How to Use Predictive Analytics for Business Intelligence

From banking to hospitality, retail to the government, oil and gas to manufacturing, predictive analytics can be useful to a variety of organizations. Here are a few scenarios in which predictive analytics are being used today:

Credit scores. Credit bureaus are running predictive analytics on individuals in order to determine their creditworthiness, which factors into their credit score.

Fraud detection. Predictive analytics can help financial institutions detect fraud before the fraud occurs.

Stock trading. Analytics can train financial systems to automatically buy or sell a stock based on past data, with the goal of maximizing returns.

Marketing campaigns. Targeted advertising campaigns use analytics to target the users most likely to click on search or social media ads. Platforms such as Facebook will also display better performing ads within an ad set more often, based on analytics.

Store layouts. Analytics can tell stores where to place different items in order to drive the most sales. If you always see chocolate bars in the checkout aisle, there is probably some data driving that decision.

Store inventory. Likewise, analytics can predict inventory needs so that stores carry the right products in the right amounts. (No analytics could have predicted the Great Toilet Paper Shortage of 2020, mind you.)

Ticket prices. If you follow a pro sports team, you might have noticed that not every game costs the same. Analytics can help determine which days of the week and opponents are more/less popular and adjust prices accordingly. This even applies to ticket resale sites, which use analytics to (re-)price tickets based on demand. And if you’ve used Google Flights, you know that airfares can fluctuate widely, even from one day to the next, based on predictive data.

Cybersecurity. The latest cybersecurity tools use predictive analytics to detect network anomalies and spot unusual behaviour to help catch intruders before the malware drops. Analytics are also used to help filter spam out of your inbox.

Industry 4.0. Organizations use analytics to analyze their operations data and train models to improve operational efficiency and maximize revenue.

Risk reduction. Analytics can help reduce risk for critical operations, such as mining, where lives are potentially at stake.

A predictive model is a set of mathematical calculations that aims to predict future outcomes based on past behaviour and assigns a predictive score. “How do we build a predictive model? It’s really complicated to get there if a human brain is building that model,” says Singh. “But in order to do that, we can turn to machine learning (ML).”

“With machine learning, we can build these models by reverse engineering the data to uncover patterns within the data. Those patterns help the system to build a predictive model. But in order to get there, machine learning algorithms and systems need to be trained,” Singh says.

Machine learning can be broken down into supervised and unsupervised learning. “Supervised learning is similar to a teacher teaching a bunch of students in a classroom,” says Singh. “You know what to teach, you know the data and how the system has responded in the past, so you use that past data to train the algorithm. The data sets used for supervised learning are labelled. We know the past outcomes, and we use the learning from those outcomes to predict the future.”

Singh points out that supervised learning is used in spam detection, sentiment analysis, weather forecasting and price predictions. “This type of machine learning has higher accuracy in terms of predictions because we’ve trained our system based on past experiences.”

Unsupervised learning, on the other hand, does not have any past data, and uses unlabelled data sets. “We’re just trying to uncover patterns by ingesting large volumes of data into the algorithm,” says Singh. “The goal is to get insights from the learnings by using large amounts of data.” Unsupervised learning is used within cybersecurity to detect network traffic anomalies, as well as in areas like building customer personas in marketing or for recommending shows you’d like on streaming services.

“Especially with unsupervised learning, we’re processing lots and lots of data,” says Singh. “There’s a lot of computations happening in the background, and a lot of compute being used. At the same time, we are also using a lot of storage. All that data we have to process has to be stored somewhere that is accessible and has a reasonable amount of latency, so that we can quickly process the information. We need lots of compute and storage for this, and as you can imagine, the cloud really comes in handy.”

The good news is you can manage and scale entire data analytics platforms without buying any hardware, as major cloud providers like AWS, Microsoft Azure and Google all offer analytics solutions. “Those data warehouse platforms all have artificial intelligence and machine learning built in,” says Singh. “You can use machine learning in a data warehouse, which is built in the cloud, and run predictive analytics on top of your data warehouse. There are tools available to build scalable models for your organization.”

The cloud also offers enhanced security and a pay-per-use model, which can lead to cost savings. “If you need a job to complete quickly, you can throw in more compute resources or add faster storage to it and finish the job sooner. Or, if you want to save costs, you can reduce your compute instances, so you have that control to efficiently make use of the cloud.”