Prescriptive vs. Predictive Analytics: Examples & Use Cases
Explore the differences between prescriptive and predictive analytics. Understand their unique benefits, applications, and how they can revolutionize your data strategy.
We’ve all wondered about the future. What will happen? How will it happen? While these questions can feel existential, they also serve a strategic purpose. Nearly every business that’s been hailed as innovative has had this forward-thinking mindset – considering everything from emerging market trends to potential supply chain disruptions.
But there’s a difference between guesswork and a shrewd prediction, and that comes down to: analytics. We can further define analytics between four broad categories:
Descriptive: This uses historical data to identify specific patterns or relationships, that is: what happened?
Diagnostic: This type of analytics builds on descriptive analytics to answer the question of why did this happen? Diagnostic analytics looks at the relationship between variables and the relationship between cause and effect.
Predictive: Predictive analytics looks to the future, asking: what might happen? This uses a combination of historical data analysis, statistical modeling, machine learning and AI to make these forecasts.
Prescriptive: Prescriptive analytics goes deeper than predictive analytics, looking at the relationship between certain variables and how these factors impact the final outcome. It focuses on, what steps can we take to make this happen?
In this article, we’ll be focusing on the last two types of analytics: predictive and prescriptive.
What is predictive analytics?
Predictive analytics is concerned with what could occur, based on the trends, patterns, and behaviors we’ve observed in the past and at present. Predictive analytics uses data mining, statistical modeling, and AI to analyze historical and real-time data, delivering forecasts that businesses use to mitigate risk, predict churn, provide personalized product recommendations, and more.
One of the most well-known applications of predictive models is the predictions Netflix makes about what a subscriber is likely to watch next based on their viewing history. But predictive models go further back than Netflix – they’ve also been used for decades in credit scoring.
Of course, predictive analytics also has its limitations. One of them is overfitting – an issue that arises when a predictive model doesn’t work on any other data but the data it was trained on. There is also the risk of training the model on biased or incorrect data, resulting in inaccurate or discriminatory forecasts.
What is prescriptive analytics?
Prescriptive analytics seeks to answer the question, “What should be done next?” It ingests data from numerous sources, applies AI solutions (e.g., machine learning algorithms), and recommends the best course of action.
With prescriptive models, companies can determine how a specific decision could affect a business outcome – and adjust it to increase their chances of success. (However, it’s important to note that with the inherent unpredictability of the future, prescriptive analytics might not always be the perfect forecast.)
Prescriptive vs. predictive analytics
Both prescriptive and predictive analytics support improved decision-making in their own way. The table below outlines the key differences in how they support organizations:
Despite the differences between prescriptive and predictive analytics, you don’t have to choose between one or the other – in fact, combining these two types of analytics yields more granular, actionable insights.
For example, if you’re planning a flash sale, then prescriptive analytics could suggest you use the SMS channel to announce the sale for maximum engagement, and predictive analytics could estimate that doing so will grow sales by 15%.
Examples of prescriptive and predictive analytics (+ use cases)
Below are just a handful of examples that illustrate how organizations across industries are taking advantage of prescriptive and predictive analytics.
Successful email marketing campaigns are backed by data. By analyzing how your subscribers interact with your emails and what drives engagement, you’re able to make email communications more relevant and personalized for your audience.
You can leverage predictive analytics to estimate when a prospect is likely to make a purchase or unsubscribe from your newsletter. With these insights, you can craft emails that address their unique needs and improve customer retention.
In healthcare, we can find use cases of big data analytics on the level of both resource planning and diagnostics. For example, a combination of prescriptive and predictive analytics can help healthcare administrators predict future demand and then allocate the right amount of staff to prevent long waiting times.
Recent research into the use of predictive analytics for cancer detection has also found that it can identify breast cancer with a 100% accuracy rate. In the case of patients with oral cancer, predictive models were also highly accurate in estimating their survival rate.
Prescriptive and predictive analytics allow financial institutions to improve fraud detection and prevention. The analytics protects the banks’ customers from bad actors while saving resources they would otherwise have spent on investigating fraud.
Behavioral biometrics, which authenticates users based on their address, typical typing speed, and the way they hold their phone, is one way analytics enhances fraud prevention. A U.S. bank that integrated behavioral biometrics found that 96% of real users were characterized by familiar input behavior (the way they usually log in), while 87% of high-risk users had an unusual login pattern.
Product managers gather all sorts of data during product development, from market research and customer surveys to usage data. When you feed this data into a prescriptive model, it can recommend the best ways to improve a product or which features to prioritize.
McKinsey has found that prescriptive analytics can “anticipate opportunities and recommend the timing of promotions and price and assortment changes in specific categories to raise foot traffic, basket size or any other KPI.”
In the case of a home improvement retailer, prescriptive models found that its highest-spending customers bought moving boxes. With this information, the retailer started offering big discounts on boxes, which drove revenue on other merchandise.
How to get the most out of prescriptive and predictive analytics
Whether you implement predictive, prescriptive, or both types of analytics, there are a few best practices to follow to get the most out of these tools.
1. Identify your goals
As we explained in our guide to analytics, identifying your business goals is the first step in installing analytics in your organization. A goal could be as simple as:
Discovering ways to grow an existing product’s market share
Developing a method of quickly answering analytics questions
2. Gather data
Unifying all of your customer data is essential to creating reliable data sets. A unified customer view will allow you to better predict their future behavior and find the most effective way of engaging them.
For example, you might discover that customers between the ages of 25 and 30 show more interest in a specific product. Prescriptive analytics could suggest you target them with a personalized email marketing campaign to boost sales of that product and introduce this demographic to any related products they would find interesting.
3. Adopt an analytics tool
A data analytics tool can help make sense of large amounts of data and provide actionable insight (Google Analytics is a common example of an analytics tool, which helps provide insight into web traffic, user acquisition, on-site conversions, etc.).
When looking at a data analytics tool for your business, we recommend considering the following:
The business goals you’re aiming to solve for (e.g., looking to optimize ad spend and in-app user experiences are two different goals, owned by different departments, that may require different tools or features).
The volume of data you’ll be working with. This can impact pricing (many analytics tools offer tiered subscriptions) and how quickly and effectively you're able to mine data for insights.
But as is the case with any form of analytics, the data that’s being used needs to be complete and up to date. This is where a customer data platform (CDP) can be instrumental, helping to ensure data accuracy at scale while seamlessly connecting to downstream tools.
Predictive analytics magic with Segment & AWS Redshift
Predictive analytics can be a massive undertaking – but what if you could implement it without writing any code (or having to manually construct an expensive data science infrastructure)? Whether you have little machine learning experience or you’re an expert, you can pair Twilio Segment’s CDP with AWS Redshift to:
Easily set up and run predictive analytics with SQL commands
Run an analysis of data from all of your sources directly in Redshift
Avoid slow queries with compressed data type
1. Capture customer data with Segment
Segment is able to collect, clean, and consolidate customer data at scale – data that can then be used to build machine learning models. Knowing what customer data to collect will depend on what ML model you’re hoping to build. For example, if you were an e-commerce store and wanted to predict customers' lifetime value, you may track events like: order_completed, order_refunded, coupon_applied, payment_info_entered, etc.
Here’s our e-commerce tracking spec if you’re interested in our suggestions on what to track for this industry.
2. Configure the ML inputs
In machine learning, an input refers to the variables or features the model is trained on (e.g., events, user traits). Following the customer lifetime value example, if we were building an ML model for customer LTV, we would use event data, like product views, items added to shopping cart, checkout completed, and pair it with customers actual lifetime values. This way the ML model can start to identify relationships and patterns that could be used to make predictions based on other customers’ behavior.
With Segment, we create Computed Traits that would act as our ML inputs, which could be sent to Redshift with a single click. Here’s a preview of the different types of Computed Traits in Segment:
3. Build the ML model with SQL & AWS Sagemaker
You can then use any SQL editor and connect it to Redshift. Here’s a sample query, in line with trying to predict customer LTV.
After running the query, AWS Sagemaker will generate multiple models based on your inputs (and pick the one they think is best). From there, you can import that model into Segment as a new Computed Trait, which can be used to build audiences or campaigns.
Interested in hearing more about how Segment can help you?
Connect with a Segment expert who can share more about what Segment can do for you.