One of the most important capabilities to develop in any large business is being able to take a systematic approach to analytics. Making the right decisions is a challenge for businessmen, more so if there is limited data to support the decision-making process. What’s more, past experiences aren’t always reliable indicators of future outcomes, so decisions based only on historical data may sometimes be incorrect. Getting started with prescriptive analytics.
But there are techniques, technology, and tools to help enterprises in this. Predictive analytics is now a popular way to predict future outcomes by aggregating and analyzing historical data. Often, predictive analytics is referred to as the “proactive component” of data analytics.
But there’s one more technique that’s being adopted by businesses as part of their decision-making process, and that is prescriptive analytics.
As we had written before in a previous post, Gartner’s Analytic Ascendancy Model explains it well:
- Descriptive analytics: What happened?
- Diagnostic analytics: Why did it happen?
- Predictive Analytics: What will happen?
- Prescriptive Analytics: How can we make it happen?
Prescriptive analytics provides specific options that can be implemented, which are then compared to criteria to determine which are the most appropriate to be adopted. It can generate recommendations or insights that will help business leaders make better, faster decisions to optimize the value of their data.
For example, if a business wants to improve its service, prescriptive analytics can assist in determining what changes can be made to the customer experience (including price and product selection).
The Express Analytics CDP Oyster helps companies integrate data and get deeper insights with its mature ID resolution capability.
What is Prescriptive Analytics?
It is an area of business analytics (BA) that is devoted to determining the best course of action to be taken, given a specific set of circumstances or opportunities. Prescriptive analytics can also provide options for how to maximize a future opportunity or minimize a future threat, as well as explain the implications of each alternative.
Prescriptive Analytics Definition
Prescriptive analytics is the branch of analytics that seeks to provide “what should be done, or what can be done” in light of data. The basic concept of prescriptive analytics is to consider the business impact on various aspects. Prescriptive analytics involves the use of advanced mathematical models and algorithms, using artificial intelligence, to provide an automated solution to a problem.
What Is The Difference Between Predictive And Prescriptive Analytics
The key difference between prescriptive analytics and traditional predictive analytics is that the former tells you how to implement changes.
Due to the nature of predictive analytics, it indicates what may unfold but does not provide guidance as to how to proceed so that a business can be in the driver’s seat to control these “future” events. Traditional predictive analytics involves the use of data-driven models and algorithms to build a model that provides forecasting or inferences about what will happen in the future based on existing data. However, the output of the model is not data-driven, instead, it is a series of predictions or inferences about what will happen in the future.
Prescriptive analytics, on the other hand, helps determine the most profitable decision. The decision-making process can be accomplished through the use of natural language processing, clustering, clustering, classification, decision trees, neural networks, fuzzy logic, Bayesian networks, rule-based analytics, and so on.
While the problem may be predicting the future, this is not the primary objective of prescriptive analytics; that is best left to predictive analytics, especially in industries that require high accuracy. For example, if you are operating a nuclear power plant or manufacturing airplane parts, then predictive analytics can help reduce the risk of accidents. Predictive analytics relies on historical data to predict future events. This method is used to forecast customer behavior, product demand, and financial performance.
Prescriptive analytics is largely dedicated to the question “how to?” rather than “what if”.
While one could think of predictive analytics as determining the outcome of a forecast, it’s still about achieving an outcome. Prescriptive analytics, on the other hand, don’t really care what the outcome is: all it is concerned about is what should happen and how to achieve it. While predictive analytics and prescriptive analytics might seem like two sides of the same coin, they’re really not. Predictive analytics is about providing support for an existing decision, while prescriptive analytics helps you develop a new decision — there’s still a lot of overlap between the two, but we’ll get to that.
One example to differentiate between predictive and prescriptive is: You need a predictive model to find out what customers are likely to buy. But you need prescriptive models to find out which is the best communication channel, or how many emails you should send to increase the chance of purchase.
How Does Prescriptive Analytics Work In The Overall Business Decision Process?
First, the company should understand what is happening in its business based on current trends and historic data. Based on these inputs, prescriptive analytics can be used to change the business for the better.
According to us, any company that is using predictive analytics should also be using prescriptive analytics to identify opportunities, address risks and improve business performance. Using analytical models and algorithms to make decisions and recommendations, businesses can use prescriptive analytics to go a step beyond just trying to understand what is likely to happen in the future. Prescriptive analytics examines information to understand the impact on the business and prescribe the appropriate action or next best step for improvement.
Types Of Prescriptive Analytics
When it comes to prescriptive analytics, the three main subtypes are:
1. Operational Prescriptive Analytics – typically used in business settings. This type of prescriptive analytics is the most common because it is most suited to business-driven decision-making.
2. Financial Prescriptive Analytics – often used in conjunction with Operational Prescriptive Analytics.
3. Prospective/Strategic Prescriptive Analytics – this prescriptive analytics type is most often used in an environment where business plans are made and implemented with the full understanding of short-term goals.
Prescriptive Analytics: Getting Started
This post focuses on prescriptive analytics and how to go about using it to help your business. Prescriptive analytics works on the basis of what should be done. It takes into account the problem at hand and suggests potential solutions for solving it. For that, we need to put things in perspective first, vis-a-vis predictive analytics.
As we have said before in this post, the accuracy of predictive analytics is dependent on the quality of historical data. If the quality of the latter is poor or sketchy, the forecast could get affected. Predictive analytics also works well in areas where sample sizes are low and the population is not well-defined. This method can help estimate future events in areas such as finance, IT, and human resources.
The majority of existing predictive analytics techniques are based on the idea that there is some sort of underlying causal relationship between the different variables in your data set. These techniques are typically used when there are two or more independent variables that are measured in a specific process, and then you want to use these variables to predict an outcome that depends on these independent variables.
This type of modeling is useful when you want to know what will happen with the dependent variable (the outcome) once the independent variables change.
Prescriptive analytics, on the other hand, doesn’t predict the future. It can only describe the cause-and-effect relationship between the variables of interest. It is used to identify actions that can be taken to affect the outcome. The output is not a prediction of what might happen, but rather what should happen to achieve desired outcomes.
Scenario – The first step in prescriptive analytics is to identify the objective. Scenarios should be clear and concrete; they should show exactly what will happen as a result of a change in your data. Two or three scenarios (such as the following) are normally sufficient. Scenarios should cover all relevant aspects of the project, both positive and negative.
Inputs – What does the output look like? What data should be collected? What methods will be used for the coalition of the data?
Business objective – What is the business objective? What is the expected value of the product/service/whatever? What are the most important constraints?
Outputs – What is the expected outcome? How will it be measured?
We do this so we know what outcome we are eventually trying to achieve.
Prescriptive analytics is typically used in situations where there is no causal relationship between two or more variables. It is useful when you want to know how to change the condition of the dependent variable in order to maximize the likelihood of the outcome you are trying to achieve.
Now, let’s discuss a working example of prescriptive analytics to understand how it can be used in the business context.
Our data analyst Pankaj Katkar shows you how to build a prescriptive analytics model, using a hypothetical example.
Let’s first set up our problem statement – consider we have a food company called Express Restaurant which has multiple locale outlets. The senior management, though, is always worried about the rising customer attrition levels. Its efforts to retain customers so far have been largely reactive. Only when the customer has not visited for many months is when the management takes action. That’s not a great strategy, is it? So the management team is now keen to take more proactive steps to stop the churn.
As data scientists let’s analyze the Express Restaurant data, derive insights, predict the potential behavior of customers, and then recommend steps to improve customer retention.
For building our model we will need the data of customers. For that, we can collect customer reviews which will be readily available on the site and channels like social media. After gathering reviews of the customers, we will also have to collect the data about customers’ visits to the restaurant to find out customer churn. We can consider customer churn for those customers who have not visited the restaurant in the last two months.
After gathering the data and finding the customers who have stopped visiting the restaurant, we need to find what are the features affecting the customer churn. For that, we can use customer reviews, and out of that, we will extract the topics mentioned by customers. Those extracted topics can be used as features to determine customer churn as those topics were faced by the customers while visiting the restaurant.
To extract topics, we will use the VOCA topic model by Express Analytics. The main list of topics that were found are:
- Price and quality
- Staff behavior
Radar chart of all the topics is shown below:
These topics are pretty natural as these topics contribute towards customer churn on which senior management has to take action. If the price is high and the quality of food is not good, customers will not re-visit. Also, if the staff behavior is rude people will most probably not visit again. Covid-19 also has an impact on customer churn.
After extracting the topics, we will have to find out which topics are affecting customer churn, and to what extent. For this, we have to build models for the prediction of customer churn based on the features.
To build the model, we can start from simple linear models to complex non-parametric and non-linear ones. We can try linear regression, logistic regression. I have tried different models and then fine-tuned them to fit our use case.
Proactive Strategy to Improve Performance – Prescriptive Analysis
And now comes the part we’ve been waiting for – prescriptive analytics. Let’s see what recommendations we can come up with to reduce the customer churn.
The following variables have a strong probability of changing customer decisions. These variables were generated by using the logistic regression model.
- Price of food
- Quality of food
- Staff behavior
- Customer service
- Covid-19 fallout
We saw these variables earlier in our discussion. Let’s put down our recommendations based on what we’ve understood from our model.
So these two variables, price, and quality of food have a high impact on the customer churn. These variables seem to be “actionable” by senior management. Thus, senior management needs to take proactive action with the help of their chefs to improve the quality of food and also focus on the price of food.
Coming to variable staff behavior and customer service, management needs to go through the feedback of the customers using sentiment score and needs to work on it.
What if the budget is limited? Then the company can optimize its retention efforts by reaching out to targeted customers. The company can focus on a particular location or region and then focus on the customers in that region or location by making necessary retention offers.
Thus, this is the prescription that is given, using a model, to help Express Restaurant and its outlets to stop losing customers.
Tools Required For Prescriptive Analytics
Prescriptive analytics tools range from high-level programming languages to integrated ERP tools to solution-specific software packages. These include SAS, Excel, SQL, Visual Basic, Visual C++, Visual Basic for Applications (VBA), etc.
Most of these tools have a prescriptive analytics capability built into them, although some may require custom programming.
Pre-built data analysis toolkits: The most commonly used of these is SAS. The toolkit includes operations and analytical tools such as scatter plots, ratio analysis, and linear regression that can be used for prescriptive analytics.
Tools that can be programmed into specific enterprise applications such as SAP and Oracle also include prescriptive analytics capabilities.
Custom tools: These are primarily developed to the specific needs of the user. Many of these tools are freeware or have shareware versions that are free to use without the need for a license.
Examples of these tools include network flows, graphical user interfaces for data processing, and visualization tools such as Excel.
Operating systems: In addition to the various prescriptive analytics platforms, different operating systems have been designed with features specifically designed to aid in prescriptive analytics. For example, Microsoft’s SQL Server, Red Hat’s Enterprise Linux, and IBM’s AIX are three systems designed to facilitate the implementation of prescriptive analytics by offering faster processing speeds with more robust memory capabilities.
In conclusion: Decision-making is an important capability of any organization. With the use of predictive analytics, organizations can systematically approach future outcomes and manage risk. Predictive analytics helps large businesses to make decisions with the assistance of historical data. Together, predictive analytics and prescriptive analytics help to make business decisions and manage risk. Essentially, prescriptive analytics is a branch of business analytics that deals with determining the best course of action to take, based on a set of circumstances or opportunities. Prescriptive analytics is already being applied in industries such as healthcare, insurance, and finance.
An Engine That Drives Customer Intelligence
Oyster is not just a customer data platform (CDP). It is the world’s first customer insights platform (CIP). Why? At its core is your customer. Oyster is a “data unifying software.”
Liked This Article?
Gain more insights, case studies, information on our product, customer data platform