluate the likely effects that different decisions will yield.
As with all attempts to predict the future, predictive analytics becomes less accurate as the prediction horizon increases. As a result, the best practice is to use predictive analytics for short-term projections as the uncertainty will be too high with longer timelines.