Predictive vs. prescriptive analytics: augmenting your data strategy

Becca Weng - December 13th, 2022

Approaching the field of data science and machine learning can be daunting. The field is ever-changing and highly dynamic, with what seems like a plethora of new terms and concepts cropping up every time you check Google or go to a conference. Additionally, many terms seem to be used interchangeably, or to have huge swaths of overlap. As the field continues to grow over the years, we may begin to see clearer divides. Even as the concepts are developing, it is important to have a general understanding of certain key terms, so that you can start implementing solutions.

In this article, we will provide a high-level overview of the four main kinds of analytics, with a focus on predictive and prescriptive analytics. Then we’ll go over a few common data analytics models that you may encounter as well as some industry-specific use-cases, so you feel empowered to start leveraging data science to propel your business forward.

Gartner’s Analytics Capabilities Spectrum

Gartner's 2022 Analytics Capabilities SpectrumGartner's 2022 Analytics Capabilities Spectrum

Although predictive analytics and prescriptive analytics fall under the broader umbrella of “analytics,” remember that analytics is a core building block for data science and machine learning.

At their 2022 Data and Analytics Summit, Gartner defined four main kinds of analytics, according to what they call “The Analytics Capabilities Spectrum.” These analytics help to answer the questions of what, why, and how with regard to the data:

  1. Descriptive analytics: what happened?
  2. Diagnostic analytics: why and how did it happen?
  3. Predictive analytics: what is likely to happen?
  4. Prescriptive analytics: what should we do? What can we do to make ___ happen?

The phases affiliated with the spectrum are observation, orientation, decision making, and action. As you progress through the four kinds of analytics, you are able to take a more proactive role in leading and advancing your business or organization, with a focus on automation and optimization.

Descriptive analytics

Descriptive analytics focuses on examining existing data to answer the question “What happened?” This tends to be one of the first steps after data collection. As one of the first steps, it is also fundamental to the success of the rest of your analytics framework. A big part of descriptive analytics is creating visualizations, such as box plots, scatter plots, and histograms, as well as summary statistics, such as minimum and maximum values, standard deviation, and the mean. A common term used in the data science space is exploratory data analysis (EDA). While EDA can be performed using tools like Excel or dashboarding tools, it is important for data scientists to be able to perform EDA using programming languages like Python.

With Einblick, you can easily perform descriptive analytics by

The canvas approach to data science gives you the ability to explore your data and thought processes freely without constraints. You can read our blog post to learn more about how to perform exploratory data analysis in Python.

Try Einblick for Free

Diagnostic analytics

Like descriptive analytics, diagnostic analytics also uses existing or historical data. But the guiding questions and techniques are different. After determining WHAT happened using descriptive analytics, you can use diagnostic analytics to understand WHY and HOW. Diagnostic analytics use techniques such as regression analysis, key driver analysis, and root cause analysis to understand the relationships between variables, and begin understanding the factors that led to certain outcomes. Note that certain tools and techniques relevant to diagnostic analytics can also be used for predictive analytics, but the questions we are asking are different.

Predictive analytics

Moving on to predictive analytics, this is where the focus starts shifting from just extracting observations and orienting ourselves via past events, to being future-forward. Predictive analytics uses existing data, statistical techniques, and machine learning to predict future events and outcomes. Supervised machine learning techniques such as linear regression, logistic regression, decision trees, random forests, and time series analysis are commonly used in predictive analytics. The techniques help data experts determine the likelihood of particular outcomes, and this in turn can then help leaders determine what the best course of action is.

Prescriptive analytics

Lastly, we have prescriptive analytics. Put very simply, prescriptive analytics helps you determine the best course of action based on your existing data. Prescriptive analytics answers the question “what should we do?” or “what can we do to make ___ happen?” A common implementation of prescriptive analytics is a content recommendation engine. For any streaming service, the business goals may include, retaining as many users as possible or increasing minutes of streaming per user. Therefore, suggesting videos that users are most likely to enjoy watching is one course of action that will maximize the likelihood of the desired outcomes.

Data analytics models

Below, we provide a non-exhaustive list of some common models and techniques you may encounter while digging into data and analytics. Each technique has its own particular use-cases, as well as evaluation metrics. Some of these models may have different names or different use-cases, so having a general understanding of the techniques out there will give you a good foundation to build your predictive, and ultimately prescriptive analytics strategy. The one that you choose will depend on the questions you’re asking, and the data that you have.

Implement a versatile data science tool

Before we dive into the models that you'll see, check out this data workflow that was used to predict system failures early using sensor readings from an industrial process. With a platform like Einblick, you can use any of the following methods easily both with Python and with our code-optional operators, so you can run your entire data science process, while collaborating easily with stakeholders and teammates.

Key driver analysis

Key driver analysis (KDA) is a statistical technique that helps you understand how much different factors contribute to a particular outcome. For example, if you’re running a tutoring center, you might want to understand what is driving enrollment at your program. Is it the placement of advertisements or the customer service or proximity to schools? Key driver analysis is a technique that can help you quantify which of these factors matter the most to your customers. You can read more about how to use Einblick's key driver analysis operator on our documentation page.

Regression analysis

Regression analysis is a family of statistical techniques that can be used to understand the relationships between variables, as well as predict outcomes. For example, if you are manufacturing cars, you may want to keep an ongoing prediction of the cost of production, based on factors such as supply chain delays, wages, operating costs, global economic conditions, cost of raw materials, and more. Regression can help you determine what factors positively or negatively correlate with production costs. Keep in mind that regression is a broad term, and that there are a number of different kinds of regression, from linear regression to Lasso and Ridge regression.

Classification models

Classification models are aptly named in that they help you to predict which category a data point falls into. Classification models can be particularly useful in clinical settings where you are trying to determine the likelihood of someone developing a medical condition or benefiting from a particular kind of treatment. You can read more about how to use Einblick's AutoML operator to build out machine learning pipelines using regression and classification models on our documentation page.

Clustering models

Clustering is a kind of unsupervised machine learning, which means the model is trained on input variables only. Clustering algorithms learn about the input data and try to create coherent groups from the data. An example of this is customer segmentation. Let’s say you’ve just launched your tutoring company, and you’re gaining some traction, but you want to understand who your customers are. You can use clustering algorithms to create groupings of customers based on their underlying characteristics. You can then use the results of your analysis to make targeted forays into each market.

Recommendation systems

Recommendation systems are used to offer optimized suggestions to users. These systems work by examining user choices, predicting what they would most like, and then offering that suggestion to them in an accessible way. We see recommendation systems every day, particularly with content, on any streaming service, or content platform. The goal of many of these recommendations is to keep the user in-product or in-service as long as possible.

Prescriptive vs. predictive analytics

At times it can be difficult to distinguish between predictive and prescriptive analytics because the two are so intertwined. Prescriptive analytics is like the final evolution of predictive analytics. As a result, your prescriptive analytics can only ever be as good as your predictions, and the data those are based off of. For example, if you work at an educational institution, whether a school or a tutoring center or bootcamp, it’s important to have a solid prediction of student outcomes. But what’s even more important is how you intervene based on your predictions to optimize outcomes. That is where prescriptive analytics becomes the critical factor. To help differentiate between prescriptive analytics and predictive analytics, see below for some concrete examples of each.

Manufacturing: supply chain management

Context and business problem

As people become more accustomed to same-day or two-day delivery, production slowdowns can really damage a company’s reputation among customers. This means that managing the supply chain of production and manufacturing, without sacrificing product quality, becomes increasingly important to maintain good customer relationships.

Predictive analytics approach

If you’re working in manufacturing, you probably have a ton of data about your product as well as the sourcers, production plants, and all the stakeholders involved in manufacturing your car, computer, tablet, water bottles, yoga mats, or whatever else you’re selling. For example, plant location, materials used in your product, name of each part required to build your product, how critical each part is to production, cost of shipping by boat or by air freight, time to assemble the product once all the parts have arrived, and more. You can use all of this information to calculate the probability of production being late, and identify the most important factors leading to slowdowns.

Prescriptive analytics approach

The next step after setting up a system to identify slowdowns, could be setting up alerts that identify the right stakeholders to take action. For example, they can call their suppliers two weeks in advance to switch the shipping method from ship to air freight, if it’s more cost effective to do so. They can then use whatever information that they get from speaking with their suppliers to update the predictive model, and from that information determine what best next steps are.

Service providers: customer churn

Context and business problem

Whether you are a financial institution, a digital subscription service, a food delivery service, or anything in between, you have customers, and you want them to stay as long as possible. You want them to renew their subscription, keep using your app, and keep using your services.

Predictive analytics approach

One form of predictive analytics that you can use to ensure customer loyalty is called customer churn analysis. By using classification models and other supervised learning techniques, you can input several factors you think might contribute to customer churn–for example, frequency of marketing emails, frequency of purchases, how the customer heard about the service, delivery time, changes in the user interface of the app–and then try to predict whether or not a customer will churn, or when they are likely to churn.

Prescriptive analytics approach

Next, you can use your predictions to identify customers most likely to churn, and determine when to intervene. For example, perhaps that means implementing a recommendation system to show new products to buy, or by offering a coupon to re-engage with the service around the holidays. You might change the headings in your emails. Prescriptive analytics helps you systematically take action to move towards your desired goal.

Check out how we used Einblick to predict banking churn! Predictive analytics is an important starting point so you can start determining best action steps.

Education: student retention

Context and business problem

One of the main outcomes for many educational institutions is student retention. If you’re a teacher, a school counselor, a tutor, a professor, or anyone that is working with students you are collecting data about the students you’re working with. Some of it may be quantitative, like test scores, attendance rate, essay grades, and participation percentage. You may also collect qualitative data such as noticing changes in demeanor, attitude shifts, and peer feedback.

Predictive analytics approach

Partway through working with your students, you can use classification analysis to predict the likelihood of a student dropping out of a class or program. Predictive analytics can help you identify the students that are likely to drop a class or program.

Prescriptive analytics approach

If your ultimate goal is student retention, then you may want to intervene with the students most likely to drop your class or program. By using machine learning techniques, you can use the predictions to then determine what the next best solution is, as soon as possible. But whether or not you’ve taken the right course of action is predicated on having solid predictive analytics tools and techniques in place.

Key Takeaways

  • There are four main kinds of analytics: descriptive, diagnostic, predictive, and prescriptive.
  • Predictive analytics uses existing data, statistical techniques, and machine learning to predict future events and outcomes, and answers the question "what is likely to happen?"
  • Prescriptive analytics helps you determine the best course of action based on your existing data, and answers the question "what should I do?"
  • Prescriptive analytics builds off of predictive analytics, and can be used in any setting, from supply chain management to customer churn to student retention.
  • It is likely that you are already implementing some predictive analytics and prescriptive analytics practices on a daily basis, whether that’s noting down customer feedback, or checking in on your suppliers or modifying your campaign subject lines.
  • Using a data science canvas like Einblick eliminates many data science pain points, will serve to automate your analytics approach, saving you time and manual work.

About

Einblick is an agile data science platform that provides data scientists with a collaborative workflow to swiftly explore data, build predictive models, and deploy data apps. Founded in 2020, Einblick was developed based on six years of research at MIT and Brown University. Einblick customers include Cisco, DARPA, Fuji, NetApp and USDA. Einblick is funded by Amplify Partners, Flybridge, Samsung Next, Dell Technologies Capital, and Intel Capital. For more information, please visit www.einblick.ai and follow us on LinkedIn and Twitter.

Start using Einblick

Pull all your data sources together, and build actionable insights on a single unified platform.

  • All connectors
  • Unlimited teammates
  • All operators